Do not let diskless.conf persist into EL9 diskless images

It fouls run of kdump building the kdump image.

README.md

@@ -13,7 +13,7 @@ If you're coming from xCAT, check out [this comparison](https://hpc.lenovo.com/u
 
 # Documentation
 
-Confluent documentation is hosted on hpc.lenovo.com: https://hpc.lenovo.com/users/documentation/
+Confluent documentation is hosted on: https://xcat2.github.io/confluent-docs/
 
 # Download
 

confluent_client/addattribs.py

@@ -14,11 +14,16 @@ import shutil
 
 shutil.copyfile('doc/man/nodeattrib.ronn.tmpl', 'doc/man/nodeattrib.ronn')
 shutil.copyfile('doc/man/nodegroupattrib.ronn.tmpl', 'doc/man/nodegroupattrib.ronn')
-with open('doc/man/nodeattrib.ronn', 'a') as outf:
-    for field in sorted(attr.node):
-        outf.write('\n* `{0}`:\n  {1}\n'.format(field, attr.node[field]['description']))
-with open('doc/man/nodegroupattrib.ronn', 'a') as outf:
-    for field in sorted(attr.node):
-        outf.write('\n* `{0}`:\n  {1}\n'.format(field, attr.node[field]['description']))
 
+def append_attributes(filename):
+    with open(filename, 'a') as outf:
+        for field in sorted(attr.node):
+            outf.write('\n* `{0}`:\n  {1}\n'.format(field, attr.node[field]['description']))
+            # Optionally write valid values if they exist
+            for key, values in attr.node[field].items():
+                if key.startswith('valid'):
+                    values_formatted = ', '.join("'{0}'".format(v) for v in values)
+                    outf.write(f'\n    Valid values: {values_formatted}\n')
 
+append_attributes('doc/man/nodeattrib.ronn')
+append_attributes('doc/man/nodegroupattrib.ronn')

confluent_client/bin/confetty

@@ -1,4 +1,4 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
 # Copyright 2014 IBM Corporation
@@ -45,6 +45,7 @@ import math
 import getpass
 import optparse
 import os
+import re
 import select
 import shlex
 import signal
@@ -654,13 +655,17 @@ def quitconfetty(code=0, fullexit=False, fixterm=True):
 def get_session_node(shellargs):
     # straight to node console
     if len(shellargs) == 1 and ' ' not in shellargs[0]:
-        return shellargs[0]
+        targ = "/nodes/%s/console/session" % shellargs[0]
+        return targ, shellargs[0]
     if len(shellargs) == 2 and shellargs[0] == 'start':
         args = [s for s in shellargs[1].split('/') if s]
         if len(args) == 4 and args[0] == 'nodes' and args[2] == 'console' and \
                 args[3] == 'session':
-            return args[1]
-    return None
+            return shellargs[1], args[1]
+        if len(args) == 5 and args[0] == 'nodes' and args[2] == 'shell' and \
+                args[3] == 'sessions':
+            return shellargs[1], args[1]
+    return None, None
 
 
 def run_inline_command(path, arg, completion, **kwargs):
@@ -917,10 +922,10 @@ def main():
     doexit = False
     inconsole = False
     pendingcommand = ""
-    session_node = get_session_node(shellargs)
+    targ, session_node = get_session_node(shellargs)
     if session_node is not None:
         consoleonly = True
-        do_command("start /nodes/%s/console/session" % session_node, netserver)
+        do_command("start %s" % targ, netserver)
         doexit = True
     elif shellargs:
         do_command(shellargs, netserver)
@@ -965,8 +970,15 @@ def main():
                     sys.stdout.write('Lost connection to server')
     quitconfetty(fullexit=True)
 
+sgr_re = re.compile(r'(\x1b\[[0-9;]*m)')
+sgr_parameters_re = re.compile(r'\x1b\[([0-9;]*)m')
+fgcolor = None
+bgcolor = None
+fgshifted = False
+pendseq = ''
 def consume_termdata(fh, bufferonly=False):
     global clearpowermessage
+    global fgcolor, bgcolor, fgshifted, pendseq
     try:
         data = tlvdata.recv(fh)
     except Exception:
@@ -975,7 +987,59 @@ def consume_termdata(fh, bufferonly=False):
         updatestatus(data)
         return ''
     if data is not None:
-        data = client.stringify(data)
+        indata = pendseq + client.stringify(data)
+        pendseq = ''
+        data = ''
+        for segment in sgr_re.split(indata):
+            if sgr_re.match(segment):  # it is an sgr, analyze, maybe replace
+                params = []
+                for parameters in sgr_parameters_re.findall(segment):
+                    for param in parameters.split(';'):
+                        params.append(param)
+                        if param == '0':
+                            fgcolor = None
+                            bgcolor = None
+                        try:
+                            ival = int(param)
+                        except ValueError:
+                            continue
+                        if 40 <= ival <= 47 or 100 <= ival <= 107:
+                            bgcolor = ival
+                        if 30 <= ival <= 37 or 90 <= ival <= 97:
+                            fgcolor = ival
+                if bgcolor is not None:
+                    fgindicated = False
+                    for idx, param in enumerate(params):
+                        try:
+                            ival = int(param)
+                        except ValueError:
+                            continue
+                        if 30 <= ival <= 37 and (bgcolor % 10 == ival % 10):
+                            fgindicated = True
+                            fgshifted = True
+                            ival += 60
+                            params[idx] = str(ival)
+                    if not fgindicated and fgcolor is not None:
+                        if bgcolor and (bgcolor % 10) == (fgcolor % 10):
+                            fgshifted = True
+                            params.append(str((fgcolor % 10) + 90))
+                        elif fgshifted:
+                            params.append(str(fgcolor))
+                segment = '\x1b[' + ';'.join(str(p) for p in params) + 'm'
+            data += segment
+        # defer any partial ansi escape sequence for a later pass
+        escidx = segment.rfind('\x1b[')
+        if escidx >= 0:
+            for chr in segment[escidx + 2:]:
+                if 0x40 <= ord(chr) <= 0x7e:
+                    break
+            else:
+                # incomplete escape sequence, don't print it yet
+                data = data[:-len(segment) + escidx]
+                pendseq = segment[escidx:]
+        if not pendseq and segment and segment[-1] == '\x1b':
+            data = data[:-1]
+            pendseq = '\x1b'
         if clearpowermessage:
             sys.stdout.write("\x1b[2J\x1b[;H")
             clearpowermessage = False

confluent_client/bin/confluent2ansible

@@ -1,8 +1,9 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 import optparse
 import signal
 import sys
 import os
+
 try:
     signal.signal(signal.SIGPIPE, signal.SIG_DFL)
 except AttributeError:
@@ -17,18 +18,16 @@ import confluent.client as client
 import confluent.sortutil as sortutil
 
 
-def lookupdata(data, key):
-    ret = data.get(key, {}).get('value', '')
-    if ret is None:
-        ret = ''
-    return ret
+
 
 def main():
     argparser = optparse.OptionParser(
-    usage='''\n       %prog noderange -o ansible.hosts
-             \n ''')
+        usage='''\n       %prog noderange -o ansible.hosts -a
+        ''')
     argparser.add_option('-o', '--output',
-                         help='Ansible hosts file')
+                         help='Writes an Ansible hosts file')
+    argparser.add_option('-a', '--append', action='store_true',
+                         help='Appends to existing hosts file')
     (options, args) = argparser.parse_args()
     try:
         noderange = args[0]
@@ -45,24 +44,42 @@ def main():
             if node not in databynode:
                 databynode[node] = {}
             databynode[node].update(res['databynode'][node])
+
     nodesbygroup = {}
-    with open(options.output, 'w') as importfile:
-        needempty = False
-        for node in sortutil.natural_sort(databynode):
-            data = databynode[node]
-            if not data.get('groups', []):
-                importfile.write(node + '\n')
-                needempty = True
-            for g in data.get('groups', []):
-                if g not in nodesbygroup:
-                    nodesbygroup[g] = set([node])
+    for node in sortutil.natural_sort(databynode):
+        data = databynode[node]
+        groups = data.get('groups', [])
+        if not groups:
+            nodesbygroup.setdefault('', set()).add(node.strip().lower())
+        else:
+            for g in groups:
+                nodesbygroup.setdefault(g, set()).add(node.strip().lower())
+    existing_data = {}
+    if options.append and os.path.exists(options.output):
+        current_group = ''
+        with open(options.output, 'r') as f:
+            for line in f:
+                line = line.strip().lower()
+                if not line:
+                    continue
+                if line.startswith('[') and line.endswith(']'):
+                    current_group = line[1:-1]
+                    existing_data.setdefault(current_group, set())
                 else:
-                    nodesbygroup[g].add(node)
-        if needempty:
-            importfile.write('\n')
-        for group in sortutil.natural_sort(nodesbygroup):
-            importfile.write('[{0}]\n'.format(group))
-            for node in sortutil.natural_sort(nodesbygroup[group]):
+                    existing_data.setdefault(current_group, set()).add(line)
+
+    for group, nodes in nodesbygroup.items():
+        nodes = {n.strip().lower() for n in nodes}
+        current_nodes = existing_data.get(group, set())
+        new_nodes = nodes - current_nodes
+        if new_nodes:
+            existing_data.setdefault(group, set()).update(nodes)
+
+    with open(options.output, 'w') as importfile:
+        for group in sortutil.natural_sort(existing_data.keys()):
+            if group:
+                importfile.write('[{0}]\n'.format(group))
+            for node in sortutil.natural_sort(existing_data[group]):
                 importfile.write('{0}\n'.format(node))
             importfile.write('\n')
 

confluent_client/bin/confluent2hosts

@@ -118,6 +118,7 @@ def main():
     ap.add_argument('-a', '--attrib', help='Pull ip addresses and hostnames from attribute database', action='store_true')
     ap.add_argument('-i', '--ip', help='Expression to generate addresses (e.g. 172.16.1.{n1} or fd2b:246f:8a50::{n1:x})')
     ap.add_argument('-n', '--name', help='Expression for name to add ({node}-compute, etc). If unspecified, "{node} {node}.{dns.domain}" will be used', action='append')
+    ap.add_argument('-f', '--fqdn-first', help='Put the FQDN first in the hosts entries', action='store_true')
     args = ap.parse_args()
     c = client.Command()
     if args.name:
@@ -173,7 +174,13 @@ def main():
                                     break
                             else:
                                 for name in list(names):
-                                    names.append('{0}.{1}'.format(name, mydomain))
+                                    fqdn = '{0}.{1}'.format(name, mydomain)
+                                    if args.fqdn_first:
+                                        # Insert FQDN at the beginning if --fqdn-first flag is set
+                                        names.insert(0, fqdn)
+                                    else:
+                                        # Otherwise, append FQDN at the end (original behavior)
+                                        names.append(fqdn)
                     names = ' '.join(names)
                     merger.add_entry(ipdb[node][currnet], names)
         merger.write_out('/etc/whatnowhosts')

confluent_client/bin/dir2img

@@ -8,30 +8,35 @@ import os
 import subprocess
 import sys
 
-def create_image(directory, image, label=None):
-    ents = 0
-    datasz = 512
-    for dir in os.walk(sys.argv[1]):
-        ents += 1
-        for filen in dir[2]:
+def create_image(directory, image, label=None, esize=0, totalsize=None):
+
+    if totalsize:
+        datasz = totalsize * 1048576
+    else:
+        ents = 0
+        datasz = 512 + (esize * 1048576)
+        for dir in os.walk(sys.argv[1]):
             ents += 1
-            filename = os.path.join(dir[0], filen)
-            currsz = os.path.getsize(filename)
-            # assuming up to 65k cluster
-            currsz = (currsz // 512 +1) * 512
-            datasz += currsz
-    datasz += ents * 32768
-    datasz = datasz // 16384 + 1
+            for filen in dir[2]:
+                ents += 1
+                filename = os.path.join(dir[0], filen)
+                currsz = os.path.getsize(filename)
+                # assuming up to 65k cluster
+                currsz = (currsz // 512 + 1) * 512
+                datasz += currsz
+        datasz += ents * 32768
+    datasz = datasz // 65536 + 1
     with open(image, 'wb') as imgfile:
-        imgfile.seek(datasz * 16384 - 1)
+        imgfile.seek(datasz * 65536 - 1)
         imgfile.write(b'\x00')
     if label:
+        # 4 heads, 32 sectors, means 65k per track
         subprocess.check_call(['mformat', '-i', image, '-v', label,
                                '-r', '16', '-d', '1', '-t', str(datasz),
-                               '-s', '16','-h', '2', '::'])
+                               '-s', '32','-h', '4', '::'])
     else:
         subprocess.check_call(['mformat', '-i', image, '-r', '16', '-d', '1', '-t',
-                                str(datasz), '-s', '16','-h', '2', '::'])
+                                str(datasz), '-s', '32','-h', '4', '::'])
     # Some clustered filesystems will have the lock from mformat
     # linger after close (mformat doesn't unlock)
     # do a blocking wait for shared lock and then explicitly
@@ -56,6 +61,21 @@ if __name__ == '__main__':
             sys.argv[0]))
         sys.exit(1)
     label = None
-    if len(sys.argv) > 3:
-        label = sys.argv[3]
-    create_image(sys.argv[1], sys.argv[2], label)
+    args = sys.argv
+    esize = 0
+    try:
+        earg = args.index('-e')
+        esize = int(args[earg + 1])
+        args = args[:earg] + args[earg +2:]
+    except ValueError:
+        pass
+    totsize = None
+    try:
+        earg = args.index('-s')
+        totsize = int(args[earg + 1])
+        args = args[:earg] + args[earg +2:]
+    except ValueError:
+        pass
+    if len(args) > 3:
+        label = args[3]
+    create_image(args[1], args[2], label, esize, totsize)

confluent_client/bin/nodeapply

@@ -68,6 +68,7 @@ def run():
 
     currprocs = 0
     all = set([])
+    poller = select.epoll()
     pipedesc = {}
     pendingexecs = deque()
     exitcode = 0
@@ -102,19 +103,23 @@ def run():
         cmdv = ['ssh', sshnode] + cmdvbase + cmdstorun[0]
         if currprocs < concurrentprocs:
             currprocs += 1
-            run_cmdv(sshnode, cmdv, all, pipedesc)
+            run_cmdv(sshnode, cmdv, all, poller, pipedesc)
         else:
             pendingexecs.append((sshnode, cmdv))
     if not all or exitcode:
         sys.exit(exitcode)
-    rdy, _, _ = select.select(all, [], [], 10)
+    rdy = poller.poll(10)
     while all:
         pernodeout = {}
         for r in rdy:
+            r = r[0]
             desc = pipedesc[r]
+            r = desc['file']
             node = desc['node']
             data = True
-            while data and select.select([r], [], [], 0)[0]:
+            singlepoller = select.epoll()
+            singlepoller.register(r, select.EPOLLIN)
+            while data and singlepoller.poll(0):
                 data = r.readline()
                 if data:
                     if desc['type'] == 'stdout':
@@ -131,15 +136,17 @@ def run():
                     if ret is not None:
                         exitcode = exitcode | ret
                         all.discard(r)
+                        poller.unregister(r)
                         r.close()
                         if desc['type'] == 'stdout':
                             if idxbynode[node] < len(cmdstorun):
                                 cmdv =  ['ssh', sshnode] + cmdvbase + cmdstorun[idxbynode[node]]
                                 idxbynode[node] += 1
-                                run_cmdv(node, cmdv, all, pipedesc)
+                                run_cmdv(node, cmdv, all, poller, pipedesc)
                             elif pendingexecs:
                                 node, cmdv = pendingexecs.popleft()
-                                run_cmdv(node, cmdv, all, pipedesc)
+                                run_cmdv(node, cmdv, all, poller. pipedesc)
+            singlepoller.close()
         for node in sortutil.natural_sort(pernodeout):
             for line in pernodeout[node]:
                 line = client.stringify(line)
@@ -147,19 +154,21 @@ def run():
                 sys.stdout.write('{0}: {1}'.format(node, line))
         sys.stdout.flush()
         if all:
-            rdy, _, _ = select.select(all, [], [], 10)
+            rdy = poller.poll(10)
     sys.exit(exitcode)
 
 
-def run_cmdv(node, cmdv, all, pipedesc):
+def run_cmdv(node, cmdv, all, poller, pipedesc):
     nopen = subprocess.Popen(
         cmdv, stdin=devnull, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-    pipedesc[nopen.stdout] = {'node': node, 'popen': nopen,
-                              'type': 'stdout'}
-    pipedesc[nopen.stderr] = {'node': node, 'popen': nopen,
-                              'type': 'stderr'}
+    pipedesc[nopen.stdout.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stdout', 'file': nopen.stdout}
+    pipedesc[nopen.stderr.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stderr', 'file': nopen.stderr}
     all.add(nopen.stdout)
+    poller.register(nopen.stdout, select.EPOLLIN)
     all.add(nopen.stderr)
+    poller.register(nopen.stderr, select.EPOLLIN)
 
 
 if __name__ == '__main__':

confluent_client/bin/nodecertutil

@@ -0,0 +1,89 @@
+#!/usr/bin/python3
+
+import os
+import sys
+from cryptography import x509
+from cryptography.hazmat.primitives import hashes
+
+path = os.path.dirname(os.path.realpath(__file__))
+path = os.path.realpath(os.path.join(path, '..', 'lib', 'python'))
+if path.startswith('/opt'):
+    sys.path.append(path)
+import confluent.client as client
+
+def removebmccacert(noderange, certid, cmd):
+    for res in cmd.delete(f'/noderange/{noderange}/configuration/management_controller/certificate_authorities/{certid}'):
+        print(repr(res))
+
+def listbmccacerts(noderange, cmd):
+    certids = []
+    for res in cmd.read(f'/noderange/{noderange}/configuration/management_controller/certificate_authorities'):
+        certids.append(res.get('item', {}).get('href', ''))
+    for certid in certids:
+        for res in cmd.read(f'/noderange/{noderange}/configuration/management_controller/certificate_authorities/{certid}'):
+            for node in res.get('databynode', {}):
+                certdata = res['databynode'][node].get('pem', {}).get('value', '')
+                summary = ''
+                if not certdata:
+                    continue
+                san = res['databynode'][node].get('san', {}).get('value', '')
+                if san:
+                    summary += f" SANs: {san}"
+                subject = res['databynode'][node].get('subject', {}).get('value', '')
+                if subject:
+                    summary = subject
+                try:
+                    cert = x509.load_pem_x509_certificate(certdata.encode())
+                    sha256 = cert.fingerprint(hashes.SHA256()).hex().upper()
+                except Exception as e:
+                    print(f"Error processing certificate for {node}: {e}", file=sys.stderr)
+                    continue
+                summary += f" (SHA256={sha256})"
+                print(f"{node}: {certid}: {summary}")
+
+def installbmccacert(noderange, certfile, cmd):
+    if certfile:
+        try:
+            with open(certfile, 'r') as f:
+                certdata = f.read()
+        except Exception as e:
+            print(f"Error reading certificate file: {e}", file=sys.stderr)
+            sys.exit(1)
+
+        # Simple validation: check if it starts and ends with the correct PEM markers
+        if not (certdata.startswith("-----BEGIN CERTIFICATE-----") and certdata.strip().endswith("-----END CERTIFICATE-----")):
+            print("Invalid certificate format. Must be a PEM encoded certificate.", file=sys.stderr)
+            sys.exit(1)
+        payload = {'pem': certdata}
+        for res in cmd.update(f'/noderange/{noderange}/configuration/management_controller/certificate_authorities', payload):
+            print(repr(res))
+
+    
+
+if __name__ == '__main__':
+    import argparse
+
+    parser = argparse.ArgumentParser(description='Node certificate utility')
+    parser.add_argument('noderange', help='Node range to operate on')
+    subparsers = parser.add_subparsers(dest='command', help='Available commands')
+
+    # installbmccacert subcommand
+    install_parser = subparsers.add_parser('installbmccacert', help='Install BMC CA certificate')
+    install_parser.add_argument('filename', help='Certificate file to install')
+
+    remove_parser = subparsers.add_parser('removebmccacert', help='Remove BMC CA certificate')
+    remove_parser.add_argument('id', help='Certificate id to remove')
+
+    list_parser = subparsers.add_parser('listbmccacerts', help='List BMC CA certificates')
+
+    args = parser.parse_args()
+    c = client.Command()
+    if args.command == 'installbmccacert':
+        installbmccacert(args.noderange, args.filename, c)
+    elif args.command == 'removebmccacert':
+        removebmccacert(args.noderange, args.id, c)
+    elif args.command == 'listbmccacerts':
+        listbmccacerts(args.noderange, c)
+    else:
+        parser.print_help()
+        sys.exit(1)

confluent_client/bin/nodeconfig

@@ -1,7 +1,7 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
-# Copyright 2017 Lenovo
+# Copyright 2025 Lenovo
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -96,6 +96,12 @@ cfgpaths = {
     'bmc.static_ipv6_gateway': (
         'configuration/management_controller/net_interfaces/management',
         'static_v6_gateway'),
+    'bmc.vlan_id': (
+        'configuration/management_controller/net_interfaces/management',
+        'vlan_id'),
+    'bmc.mac_address': (
+        'configuration/management_controller/net_interfaces/management',
+        'hw_addr'),
     'bmc.hostname': (
         'configuration/management_controller/hostname', 'hostname'),
 }

confluent_client/bin/nodeconsole

@@ -15,6 +15,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import base64
 import optparse
 import os
 import subprocess
@@ -28,8 +29,31 @@ import confluent.client as client
 import confluent.sortutil as sortutil
 import confluent.logreader as logreader
 import time
+import select
+import signal
 import socket
 import re
+import tty
+import termios
+import fcntl
+import confluent.screensqueeze as sq
+try:
+    from PIL import Image, ImageDraw, ImageFont
+except ImportError:
+    Image = None
+
+try:
+    # sixel is optional, attempt to import but stub out if unavailable
+    import io
+    import sixel
+
+    class DumbWriter(sixel.SixelWriter):
+        def restore_position(self, output):
+            return
+except ImportError:
+    class DumbWriter():
+        def draw(self, imgfile):
+            sys.stderr.write("PySixel not detected, Sixel format display not supported\n")
 
 confettypath = os.path.join(os.path.dirname(sys.argv[0]), 'confetty')
 argparser = optparse.OptionParser(
@@ -46,6 +70,11 @@ argparser.add_option('-l', '--log', action='store_true', default=False,
 argparser.add_option('-T', '--Timestamp', action='store_true', default=False,
                      help= 'Dump log in stdout with timestamps')
 
+argparser.add_option('-s', '--screenshot', action='store_true', default=False,
+                     help='Attempt to grab screenshot and render using kitty image protocol')
+argparser.add_option('-i', '--interval', type='float',
+                     help='Interval in seconds to redraw the screenshot. Currently only '
+                          'works for one node')
 argparser.add_option('-w','--windowed', action='store_true', default=False,
                      help='Open terminal windows for each node.  The '
                           'environment variable NODECONSOLE_WINDOWED_COMMAND '
@@ -69,6 +98,234 @@ argparser.add_option('-w','--windowed', action='store_true', default=False,
 
 (options, args) = argparser.parse_args()
 
+oldtcattr = None
+oldfl = None
+
+def get_coords():
+    sys.stdout.write('\x1b[6n') #
+    sys.stdout.flush()
+    gotreply = select.select([sys.stdin,], [], [], 0.250)[0]
+    if gotreply:
+        response = ''
+        while select.select([sys.stdin,], [], [], 0.1)[0] and 'R' not in response:
+            response += sys.stdin.read()
+        coords = response.replace('R', '').split('[')[1].split(';')
+    #sys.stdout.write('\x1b[{}:{}H'.format(*coords))
+
+def direct_console():
+    global oldtcattr
+    global oldfl
+    oldtcattr = termios.tcgetattr(sys.stdin.fileno())
+    oldfl = fcntl.fcntl(sys.stdin.fileno(), fcntl.F_GETFL)
+    tty.setraw(sys.stdin.fileno())
+    fcntl.fcntl(sys.stdin.fileno(), fcntl.F_SETFL, oldfl | os.O_NONBLOCK)
+
+def indirect_console():
+    fcntl.fcntl(sys.stdin.fileno(), fcntl.F_SETFL, oldfl & ~os.O_NONBLOCK)
+    termios.tcsetattr(sys.stdin.fileno(), termios.TCSANOW, oldtcattr)
+
+def determine_tile_size(numnodes):
+    # for now, smash everything to a common aspect ratio.  16:11
+    # is pretty much wrong for everything, making 4:3 a bit too wide
+    # and 16:9 significantly too narrow, but it is serviceable
+    # An improvement could come with us owning the scaling
+    # instead of delegating to Kitty, which says if we specify both,
+    # we get stretching.  In theory we should be able to get aspect correct
+    # from kitty by omitting, but:
+    # then we don't know how much to move the cursor left after draw_image
+    # Konsole won't scale at all with only partial scaling specified
+    cheight, cwidth, pixwidth, pixheight = sq.get_screengeom()
+    # 16:12 is to roughly account for the 'titles' of the tiles
+    ratio = (pixwidth / 16) / (pixheight / 12)
+    bestdeviation = None
+    bestdims = []
+    for i in range(1, numnodes + 1):
+        number = numnodes
+        while number % i != 0:
+            number += 1
+        columns = i
+        rows = number // i
+        deviation = abs(ratio -  (columns / rows))
+        if bestdeviation is None:
+            bestdeviation = deviation
+            bestdims = [columns, rows]
+        elif deviation < bestdeviation:
+            bestdeviation = deviation
+            bestdims = [columns, rows]
+    # ok, the above algorithm can still pick things like
+    # 1 2 3
+    # 4
+    # So we will let it pick the number of rows, and
+    # then see if we can chop columns and still fit
+    while (bestdims[0] - 1) * bestdims[1] >= numnodes:
+        bestdims[0] = bestdims[0] - 1
+    cellswide = cwidth // bestdims[0]
+    cellshigh = cheight // bestdims[1]
+    tilewidth = cellswide * pixwidth / cwidth
+    tileheight = cellshigh * pixheight / cheight
+    if tilewidth > (tileheight * 16 / 11):
+        tilewidth = tileheight * 16 / 11
+        cellswide = int(tilewidth // (pixwidth / cwidth))
+    if tileheight > (tilewidth * 11 /16):
+        tileheight = tilewidth * 11 / 16
+        cellshigh = int(tileheight // (pixheight / cheight))
+    bestdims = bestdims + [cellswide, cellshigh, cellshigh * bestdims[1]]
+    # incur any scrolling we might get. This allows us to accurately
+    # save/restore cursor or even get coordinates without scrolling fouling
+    # the desired target
+    sys.stdout.write('\n' * bestdims[4])
+    sys.stdout.flush()
+    cursor_up(bestdims[4])
+    return bestdims
+
+cursor_saved = False
+def sticky_cursor():
+    global cursor_saved
+# get cursor restore_position
+    if sys.stdin.isatty() and not cursor_saved:
+        try:
+            direct_console()
+            sys.stdout.write('\x1b7')
+            cursor_saved = True
+        finally:
+            indirect_console()
+    elif cursor_saved:
+        try:
+            direct_console()
+            sys.stdout.write('\x1b8')
+        finally:
+            indirect_console()
+
+def cursor_up(count=1):
+    sys.stdout.write(f'\x1b[{count}A')
+def cursor_down(count=1):
+    sys.stdout.write(f'\x1b[{count}B')
+def cursor_right(count=1):
+    sys.stdout.write(f'\x1b[{count}C')
+def cursor_left(count=1):
+    sys.stdout.write(f'\x1b[{count}D')
+def cursor_save():
+    sys.stdout.write('\x1b7')
+def cursor_restore():
+    sys.stdout.write('\x1b8')
+def cursor_hide():
+    sys.stdout.write('\x1b[?25l')
+def cursor_show():
+    sys.stdout.write('\x1b[?25h')
+
+def get_pix_dimensions(width, height):
+    cheight, cwidth, pixwidth, pixheight = sq.get_screengeom()
+    imgwidth = int(pixwidth / cwidth * width)
+    imgheight = int(pixheight / cheight * height)
+    return imgwidth, imgheight
+
+def draw_text(text, width, height):
+    if Image:
+        maxfntsize = 256
+        imgwidth, imgheight = get_pix_dimensions(width, height)
+        nerr = Image.new(mode='RGB', size=(imgwidth, imgheight), color='green')
+        nd = ImageDraw.Draw(nerr)
+        for txtpiece in text.split('\n'):
+            fntsize = 8
+            txtfont = ImageFont.truetype('DejaVuSans.ttf', size=fntsize)
+            while nd.textlength(txtpiece, font=txtfont) < int(imgwidth * 0.90):
+                fntsize += 1
+                txtfont = ImageFont.truetype('DejaVuSans.ttf', size=fntsize)
+            fntsize -= 1
+            if fntsize < maxfntsize:
+                maxfntsize = fntsize
+        hmargin = int(imgwidth * 0.05)
+        vmargin = int(imgheight * 0.10)
+        nd.text((hmargin, vmargin), text, font=txtfont)
+        nd.rectangle((0, 0, nerr.width - 1, nerr.height -1), outline='white')
+        outfile = io.BytesIO()
+        nerr.save(outfile, format='PNG')
+        data = base64.b64encode(outfile.getbuffer())
+        draw_image(data, width, height, doscale=False)
+    else:
+        sys.stdout.write(text)
+        cursor_left(len(text))
+
+def draw_image(data, width, height, doscale=True):
+    imageformat = os.environ.get('CONFLUENT_IMAGE_PROTOCOL', 'kitty')
+    if doscale and Image and width:
+        bindata = base64.b64decode(data)
+        binfile = io.BytesIO()
+        binfile.write(bindata)
+        binfile.seek(0)
+        try:
+            img = Image.open(binfile)
+        except Exception as e:
+            errstr = 'Error rendering image:\n' + str(e)
+            return draw_text(errstr, width, height)
+        imgwidth, imgheight = get_pix_dimensions(width, height)
+        nimg = Image.new(mode='RGBA', size=(imgwidth, imgheight))
+        imgwidth -= 4
+        imgheight -= 4
+        hscalefact = imgwidth / img.width
+        vscalefact = imgheight / img.height
+        if hscalefact < vscalefact:
+            rzwidth = imgwidth
+            rzheight = int(img.height * hscalefact)
+        else:
+            rzwidth = int(img.width * vscalefact)
+            rzheight = imgheight
+        img = img.resize((rzwidth, rzheight))
+        nd = ImageDraw.Draw(nimg)
+        nd.rectangle((1, 1, rzwidth + 2, rzheight + 2), outline='black')
+        nd.rectangle((0, 0, rzwidth + 3, rzheight + 3), outline='white')
+        nimg.paste(img, box=(2, 2))
+        outfile = io.BytesIO()
+        nimg.save(outfile, format='PNG')
+        data = base64.b64encode(outfile.getbuffer())
+    if imageformat == 'sixel':
+        sixel_draw(data)
+    elif imageformat == 'iterm':
+        iterm_draw(data, width, height)
+    else:
+        kitty_draw(data, width, height)
+
+
+def sixel_draw(data):
+    bindata = base64.b64decode(data)
+    binfile = io.BytesIO()
+    binfile.write(bindata)
+    binfile.seek(0)
+    DumbWriter().draw(binfile)
+
+def iterm_draw(data, width, height):
+    if not height:
+        height = 'auto'
+    if not width:
+        width = 'auto'
+    bindata = base64.b64decode(data)
+    datalen = len(bindata)
+    sys.stdout.write(
+        '\x1b]1337;File=inline=1;width={};height={};size={}:'.format(width,height,datalen))
+    sys.stdout.write(data.decode('utf8'))
+    sys.stdout.write('\a')
+    sys.stdout.flush()
+
+
+def kitty_draw(data, width, height):
+    preamble = '\x1b_Ga=T,f=100'
+    if height:
+        preamble += f',r={height},c={width}'
+    #sys.stdout.write(repr(preamble))
+    #sys.stdout.write('\xb[{}D'.format(len(repr(preamble))))
+    #return
+    first = True
+    while data:
+        chunk, data = data[:4096], data[4096:]
+        m = 1 if data else 0
+        if first:
+            sys.stdout.write('{},m={};'.format(preamble, m))
+        else:
+            sys.stdout.write('\x1b_Gm={};'.format(m))
+        sys.stdout.write(chunk.decode('utf8'))
+        sys.stdout.write('\x1b\\')
+        sys.stdout.flush()
+
 pass_through_args = []
 killcon = False
 try:
@@ -106,6 +363,201 @@ if options.Timestamp:
     logreader.dump_to_console(logname)
     sys.exit(0)
 
+def prep_node_tile(node):
+    currcolcell, currrowcell = nodepositions[node]
+    if currcolcell:
+        cursor_right(currcolcell)
+    if currrowcell:
+        cursor_down(currrowcell)
+    sys.stdout.write('▏' + node)
+    cursor_left(len(node) + 1)
+    cursor_down()
+
+def reset_cursor(node):
+    currcolcell, currrowcell = nodepositions[node]
+    if currcolcell:
+        cursor_left(currcolcell)
+    cursor_up(currrowcell + 1)
+
+nodepositions = {}
+numrows = 0
+cwidth = 0
+cheight = 0
+imagedatabynode = {}
+
+def redraw():
+    for node in imagedatabynode:
+        imgdata = imagedatabynode[node]
+        if node in nodepositions:
+            prep_node_tile(node)
+            cursor_save()
+        else:
+            if options.interval is not None:
+                if node != firstnodename:
+                    sys.stderr.write('Multiple nodes not supported for interval')
+                    sys.exit(1)
+                sticky_cursor()
+            sys.stdout.write('{}: '.format(node))
+        # one row is used by our own name, so cheight - 1 for that allowance
+        draw_image(imgdata.encode(), cwidth, cheight - 1 if cheight else cheight)
+        if node in nodepositions:
+            cursor_restore()
+            reset_cursor(node)
+        else:
+            sys.stdout.write('\n')
+        sys.stdout.flush()
+resized = False
+def do_screenshot():
+    global resized
+    global numrows
+    sess = client.Command()
+    if options.tile:
+        imageformat = os.environ.get('CONFLUENT_IMAGE_PROTOCOL', 'kitty')
+        if imageformat not in ('kitty', 'iterm'):
+            sys.stderr.write('Tiled screenshots only supported with kitty or iterm protocol')
+            sys.exit(1)
+        allnodes = []
+        numnodes = 0
+        for res in sess.read('/noderange/{}/nodes/'.format(args[0])):
+            allnodes.append(res['item']['href'].replace('/', ''))
+            numnodes += 1
+        resized = False
+        def do_resize(a=None, b=None):
+            global resized
+            if a:
+                resized = True
+                # on a window resize, clear the old stuff
+                # ideally we'd retain the images and redraw them
+                sys.stdout.write('\x1bc')
+            global numrows
+            global cwidth
+            global cheight
+            cols, rows, cwidth, cheight, numrows = determine_tile_size(numnodes)
+            currcol = 1
+            currcolcell = 0
+            currrowcell = 0
+            for node in allnodes:
+                nodepositions[node] = currcolcell, currrowcell
+                if currcol < cols:
+                    currcol += 1
+                    currcolcell += cwidth
+                else:
+                    currcol = 1
+                    currcolcell = 0
+                    currrowcell += cheight
+            if a:
+                redraw()
+        do_resize()
+        signal.signal(signal.SIGWINCH, do_resize)
+    elif options.interval is not None:
+        sys.stdout.write('\x1bc')
+    firstnodename = None
+    dorefresh = True
+    vnconly = set([])
+    while dorefresh:
+        for res in sess.read('/noderange/{}/console/ikvm_screenshot'.format(args[0])):
+            for node in res.get('databynode', {}):
+                errorstr = ''
+                if not firstnodename:
+                    firstnodename = node
+                error = res['databynode'][node].get('error')
+                if error and 'vnc available' in error:
+                    vnconly.add(node)
+                    continue
+                elif error:
+                    errorstr = error
+                imgdata = res['databynode'][node].get('image', {}).get('imgdata', None)
+                if imgdata:
+                    if len(imgdata) < 32: # We were subjected to error
+                        errorstr = f'Unable to get screenshot'
+                if errorstr or imgdata:
+                    draw_node(node, imgdata, errorstr, firstnodename, cwidth, cheight)
+        if asyncvnc:
+            urlbynode = {}
+            for node in vnconly:
+                for res in sess.update(f'/nodes/{node}/console/ikvm', {'method': 'unix'}):
+                    url = res.get('item', {}).get('href')
+                    if url:
+                        urlbynode[node] = url
+            draw_vnc_grabs(urlbynode, cwidth, cheight)
+            if resized:
+                do_resize(True)
+                resized = False
+        elif vnconly:
+            sys.stderr.write("Require asyncvnc installed to do VNC screenshotting\n")
+        if options.interval is None:
+            dorefresh = False
+        else:
+            dorefresh = True
+            time.sleep(options.interval)
+    sys.exit(0)
+
+try:
+    import asyncio, asyncvnc
+except ImportError:
+    asyncvnc = None
+
+def draw_vnc_grabs(urlbynode, cwidth, cheight):
+    asyncio.run(grab_vncs(urlbynode, cwidth, cheight))
+async def grab_vncs(urlbynode, cwidth, cheight):
+    tasks = []
+    for node in urlbynode:
+        url = urlbynode[node]
+        tasks.append(asyncio.create_task(do_vnc_screenshot(node, url, cwidth, cheight)))
+    await asyncio.gather(*tasks)
+
+async def my_opener(host, port):
+    # really, host is the unix
+    return await asyncio.open_unix_connection(host)
+
+async def do_vnc_screenshot(node, url, cwidth, cheight):
+    async with asyncvnc.connect(url, opener=my_opener) as client:
+        # Retrieve pixels as a 3D numpy array
+        pixels = await client.screenshot()
+        # Save as PNG using PIL/pillow
+        image = Image.fromarray(pixels)
+        outfile = io.BytesIO()
+        image.save(outfile, format='PNG')
+        imgdata = base64.b64encode(outfile.getbuffer()).decode()
+        if imgdata:
+            draw_node(node, imgdata, '', '', cwidth, cheight)
+
+def draw_node(node, imgdata, errorstr, firstnodename, cwidth, cheight):
+    imagedatabynode[node] = imgdata
+    if node in nodepositions:
+        prep_node_tile(node)
+        cursor_save()
+    else:
+        if options.interval is not None:
+            if node != firstnodename:
+                sys.stderr.write('Multiple nodes not supported for interval')
+                sys.exit(1)
+            sticky_cursor()
+        sys.stdout.write('{}: '.format(node))
+    # one row is used by our own name, so cheight - 1 for that allowance
+    if errorstr:
+        draw_text(errorstr, cwidth, cheight -1 if cheight else cheight)
+    else:
+        draw_image(imgdata.encode(), cwidth, cheight - 1 if cheight else cheight)
+    if node in nodepositions:
+        cursor_restore()
+        reset_cursor(node)
+    else:
+        sys.stdout.write('\n')
+    sys.stdout.flush()
+
+if options.screenshot:
+    try:
+        cursor_hide()
+        do_screenshot()
+    except KeyboardInterrupt:
+        pass
+    finally:
+        cursor_show()
+        cursor_down(numrows)
+        sys.stdout.write('\n')
+    sys.exit(0)
+
 def kill(noderange):
     sess = client.Command()
     envstring=os.environ.get('NODECONSOLE_WINDOWED_COMMAND')
@@ -113,29 +565,23 @@ def kill(noderange):
         envstring = 'xterm'
 
     nodes = []
-    for res in sess.read('/noderange/{0}/nodes/'.format(args[0])):
+    for res in sess.read('/noderange/{0}/nodes/'.format(noderange)):
         node = res.get('item', {}).get('href', '/').replace('/', '')
         if not node:
             sys.stderr.write(res.get('error', repr(res)) + '\n')
             sys.exit(1)
         nodes.append(node) 
 
-    for node in nodes:
-        s=socket.socket(socket.AF_UNIX)
-        winid=None
+    for node in nodes:    
+        command = "ps auxww | grep {0} | grep console | egrep '\\b{1}\\b' | grep -v grep | awk '{{print $2}}'".format(envstring, node)
+        process = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        stdout, stderr = process.communicate()
         try:
-            win=subprocess.Popen(['xwininfo', '-tree', '-root'], stdout=subprocess.PIPE)
-            wintr=win.communicate()[0]
-            for line in wintr.decode('utf-8').split('\n'):
-                if 'console: {0}'.format(node) in line or 'confetty' in line:
-                    win_obj = [ele for ele in line.split(' ') if ele.strip()]
-                    winid = win_obj[0]
-        except:
-            print("Error: cannot retrieve window id of node {}".format(node))
-
-        if winid:
-            ps_data=subprocess.Popen(['xkill', '-id', winid ], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-
+            process_id = stdout.decode('utf-8').split()[0]
+        except IndexError:
+            sys.stderr.write(node + ": console window not found \n")
+            continue
+        subprocess.Popen(["kill", process_id], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
     sys.exit(0)
 
 def handle_geometry(envlist, sizegeometry, side_pad=0, top_pad=0, first=False):    
@@ -257,7 +703,6 @@ if options.windowed:
         screenheight -= wmyo
         currx = window_width
         curry = 0
-        maxcol = int(screenwidth/window_width)
 
     for node in sortutil.natural_sort(nodes):
         if options.tile and envlist[0] == 'xterm':
@@ -265,7 +710,7 @@ if options.windowed:
             corrected_y = curry
             xgeometry = '{0}+{1}+{2}'.format(sizegeometry, corrected_x, corrected_y)
             currx += window_width
-            if currx >= screenwidth:
+            if currx + window_width >= screenwidth:
                 currx=0
                 curry += window_height
                 if curry > screenheight:

confluent_client/bin/nodedeploy

@@ -48,7 +48,18 @@ def armonce(nr, cli):
         pass
 
 
-def setpending(nr, profile, cli):
+def setpending(nr, profile, profilebynodes, cli):
+    if profilebynodes:
+        for node in sortutil.natural_sort(profilebynodes):
+            prof = profilebynodes[node]
+            args = {'deployment.pendingprofile': prof, 'deployment.state': '', 'deployment.state_detail': ''}
+            if not prof.startswith('genesis-'):
+                args['deployment.stagedprofile'] = ''
+                args['deployment.profile'] = ''
+            for rsp in cli.update('/nodes/{0}/attributes/current'.format(node),
+                                  args):
+                pass
+        return
     args = {'deployment.pendingprofile': profile, 'deployment.state': '', 'deployment.state_detail': ''}
     if not profile.startswith('genesis-'):
         args['deployment.stagedprofile'] = ''
@@ -69,6 +80,7 @@ def main(args):
     ap.add_argument('-n', '--network', help='Initiate deployment over PXE/HTTP', action='store_true')
     ap.add_argument('-p', '--prepareonly', help='Prepare only, skip any interaction with a BMC associated with this deployment action', action='store_true')
     ap.add_argument('-m', '--maxnodes', help='Specifiy a maximum nodes to be deployed')
+    ap.add_argument('-r', '--redeploy', help='Redeploy nodes with the current or pending profile', action='store_true')
     ap.add_argument('noderange', help='Set of nodes to deploy')
     ap.add_argument('profile', nargs='?', help='Profile name to deploy')
     args, extra = ap.parse_known_args(args)
@@ -78,7 +90,7 @@ def main(args):
     if args.profile and not args.network:
         sys.stderr.write('-n is a required argument currently to perform an install, optionally with -p\n')
         return 1
-    if not args.profile and args.network:
+    if not args.profile and args.network and not args.redeploy:
         sys.stderr.write('Both noderange and a profile name are required arguments to request a network deployment\n')
         return 1
     if args.clear and args.profile:
@@ -96,29 +108,68 @@ def main(args):
         if 'error' in rsp:
             sys.stderr.write(rsp['error'] + '\n')
             sys.exit(1)
+    profilebynode = {}
     if args.clear:
         cleararm(args.noderange, c)
         clearpending(args.noderange, c)
-    elif args.profile:
-        profnames = []
-        for prof in c.read('/deployment/profiles/'):
-            profname = prof.get('item', {}).get('href', None)
-            if profname:
-                profname = profname.replace('/', '')
-                profnames.append(profname)
-                if profname == args.profile:
-                    break
-        else:
-            sys.stderr.write('The specified profile "{}" is not an available profile\n'.format(args.profile))
-            if profnames:
-                sys.stderr.write('The following profiles are available:\n')
-                for profname in profnames:
-                    sys.stderr.write('    ' + profname + '\n')
-            else:
-                sys.stderr.write('No deployment profiles available, try osdeploy import or imgutil capture\n')
+    elif args.redeploy:
+        hadpending = {}
+        for rsp in c.read('/noderange/{0}/attributes/current'.format(args.noderange)):
+            for node in rsp.get('databynode', {}):
+                nodeinfo = rsp['databynode'][node]
+                for attr in nodeinfo:
+                    if attr == 'deployment.pendingprofile':
+                        curr = nodeinfo[attr].get('value', '')
+                        if curr:
+                            hadpending[node] = True
+                            profilebynode[node] = curr
+                    if attr == 'deployment.stagedprofile':
+                        curr = nodeinfo[attr].get('value', '')
+                        if curr and node not in hadpending:
+                            profilebynode[node] = curr
+                    if attr == 'deployment.profile':
+                        curr = nodeinfo[attr].get('value', '')
+                        if curr and node not in profilebynode:
+                            profilebynode[node] = curr
+        for lockinfo in c.read('/noderange/{0}/deployment/lock'.format(args.noderange)):
+            for node in lockinfo.get('databynode', {}):
+                lockstate = lockinfo['databynode'][node]['lock']['value']
+                if lockstate == 'locked':
+                    lockednodes.append(node)     
+    if args.profile and profilebynode:
+        sys.stderr.write('The -r/--redeploy option cannot be used with a profile, it redeploys the current or pending profile\n')
+        return 1
+    if args.profile or profilebynode:
+        lockednodes = []
+        for lockinfo in c.read('/noderange/{0}/deployment/lock'.format(args.noderange)):
+            for node in lockinfo.get('databynode', {}):
+                lockstate = lockinfo['databynode'][node]['lock']['value']
+                if lockstate == 'locked':
+                    lockednodes.append(node)
+        if lockednodes:
+            sys.stderr.write('Requested noderange has nodes with locked deployment: ' + ','.join(lockednodes))
+            sys.stderr.write('\n')
             sys.exit(1)
+        if args.profile:
+            profnames = []
+            for prof in c.read('/deployment/profiles/'):
+                profname = prof.get('item', {}).get('href', None)
+                if profname:
+                    profname = profname.replace('/', '')
+                    profnames.append(profname)
+                    if profname == args.profile:
+                        break
+            else:
+                sys.stderr.write('The specified profile "{}" is not an available profile\n'.format(args.profile))
+                if profnames:
+                    sys.stderr.write('The following profiles are available:\n')
+                    for profname in profnames:
+                        sys.stderr.write('    ' + profname + '\n')
+                else:
+                    sys.stderr.write('No deployment profiles available, try osdeploy import or imgutil capture\n')
+                sys.exit(1)
         armonce(args.noderange, c)
-        setpending(args.noderange, args.profile, c)
+        setpending(args.noderange, args.profile, profilebynode, c)
     else:
         databynode = {}
         for r in c.read('/noderange/{0}/attributes/current'.format(args.noderange)):

confluent_client/bin/nodediscover

@@ -123,7 +123,7 @@ def process_header(header):
             fields.append('serial')
         elif datum == 'uuid':
             fields.append('uuid')
-        elif datum in ('bmc', 'imm', 'xcc'):
+        elif datum in ('bmc', 'imm', 'xcc', 'ip'):
             fields.append('hardwaremanagement.manager')
         elif datum in ('bmc gateway', 'xcc gateway', 'imm gateway'):
             fields.append('net.bmc.ipv4_gateway')
@@ -191,6 +191,7 @@ def import_csv(options, session):
             if field in unique_fields:
                 unique_data[field] = set([])
         broken = False
+        alldata=[]
         for record in records:
             currfields = list(fields)
             nodedatum = {}
@@ -207,9 +208,15 @@ def import_csv(options, session):
                 nodedatum[currfield] = datum
             if not datum_complete(nodedatum):
                 sys.exit(1)
+            alldata.append(nodedatum)
+        allthere = True
+        for nodedatum in alldata:
             if not search_record(nodedatum, options, session) and not broken:
+                allthere = False
                 blocking_scan(session)
-            if not search_record(nodedatum, options, session):
+                break
+        for nodedatum in alldata:
+            if not allthere and not search_record(nodedatum, options, session):
                 sys.stderr.write(
                     "Could not match the following data: " +
                     repr(nodedatum) + '\n')
@@ -230,8 +237,12 @@ def import_csv(options, session):
                 print('Defined ' + res['created'])
             else:
                 print(repr(res))
+        child = os.fork()
+        if child:
+            continue
         for mac in maclist:
-            for res in session.update('/discovery/by-mac/{0}'.format(mac),
+            mysess = client.Command()
+            for res in mysess.update('/discovery/by-mac/{0}'.format(mac),
                                       {'node': nodename}):
                 if 'error' in res:
                     sys.stderr.write(res['error'] + '\n')
@@ -241,6 +252,12 @@ def import_csv(options, session):
                     print('Discovered ' + res['assigned'])
                 else:
                     print(repr(res))
+        sys.exit(0)
+    while True:
+        try:
+            os.wait()
+        except ChildProcessError:
+            break
     if exitcode:
         sys.exit(exitcode)
 

confluent_client/bin/nodeeventlog

@@ -78,8 +78,11 @@ exitcode = 0
 def format_event(evt):
     retparts = []
     if 'timestamp' in evt and evt['timestamp'] is not None:
-        display = dt.strptime(evt['timestamp'], '%Y-%m-%dT%H:%M:%S')
-        retparts.append(display.strftime('%m/%d/%Y %H:%M:%S'))
+        try:
+            display = dt.strptime(evt['timestamp'], '%Y-%m-%dT%H:%M:%S')
+            retparts.append(display.strftime('%m/%d/%Y %H:%M:%S'))
+        except ValueError:
+            display = ''    
     dscparts = []
     if evt.get('log_id', None):
         retparts.append(evt['log_id'] + ':')

confluent_client/bin/nodefirmware

@@ -1,4 +1,4 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
 # Copyright 2016-2017 Lenovo
@@ -56,7 +56,7 @@ components = ['all']
 
 argparser = optparse.OptionParser(
     usage="Usage: "
-          "%prog <noderange> [list][update [--backup <file>]]|[<components>]")
+          "%prog <noderange> [list][updatestatus][update [--backup <file>]]|[<components>]")
 argparser.add_option('-b', '--backup', action='store_true',
                      help='Target a backup bank rather than primary')
 argparser.add_option('-m', '--maxnodes', type='int',
@@ -65,14 +65,18 @@ argparser.add_option('-m', '--maxnodes', type='int',
              
 (options, args) = argparser.parse_args()
 upfile = None
+querystatus = False
 try:
     noderange = args[0]
     if len(args) > 1:
         if args[1] == 'update':
             upfile = args[2]
         else:
+            comps = []
             if args[1] == 'list':
                 comps = args[2:]
+            elif args[1] == 'updatestatus':
+                querystatus = True
             else:
                 comps = args[1:]
             components = []
@@ -110,11 +114,24 @@ def update_firmware(session, filename):
         upargs['bank'] = 'backup'
     noderrs = {}
     if session.unixdomain:
-        of = open(filename, 'rb')
-        try:
-            session.add_file(filename, of.fileno(), 'rb')
-        except Exception:
-            pass
+        filesbynode = {}
+        for exp in session.create('/noderange/{0}/attributes/expression'.format(noderange),
+                        {'expression': filename}):
+            if 'error' in exp:
+                sys.stderr.write(exp['error'] + '\n')
+                exitcode |= exp.get('errorcode', 1)
+            ex = exp.get('databynode', ())
+            for node in ex:
+                filesbynode[node] = ex[node]['value']
+                if not isinstance(filesbynode[node], bytes) and not isinstance(filesbynode[node], str):
+                    filesbynode[node] = filesbynode[node].encode('utf-8')
+        for node in filesbynode:
+            endfilename = filesbynode[node]
+            of = open(endfilename, 'rb')
+            try:
+                session.add_file(endfilename, of.fileno(), 'rb')
+            except Exception:
+                pass
     for res in session.create(resource, upargs):
         if 'created' not in res:
             for nodename in res.get('databynode', ()):
@@ -149,9 +166,13 @@ def show_firmware(session):
     firmware_shown = False
     nodes_matched = False
     for component in components:
+        category = 'all'
+        if component in ('adapters', 'disks', 'misc', 'core'):
+            category = component
+            component = 'all'    
         for res in session.read(
-                '/noderange/{0}/inventory/firmware/all/{1}'.format(
-                    noderange, component)):
+                '/noderange/{0}/inventory/firmware/{2}/{1}'.format(
+                    noderange, component, category)):
             nodes_matched = True
             exitcode |= client.printerror(res)
             if 'databynode' not in res:
@@ -171,7 +192,16 @@ def show_firmware(session):
 
 try:
     session = client.Command()
-    if upfile is None:
+    if querystatus:
+        for res in session.read(
+                '/noderange/{0}/inventory/firmware/updatestatus'.format(noderange)):
+            for node in res.get('databynode', {}):
+                currstat = res['databynode'][node].get('status', None)
+                if currstat:
+                    print('{}: {}'.format(node, currstat))
+                else:
+                    print(repr(res))
+    elif upfile is None:
         show_firmware(session)
     else:
         update_firmware(session, upfile)

confluent_client/bin/nodeinventory

@@ -49,7 +49,9 @@ def pretty(text):
 
 def print_mem_info(node, prefix, meminfo):
     memdescfmt = '{0}GB PC'
-    if meminfo['memory_type'] == 'DDR3 SDRAM':
+    if meminfo['memory_type'] is None:
+        memdescfmt = '{0}GB '
+    elif meminfo['memory_type'] == 'DDR3 SDRAM':
         memdescfmt += '3-{1} '
     elif 'DDR4' in meminfo['memory_type']:
         memdescfmt += '4-{1} '

confluent_client/bin/nodelicense

@@ -1,4 +1,4 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
 # Copyright 2019 Lenovo
@@ -65,16 +65,30 @@ client.check_globbing(noderange)
 
 def install_license(session, filename):
     global exitcode
+    
     resource = '/noderange/{0}/configuration/' \
                'management_controller/licenses/'.format(noderange)
     filename = os.path.abspath(filename)
     instargs = {'filename': filename}
     if session.unixdomain:
-        of = open(filename, 'rb')
-        try:
-            session.add_file(filename, of.fileno(), 'rb')
-        except Exception:
-            pass
+        filesbynode = {}
+        for exp in session.create('/noderange/{0}/attributes/expression'.format(noderange),
+                        {'expression': filename}):
+            if 'error' in exp:
+                sys.stderr.write(exp['error'] + '\n')
+                exitcode |= exp.get('errorcode', 1)
+            ex = exp.get('databynode', ())
+            for node in ex:
+                filesbynode[node] = ex[node]['value']
+                if not isinstance(filesbynode[node], bytes) and not isinstance(filesbynode[node], str):
+                    filesbynode[node] = filesbynode[node].encode('utf-8')
+        for node in filesbynode:
+            endfilename = filesbynode[node]
+            of = open(endfilename, 'rb')
+            try:
+                session.add_file(endfilename, of.fileno(), 'rb')
+            except Exception:
+                pass
     for res in session.create(resource, instargs):
         for node in res.get('databynode', []):
             if 'error' in res['databynode'][node]:

confluent_client/bin/nodemedia

@@ -1,4 +1,4 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
 # Copyright 2018 Lenovo
@@ -101,22 +101,37 @@ def detach_media(noderange, media):
 
 def upload_media(noderange, media):
     global exitcode
-    if not os.path.exists(media):
-        sys.stderr.write('Unable to locate requested file {0}\n'.format(
-            media))
-        sys.exit(404)
+
     session = client.Command()
     output = sq.ScreenPrinter(noderange, session)
     filename = os.path.abspath(media)
     resource = '/noderange/{0}/media/uploads/'.format(noderange)
+    filename = os.path.abspath(filename)
     upargs = {'filename': filename}
     noderrs = {}
     if session.unixdomain:
-        of = open(filename, 'rb')
-        try:
-            session.add_file(filename, of.fileno(), 'rb')
-        except Exception:
-            pass
+        filesbynode = {}
+        for exp in session.create('/noderange/{0}/attributes/expression'.format(noderange),
+                        {'expression': filename}):
+            if 'error' in exp:
+                sys.stderr.write(exp['error'] + '\n')
+                exitcode |= exp.get('errorcode', 1)
+            ex = exp.get('databynode', ())
+            for node in ex:
+                filesbynode[node] = ex[node]['value']
+                if not isinstance(filesbynode[node], bytes) and not isinstance(filesbynode[node], str):
+                    filesbynode[node] = filesbynode[node].encode('utf-8')
+        for node in filesbynode:
+            endfilename = filesbynode[node]
+            if not os.path.exists(endfilename):
+                sys.stderr.write('Unable to locate requested file {0}\n'.format(
+                    endfilename))
+                sys.exit(404)
+            of = open(endfilename, 'rb')
+            try:
+                session.add_file(endfilename, of.fileno(), 'rb')
+            except Exception:
+                pass
     nodeurls = {}
     for res in session.create(resource, upargs):
         if 'created' not in res:

confluent_client/bin/nodeping

@@ -1,4 +1,4 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
 # Copyright 2016-2017 Lenovo
@@ -42,6 +42,8 @@ def run():
         usage="Usage: %prog [options] noderange")
     argparser.add_option('-f', '-c', '--count', type='int', default=168,
                          help='Number of commands to run at a time')
+    argparser.add_option('-o', '--origname', action='store_true',
+                         help='Use original nodename in print out even if substituted')
     argparser.add_option('-s', '--substitutename',
                          help='Use a different name other than the nodename for ping, with {}, it is the entire name evaluated as an expression, otherwise it is used as a suffix')
     # among other things, FD_SETSIZE limits.  Besides, spawning too many
@@ -56,6 +58,7 @@ def run():
 
     currprocs = 0
     all = set([])
+    poller = select.epoll()
     pipedesc = {}
     pendingexecs = deque()
     exitcode = 0
@@ -83,19 +86,29 @@ def run():
             cmdv = ['ping', '-c', '1', '-W', '1', pingnode]
             if currprocs < concurrentprocs:
                 currprocs += 1
-                run_cmdv(pingnode, cmdv, all, pipedesc)
+                if options.origname:
+                    run_cmdv(node, cmdv, all, poller, pipedesc)
+                else:
+                    run_cmdv(pingnode, cmdv, all, poller, pipedesc)
             else:
-                pendingexecs.append((pingnode, cmdv))
+                if options.origname:
+                    pendingexecs.append((node, cmdv))
+                else:
+                    pendingexecs.append((pingnode, cmdv))
     if not all or exitcode:
         sys.exit(exitcode)
-    rdy, _, _ = select.select(all, [], [], 10)
+    rdy = poller.poll(10)
     while all:
         pernodeout = {}
         for r in rdy:
+            r = r[0]
             desc = pipedesc[r]
+            r = desc['file']
             node = desc['node']
             data = True
-            while data and select.select([r], [], [], 0)[0]:
+            singlepoller = select.epoll()
+            singlepoller.register(r, select.EPOLLIN)
+            while data and singlepoller.poll(0):
                 data = r.readline()
                 if not data:
                     pop = desc['popen']
@@ -103,6 +116,7 @@ def run():
                     if ret is not None:
                         exitcode = exitcode | ret
                         all.discard(r)
+                        poller.unregister(r)
                         r.close()
                         if desc['type'] == 'stdout':
                             if ret:
@@ -111,7 +125,8 @@ def run():
                                 print('{0}: ping'.format(node))
                             if pendingexecs:
                                 node, cmdv = pendingexecs.popleft()
-                                run_cmdv(node, cmdv, all, pipedesc)
+                                run_cmdv(node, cmdv, all, poller, pipedesc)
+            singlepoller.close()
         for node in sortutil.natural_sort(pernodeout):
             for line in pernodeout[node]:
                 line = client.stringify(line)
@@ -121,19 +136,21 @@ def run():
                     sys.stdout.write('{0}: {1}'.format(node, line))
         sys.stdout.flush()
         if all:
-            rdy, _, _ = select.select(all, [], [], 10)
+            rdy = poller.poll(10)
     sys.exit(exitcode)
 
 
-def run_cmdv(node, cmdv, all, pipedesc):
+def run_cmdv(node, cmdv, all, poller, pipedesc):
     nopen = subprocess.Popen(
         cmdv, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-    pipedesc[nopen.stdout] = {'node': node, 'popen': nopen,
-                              'type': 'stdout'}
-    pipedesc[nopen.stderr] = {'node': node, 'popen': nopen,
-                              'type': 'stderr'}
+    pipedesc[nopen.stdout.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stdout', 'file': nopen.stdout}
+    pipedesc[nopen.stderr.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stderr', 'file': nopen.stderr}
     all.add(nopen.stdout)
+    poller.register(nopen.stdout, select.EPOLLIN)
     all.add(nopen.stderr)
+    poller.register(nopen.stderr, select.EPOLLIN)
 
 
 if __name__ == '__main__':

confluent_client/bin/noderun

@@ -1,4 +1,4 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
 # Copyright 2016-2017 Lenovo
@@ -67,6 +67,7 @@ def run():
 
     currprocs = 0
     all = set([])
+    poller = select.epoll()
     pipedesc = {}
     pendingexecs = deque()
     exitcode = 0
@@ -84,19 +85,23 @@ def run():
             cmdv = shlex.split(cmd)
             if currprocs < concurrentprocs:
                 currprocs += 1
-                run_cmdv(node, cmdv, all, pipedesc)
+                run_cmdv(node, cmdv, all, poller, pipedesc)
             else:
                 pendingexecs.append((node, cmdv))
     if not all or exitcode:
         sys.exit(exitcode)
-    rdy, _, _ = select.select(all, [], [], 10)
+    rdy = poller.poll(10)
     while all:
         pernodeout = {}
         for r in rdy:
+            r = r[0]
             desc = pipedesc[r]
+            r = desc['file']
             node = desc['node']
             data = True
-            while data and select.select([r], [], [], 0)[0]:
+            singlepoller = select.epoll()
+            singlepoller.register(r, select.EPOLLIN)
+            while data and singlepoller.poll(0):
                 data = r.readline()
                 if data:
                     if desc['type'] == 'stdout':
@@ -116,10 +121,12 @@ def run():
                     if ret is not None:
                         exitcode = exitcode | ret
                         all.discard(r)
+                        poller.unregister(r)
                         r.close()
                         if desc['type'] == 'stdout' and pendingexecs:
                             node, cmdv = pendingexecs.popleft()
-                            run_cmdv(node, cmdv, all, pipedesc)
+                            run_cmdv(node, cmdv, all, poller, pipedesc)
+            singlepoller.close()
         for node in sortutil.natural_sort(pernodeout):
             for line in pernodeout[node]:
                 line = client.stringify(line)
@@ -129,11 +136,11 @@ def run():
                     sys.stdout.write('{0}: {1}'.format(node, line))
         sys.stdout.flush()
         if all:
-            rdy, _, _ = select.select(all, [], [], 10)
+            rdy = poller.poll(10)
     sys.exit(exitcode)
 
 
-def run_cmdv(node, cmdv, all, pipedesc):
+def run_cmdv(node, cmdv, all, poller, pipedesc):
     try:
         nopen = subprocess.Popen(
             cmdv, stdin=devnull, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
@@ -142,13 +149,14 @@ def run_cmdv(node, cmdv, all, pipedesc):
             sys.stderr.write('{0}: Unable to find local executable file "{1}"'.format(node, cmdv[0]))
             return
         raise
-    pipedesc[nopen.stdout] = {'node': node, 'popen': nopen,
-                              'type': 'stdout'}
-    pipedesc[nopen.stderr] = {'node': node, 'popen': nopen,
-                              'type': 'stderr'}
+    pipedesc[nopen.stdout.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stdout', 'file': nopen.stdout}
+    pipedesc[nopen.stderr.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stderr', 'file': nopen.stderr}
     all.add(nopen.stdout)
+    poller.register(nopen.stdout, select.EPOLLIN)
     all.add(nopen.stderr)
-
+    poller.register(nopen.stderr, select.EPOLLIN)
 
 if __name__ == '__main__':
     run()

confluent_client/bin/nodeshell

@@ -1,4 +1,4 @@
-#!/usr/bin/python2
+#!/usr/bin/python3
 # vim: tabstop=4 shiftwidth=4 softtabstop=4
 
 # Copyright 2016-2017 Lenovo
@@ -109,6 +109,7 @@ def run():
             ex = exp.get('databynode', ())
             for node in ex:
                 cmdparms.append((node, ex[node]['value']))
+    poller = select.epoll()
     for node, cmd in cmdparms:
         sshnode = nodemap.get(node, node)
         if not isinstance(cmd, str) and not isinstance(cmd, bytes):
@@ -121,19 +122,23 @@ def run():
         cmdv += [sshnode, cmd]
         if currprocs < concurrentprocs:
             currprocs += 1
-            run_cmdv(node, cmdv, all, pipedesc)
+            run_cmdv(node, cmdv, all, poller, pipedesc)
         else:
             pendingexecs.append((node, cmdv))
     if not all or exitcode:
         sys.exit(exitcode)
-    rdy, _, _ = select.select(all, [], [], 10)
+    rdy = poller.poll(10)
     while all:
         pernodeout = {}
         for r in rdy:
+            r = r[0]
             desc = pipedesc[r]
+            r = desc['file']
             node = desc['node']
             data = True
-            while data and select.select([r], [], [], 0)[0]:
+            singlepoller = select.epoll()
+            singlepoller.register(r, select.EPOLLIN)
+            while data and singlepoller.poll(0):
                 data = r.readline()
                 if data:
                     if desc['type'] == 'stdout':
@@ -153,10 +158,12 @@ def run():
                     if ret is not None:
                         exitcode = exitcode | ret
                         all.discard(r)
+                        poller.unregister(r)
                         r.close()
                         if desc['type'] == 'stdout' and pendingexecs:
                             node, cmdv = pendingexecs.popleft()
-                            run_cmdv(node, cmdv, all, pipedesc)
+                            run_cmdv(node, cmdv, all, poller, pipedesc)
+            singlepoller.close()
         for node in sortutil.natural_sort(pernodeout):
             for line in pernodeout[node]:
                 line = client.stringify(line)
@@ -167,19 +174,21 @@ def run():
                     sys.stdout.write('{0}: {1}'.format(node, line))
         sys.stdout.flush()
         if all:
-            rdy, _, _ = select.select(all, [], [], 10)
+            rdy = poller.poll(10)
     sys.exit(exitcode)
 
 
-def run_cmdv(node, cmdv, all, pipedesc):
+def run_cmdv(node, cmdv, all, poller, pipedesc):
     nopen = subprocess.Popen(
         cmdv, stdin=devnull, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-    pipedesc[nopen.stdout] = {'node': node, 'popen': nopen,
-                              'type': 'stdout'}
-    pipedesc[nopen.stderr] = {'node': node, 'popen': nopen,
-                              'type': 'stderr'}
+    pipedesc[nopen.stdout.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stdout', 'file': nopen.stdout}
+    pipedesc[nopen.stderr.fileno()] = {'node': node, 'popen': nopen,
+                              'type': 'stderr', 'file': nopen.stderr}
     all.add(nopen.stdout)
     all.add(nopen.stderr)
+    poller.register(nopen.stdout, select.EPOLLIN)
+    poller.register(nopen.stderr, select.EPOLLIN)
 
 
 if __name__ == '__main__':

confluent_client/bin/stats

@@ -19,6 +19,7 @@ import argparse
 import base64
 import csv
 import io
+import os
 import numpy as np
 import sys
 
@@ -72,6 +73,8 @@ def plot(gui, output, plotdata, bins, fmt):
             tdata = io.BytesIO()
         plt.savefig(tdata)
     if not gui and not output:
+        if fmt == 'environment':
+            fmt = os.environ.get('CONFLUENT_IMAGE_PROTOCOL', 'kitty')
         if fmt == 'sixel':
             writer = DumbWriter()
             writer.draw(tdata)
@@ -108,7 +111,7 @@ aparser = argparse.ArgumentParser(description='Quick access to common statistics
 aparser.add_argument('-c', type=int, default=0, help='Column number to analyze (default is last column)')
 aparser.add_argument('-d', default=None, help='Value used to separate columns')
 aparser.add_argument('-x', default=False, action='store_true', help='Output histogram in graphical format')
-aparser.add_argument('-f', default='sixel', help='Format for histogram output (sixel/iterm/kitty)')
+aparser.add_argument('-f', default='environment', help='Format for histogram output (sixel/iterm/kitty)')
 aparser.add_argument('-s', default=0, help='Number of header lines to skip before processing')
 aparser.add_argument('-g', default=False, action='store_true', help='Open histogram in separate graphical window')
 aparser.add_argument('-o', default=None, help='Output histogram to the specified filename in PNG format')

confluent_client/confluent/client.py

@@ -57,7 +57,7 @@ def stringify(instr):
     # Normalize unicode and bytes to 'str', correcting for
     # current python version
     if isinstance(instr, bytes) and not isinstance(instr, str):
-        return instr.decode('utf-8')
+        return instr.decode('utf-8', errors='replace')
     elif not isinstance(instr, bytes) and not isinstance(instr, str):
         return instr.encode('utf-8')
     return instr
@@ -464,8 +464,8 @@ def printattributes(session, requestargs, showtype, nodetype, noderange, options
 
 def _sort_attrib(k):
     if isinstance(k[1], dict) and k[1].get('sortid', None) is not None:
-        return k[1]['sortid']
-    return k[0]
+        return sortutil.naturalize_string('{}'.format(k[1]['sortid']))
+    return sortutil.naturalize_string(k[0])
 
 def print_attrib_path(path, session, requestargs, options, rename=None, attrprefix=None):
     exitcode = 0

confluent_client/confluent/screensqueeze.py

@@ -18,8 +18,9 @@ import struct
 import termios
 
 def get_screengeom():
-    return struct.unpack('hh', fcntl.ioctl(sys.stdout, termios.TIOCGWINSZ,
-                                           b'....'))
+    # returns height in cells, width in cells, width in pixels, height in pixels
+    return struct.unpack('hhhh', fcntl.ioctl(sys.stdout, termios.TIOCGWINSZ,
+                                           b'........'))
 class ScreenPrinter(object):
 
     def __init__(self, noderange, client, textlen=4):
@@ -58,7 +59,7 @@ class ScreenPrinter(object):
 
     def drawscreen(self, node=None):
         if self.squeeze:
-            currheight, currwidth = get_screengeom()
+            currheight, currwidth, _, _ = get_screengeom()
             currheight -= 2
             if currheight < 1:
                 currheight = 1
@@ -120,6 +121,7 @@ if __name__ == '__main__':
     c = client.Command()
     p = ScreenPrinter('d1-d12', c)
     p.set_output('d3', 'Upload: 67%')
+    p.set_output('d7', 'Upload: 67%')
 
 
 

confluent_client/confluent_client.spec.tmpl

@@ -1,7 +1,7 @@
 %define name confluent_client
 %define version #VERSION#
 %define fversion %{lua:
-sv, _ = string.gsub("#VERSION#", "[~+]", "-")
+sv, _ = string.gsub("#VERSION#", "[~]", "-")
 print(sv)
 }
 %define release 1

confluent_client/doc/man/confluent2hosts.ronn

@@ -13,7 +13,7 @@ noderange.  There are two general approaches.
 
 It can be used ad-hoc, using -i and -n to specify the address and name portions respectively.  This accepts the standard confluent expression syntax, allowing for things like 172.30.1.{n1} or {node}.{dns.domain} or {bmc}.
 
-It can also read from the confluent db, using `-a`.  In this mode, each net.<value>.<attribute> group is pulled together into hosts lines.  ipv4_address and ipv6_address fields are associated with the corresponding hostname attributes.
+It can also read from the confluent db, using `-a`.  In this mode, each net.<value>.<attribute> group is pulled together into hosts lines.  ipv4_address and ipv6_address fields are associated with the corresponding hostname attributes. You can use `-f` to put the FQDN first.
 
 ## EXAMPLES
 

confluent_client/doc/man/confluentdbutil.ronn

@@ -38,5 +38,8 @@ the json files (password protected, removed from the files, or unprotected).
   keys do not change and as such they do not require
   incremental backup.
   
+* `-y`, `--yaml
+  Use YAML instead of JSON as file format
+  
   * `-h`, `--help`:
   Show help message and exit

confluent_client/doc/man/nodeconsole.ronn

@@ -21,9 +21,15 @@ console process which will result in the console window closing.
 
 ## OPTIONS
 
+* `-i N`, `--interval`:
+  For screenshot mode, fetch new screenshots and overwrite old screenshots every N seconds.
+  For example, `nodeconsole r3u[21:24] -tsi 3` will tile screenshots of r3u21 through r3u24 and
+  refresh them every 3 seconds.
+
 * `-t`, `--tile`:
-  Use tmux to arrange consoles of the given noderange into a tiled layout on
-  the terminal screen
+  For text consoles, use tmux to arrange consoles of the given noderange into a tiled layout on
+  the terminal screen.  If using 'screenshot' mode, divide the terminal and display the images
+  in a grid. 
 
 * `-l`, `--log`:
   Perform a log reply on the current, local log in /var/log/confluent/consoles.
@@ -34,6 +40,14 @@ console process which will result in the console window closing.
  Dump the log with Timpstamps on the current, local log in /var/log/confluent/consoles.
   If in collective mode, this only makes sense to use on the current collective
   manager at this time.
+
+* `-s`, `--screenshot`:
+  Attempt to grab screenshot(s) and render using a terminal
+  image protocol. The image protocol defaults to kitty, and
+  can be selected by CONFLUENT_IMAGE_PROTOCOL environment variable.
+  Supported protocols are kitty, iterm, and, sixel (sixel only
+  if PySixel is installed).  This only presents screenshots, there
+  is no input supported to graphical consoles from a terminal.
   
 * `-w`, `--windowed`:
   Open terminal windows for each node.  The

confluent_client/doc/man/nodefirmware.ronn

@@ -3,7 +3,7 @@ nodefirmware(8) -- Report firmware information on confluent nodes
 
 ## SYNOPSIS
 
-`nodefirmware <noderange> [list][update [--backup <file>]]|[<components>]`
+`nodefirmware <noderange> [list][updatestatus][update [--backup <file>]]|[<components>]`
 
 ## DESCRIPTION
 
@@ -17,6 +17,9 @@ not be relevant to redfish.  Additionally, the Lenovo XCC makes certain
 information available over IPMI that is not otherwise available (for example
 the FPGA version where applicable).
 
+The updatestatus argument will describe the state of firmware updates on the
+nodes.
+
 In the update form, it accepts a single file and attempts to update it using
 the out of band facilities.  Firmware updates can end in one of three states:
 

confluent_client/doc/man/nodesensors.ronn

@@ -19,7 +19,9 @@ interval of 1 second is used.
 ## OPTIONS
 
 * `-c`, `--csv`:
-  Organize output into CSV format, one sensor per column.
+  Organize output into CSV format, one sensor per column. Note that while normally nodesensors reports
+  sensors in order as returned by server, CSV output enforces consistency by sorting after receiving
+  the results, which may have a different ordering than non-CSV usage of nodesensors.
 
 * `-i`, `--interval`=**SECONDS**:
   Repeat data gathering waiting, waiting the specified time between samples.  Unless `-n` is

confluent_osdeploy/common/initramfs/opt/confluent/bin/apiclient

@@ -3,6 +3,7 @@ try:
     import http.client as client
 except ImportError:
     import httplib as client
+import base64
 import ctypes
 import ctypes.util
 import glob
@@ -14,6 +15,14 @@ import ssl
 import sys
 import struct
 import time
+import re
+import hashlib
+try:
+    import json
+    import hmac
+except ImportError:
+    json = None
+    hmac = None
 
 class InvalidApiKey(Exception):
     pass
@@ -71,7 +80,7 @@ def get_my_addresses():
     return addrs
 
 
-def scan_confluents():
+def scan_confluents(confuuid=None):
     srvs = {}
     s6 = socket.socket(socket.AF_INET6, socket.SOCK_DGRAM)
     s6.setsockopt(socket.IPPROTO_IPV6, socket.IPV6_V6ONLY, 1)
@@ -83,14 +92,18 @@ def scan_confluents():
     s4.bind(('0.0.0.0', 1900))
     doneidxs = set([])
     msg = 'M-SEARCH * HTTP/1.1\r\nST: urn:xcat.org:service:confluent:'
-    with open('/etc/confluent/confluent.deploycfg') as dcfg:
-        for line in dcfg.read().split('\n'):
-            if line.startswith('confluent_uuid:'):
-                confluentuuid = line.split(': ')[1]
-                msg += '/confluentuuid=' + confluentuuid
-                break
-    with open('/sys/devices/virtual/dmi/id/product_uuid') as uuidin:
-        msg += '/uuid=' + uuidin.read().strip()
+    if not confuuid:
+        with open('/etc/confluent/confluent.deploycfg') as dcfg:
+            for line in dcfg.read().split('\n'):
+                if line.startswith('confluent_uuid:'):
+                    confluentuuid = line.split(': ')[1]
+                    msg += '/confluentuuid=' + confluentuuid
+                    break
+    try:
+        with open('/sys/devices/virtual/dmi/id/product_uuid') as uuidin:
+            msg += '/uuid=' + uuidin.read().strip()
+    except Exception:
+        pass
     for addrf in glob.glob('/sys/class/net/*/address'):
         with open(addrf) as addrin:
             hwaddr = addrin.read().strip()
@@ -122,6 +135,7 @@ def scan_confluents():
     srvlist = []
     if r:
         r = r[0]
+    nodename = None
     while r:
         for s in r:
             (rsp, peer) = s.recvfrom(9000)
@@ -129,6 +143,7 @@ def scan_confluents():
             current = None
             for line in rsp:
                 if line.startswith(b'NODENAME: '):
+                    nodename = line.replace(b'NODENAME: ', b'').strip().decode('utf8')
                     current = {}
                 elif line.startswith(b'DEFAULTNET: 1'):
                     current['isdefault'] = True
@@ -144,16 +159,32 @@ def scan_confluents():
         r = select.select((s4, s6), (), (), 2)
         if r:
             r = r[0]
+    if not os.path.exists('/etc/confluent/confluent.info'):
+        with open('/etc/confluent/confluent.info', 'w+') as cinfo:
+            if nodename:
+                cinfo.write('NODENAME: {0}\n'.format(nodename))
+            for srv in srvlist:
+                cinfo.write('MANAGER: {0}\n'.format(srv))
     return srvlist, srvs
 
 
-def get_net_apikey(nodename, mgr):
+def get_net_apikey(nodename, mgr, hmackey=None, confuuid=None):
     alpha = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789./'
     newpass = ''.join([alpha[x >> 2] for x in bytearray(os.urandom(32))])
     salt = '$5$' + ''.join([alpha[x >> 2] for x in bytearray(os.urandom(8))])
     newpass = newpass.encode('utf8')
     salt = salt.encode('utf8')
     crypted = c_crypt(newpass, salt)
+    if hmackey:
+        hmacvalue = hmac.new(hmackey.encode('utf8'), crypted, hashlib.sha256).digest()
+        hmacvalue = base64.b64encode(hmacvalue).decode('utf8')
+        client = HTTPSClient(host=mgr, phmac=hmacvalue, nodename=nodename, confuuid=confuuid)
+        try:
+            status, rsp = client.grab_url_with_status('/confluent-api/self/registerapikey', data=crypted, returnrsp=True)
+            if status == 200:
+                return newpass.decode('utf8')
+        except Exception:
+            pass
     for addrinfo in socket.getaddrinfo(mgr, 13001, 0, socket.SOCK_STREAM):
         try:
             clisock = socket.socket(addrinfo[0], addrinfo[1])
@@ -191,7 +222,7 @@ def get_net_apikey(nodename, mgr):
     return ''
 
 
-def get_apikey(nodename, hosts, errout=None):
+def get_apikey(nodename, hosts, errout=None, hmackey=None, confuuid=None):
     apikey = ""
     if os.path.exists('/etc/confluent/confluent.apikey'):
         apikey = open('/etc/confluent/confluent.apikey').read().strip()
@@ -200,16 +231,16 @@ def get_apikey(nodename, hosts, errout=None):
     while not apikey:
         for host in hosts:
             try:
-                apikey = get_net_apikey(nodename, host)
+                apikey = get_net_apikey(nodename, host, hmackey=hmackey, confuuid=confuuid)
             except OSError:
                 apikey = None
             if apikey:
                 break
         else:
-            srvlist, _ = scan_confluents()
+            srvlist, _ = scan_confluents(confuuid=confuuid)
             for host in srvlist:
                 try:
-                    apikey = get_net_apikey(nodename, host)
+                    apikey = get_net_apikey(nodename, host, hmackey=hmackey, confuuid=confuuid)
                 except OSError:
                     apikey = None
                 if apikey:
@@ -227,34 +258,43 @@ def get_apikey(nodename, hosts, errout=None):
     return apikey
 
 class HTTPSClient(client.HTTPConnection, object):
-    def __init__(self, usejson=False, port=443, host=None, errout=None, phmac=None, checkonly=False):
+    def __init__(self, usejson=False, port=443, host=None, errout=None, phmac=None, checkonly=False, hmackey=None, nodename=None, confuuid=None):
+        self.ignorehosts = set([])
         self.phmac = phmac
+        self.hmackey = hmackey
+        self.confuuid = confuuid
         self.errout = None
+        self.stdheaders = {}
+        if nodename:
+            self.stdheaders['CONFLUENT_NODENAME'] = nodename
         if errout:
             self.errout = open(errout, 'w')
             self.errout.flush()
-        self.stdheaders = {}
         mgtiface = None
         if usejson:
             self.stdheaders['ACCEPT'] = 'application/json'
         if host:
             self.hosts = [host]
-            with open('/etc/confluent/confluent.info') as cinfo:
-                info = cinfo.read().split('\n')
-                for line in info:
-                    if line.startswith('NODENAME:'):
-                        node = line.split(' ')[1]
-                        self.stdheaders['CONFLUENT_NODENAME'] = node
+            if not nodename:
+                with open('/etc/confluent/confluent.info') as cinfo:
+                    info = cinfo.read().split('\n')
+                    for line in info:
+                        if line.startswith('NODENAME:'):
+                            nodename = line.split(' ')[1]
+            self.stdheaders['CONFLUENT_NODENAME'] = nodename
         else:
             self.hosts = []
-            info = open('/etc/confluent/confluent.info').read().split('\n')
+            try:
+                info = open('/etc/confluent/confluent.info').read().split('\n')
+            except Exception:
+                info = []
             havedefault = '0'
             plainhost = ''
             for line in info:
                 host = ''
                 if line.startswith('NODENAME:'):
-                    node = line.split(' ')[1]
-                    self.stdheaders['CONFLUENT_NODENAME'] = node
+                    nodename = line.split(' ')[1]
+                    self.stdheaders['CONFLUENT_NODENAME'] = nodename
                 if line.startswith('MANAGER:') and not host:
                     host = line.split(' ')[1]
                     self.hosts.append(host)
@@ -289,15 +329,14 @@ class HTTPSClient(client.HTTPConnection, object):
             if plainhost and not self.hosts:
                 self.hosts.append(plainhost)
         if self.phmac:
-            with open(phmac, 'r') as hmacin:
-                self.stdheaders['CONFLUENT_CRYPTHMAC'] = hmacin.read()
+            self.stdheaders['CONFLUENT_CRYPTHMAC'] = self.phmac
         elif not checkonly:
-            self.stdheaders['CONFLUENT_APIKEY'] = get_apikey(node, self.hosts, errout=self.errout)
+            self.stdheaders['CONFLUENT_APIKEY'] = get_apikey(nodename, self.hosts, errout=self.errout, hmackey=hmackey, confuuid=self.confuuid)
         if mgtiface:
             self.stdheaders['CONFLUENT_MGTIFACE'] = mgtiface
         self.port = port
         self.host = None
-        self.node = node
+        self.node = nodename
         host = self.check_connections()
         client.HTTPConnection.__init__(self, host, port)
         self.connect()
@@ -315,6 +354,8 @@ class HTTPSClient(client.HTTPConnection, object):
         ctx.check_hostname = True
         for timeo in (0.1, 5):
             for host in hosts:
+                if host in self.ignorehosts:
+                    continue
                 try:
                     addrinf = socket.getaddrinfo(host, self.port)[0]
                     psock = socket.socket(addrinf[0])
@@ -335,7 +376,7 @@ class HTTPSClient(client.HTTPConnection, object):
                 continue
             break
         if not foundsrv:
-            srvlist, srvs = scan_confluents()
+            srvlist, srvs = scan_confluents(self.confuuid)
             hosts = []
             for srv in srvlist:
                 if srvs[srv].get('isdefault', False):
@@ -409,11 +450,101 @@ class HTTPSClient(client.HTTPConnection, object):
                 with open('/etc/confluent/confluent.apikey', 'w+') as akfile:
                     akfile.write('')
                 self.stdheaders['CONFLUENT_APIKEY'] = get_apikey(
-                    self.node, [self.host], errout=self.errout)
+                    self.node, [self.host], errout=self.errout, hmackey=self.hmackey, confuuid=self.confuuid)
+            if rsp.status == 503:  # confluent is down, but the server running confluent is otherwise up
+                authed = False
+                self.ignorehosts.add(self.host)
+                host = self.check_connections()
+                client.HTTPConnection.__init__(self, host, self.port)
+
         raise Exception(rsp.read())
 
+def get_current_vmnic_vswitch():
+    uplinkmatch = re.compile(r'^\s*Uplinks:\s*(.*)')
+    switchmatch = re.compile(r'^\s*Name:\s*(.*)')
+
+    vswinfo = subprocess.check_output(['localcli', 'network', 'vswitch', 'standard', 'list']).decode()
+
+    vmnic = None
+    vswitch_name = None
+
+    for info in vswinfo.split('\n'):
+        name_match = switchmatch.match(info)
+        if name_match:
+            vswitch_name = name_match.group(1).strip()
+
+        upinfo = uplinkmatch.match(info)
+        if upinfo:
+            vmnic = upinfo.group(1).strip()
+            if vmnic and 'vusb0' not in vmnic:
+                return vswitch_name, vmnic
+
+    return vswitch_name, vmnic
+
+def get_available_nics():
+    nicinfo = subprocess.check_output(['localcli', 'network', 'nic', 'list']).decode('utf8').split('\n')
+    available_nics = {}
+
+    # Skip headers and separators
+    parsing_started = False
+    for line in nicinfo:
+        if re.match(r'^-+', line):
+            parsing_started = True
+            continue
+        if not parsing_started or not line.strip():
+            continue
+
+        parts = re.split(r'\s{2,}', line.strip())
+        if len(parts) >= 5:
+            nic_name = parts[0]
+            nic_status = parts[4]  # "Link Status" is the 5th field
+            available_nics[nic_name] = nic_status
+
+    return available_nics
+
+def is_esxi():
+    return os.path.isdir("/etc/vmware")
+
+def fix_vswitch():
+    if is_esxi():
+        start_time = time.time()
+        while True:
+
+            current_vswitch, current_vmnic = get_current_vmnic_vswitch()
+            if current_vswitch is None:
+                raise RuntimeError("Panic: current vswitch is None")
+            if current_vmnic is None:
+                raise RuntimeError("Panic: current vmnic is None")
+
+            available_nics = get_available_nics()
+
+            if current_vmnic and available_nics.get(current_vmnic) == 'Up':
+                break
+
+            new_vmnic = next((nic for nic, status in available_nics.items() if status == 'Up'), None)
+
+            if new_vmnic and new_vmnic != current_vmnic:
+                subprocess.check_call(['localcli', 'network', 'vswitch', 'standard', 'uplink', 'remove',
+                                       '--uplink-name', current_vmnic, '--vswitch-name', current_vswitch])
+                subprocess.check_call(['localcli', 'network', 'vswitch', 'standard', 'uplink', 'add',
+                                       '--uplink-name', new_vmnic, '--vswitch-name', current_vswitch])
+            elif not new_vmnic:
+                if time.time() - start_time > 300:
+                    break
+            time.sleep(5)
+
+        time.sleep(5)
+
 if __name__ == '__main__':
     data = None
+
+    if '-f' in sys.argv:
+        try:
+            fix_vswitch()
+        except Exception as e:
+            print("fix_vswitch() error: {}".format(e))
+        sys.argv.remove('-f')
+        sys.exit(0)
     usejson = False
     if '-j' in sys.argv:
         usejson = True
@@ -448,8 +579,24 @@ if __name__ == '__main__':
         phmac = sys.argv.index('-p')
         sys.argv.pop(phmac)
         phmac = sys.argv.pop(phmac)
+        with open(phmac, 'r') as hmacin:
+            phmac = hmacin.read()
     except ValueError:
         phmac = None
+    try:
+        identfile = sys.argv.index('-i')
+        sys.argv.pop(identfile)
+        identfile = sys.argv.pop(identfile)
+        with open(identfile) as idin:
+            data = idin.read()
+            identinfo = json.loads(data)
+            nodename = identinfo.get('nodename', None)
+            hmackey = identinfo.get('apitoken', None)
+            confuuid = identinfo.get('confluent_uuid', None)
+    except ValueError:
+        hmackey = None
+        nodename = None
+        confuuid = None
     try:
         checkonly = False
         idxit = sys.argv.index('-c')
@@ -461,20 +608,21 @@ if __name__ == '__main__':
         data = open(sys.argv[-1]).read()
     if outbin:
         with open(outbin, 'ab+') as outf:
-            reader = HTTPSClient(usejson=usejson, errout=errout).grab_url(
+            reader = HTTPSClient(usejson=usejson, errout=errout, hmackey=hmackey, nodename=nodename, confuuid=confuuid).grab_url(
                 sys.argv[1], data, returnrsp=True)
             chunk = reader.read(16384)
             while chunk:
                 outf.write(chunk)
                 chunk = reader.read(16384)
         sys.exit(0)
-    client = HTTPSClient(usejson, errout=errout, phmac=phmac, checkonly=checkonly)
+
+    mclient = HTTPSClient(usejson, errout=errout, phmac=phmac, checkonly=checkonly, hmackey=hmackey, nodename=nodename, confuuid=confuuid)
     if waitfor:
         status = 201
         while status != waitfor:
-            status, rsp = client.grab_url_with_status(sys.argv[1], data)
+            status, rsp = mclient.grab_url_with_status(sys.argv[1], data)
             sys.stdout.write(rsp.decode())
     elif checkonly:
-        sys.stdout.write(client.check_connections())
+        sys.stdout.write(mclient.check_connections())
     else:
-        sys.stdout.write(client.grab_url(sys.argv[1], data).decode())
+        sys.stdout.write(mclient.grab_url(sys.argv[1], data).decode())

confluent_osdeploy/common/profile/scripts/autoconsole

@@ -0,0 +1,119 @@
+#!/usr/bin/python3
+
+# This script evaluates whether firmware redirection is likely. It uses three cues:
+# - Does the system offer up SPCR? This would indicate that the firmware is doing serial output.
+#   Otherwise, there's no indication that the firmware cares about serial console.
+# - Is the system EFI?  BIOS implementations may not intercept text draw calls after POST exit,
+#   thus even when BIOS tells us serial port is in use, it may not be doing anything when
+#   grub would be running
+# - Is the serial port connected? In the event that firmware indicates serial port, but
+#   serial port is not reporting DCD, then it doesn't look like a comfortable enough scenario
+
+import fcntl
+import os
+import os.path
+import struct
+import subprocess
+import termios
+
+
+addrtoname = {
+    0x3f8: '/dev/ttyS0',
+    0x2f8: '/dev/ttyS1',
+    0x3e8: '/dev/ttyS2',
+    0x2e8: '/dev/ttyS3',
+}
+speedmap = {
+    0: None,
+    3: 9600,
+    4: 19200,
+    6: 57600,
+    7: 115200,
+}
+
+termiobaud = {
+    9600: termios.B9600,
+    19200: termios.B19200,
+    57600: termios.B57600,
+    115200: termios.B115200,
+}
+
+
+def deserialize_grub_rh():
+    if 'console=ttyS' in open('/proc/cmdline').read():
+        return None # User manually indicated serial config
+        # they own the grub behavior too for now
+    grublines = []
+    with open('/etc/default/grub') as grubin:
+        grublines = grubin.read().split('\n')
+    with open('/etc/default/grub', 'w') as grubout:
+        for grubline in grublines:
+            if grubline.startswith('GRUB_TERMINAL'):
+                grubline = grubline.replace('serial ', '')
+            grubout.write(grubline + '\n')
+    subprocess.check_call(['grub2-mkconfig', '-o', '/boot/grub2/grub.cfg'])
+
+def fixup_ubuntu_grub_serial():
+    # Ubuntu aggressively tries to graphics up
+    # grub. We will counter that for serial
+    # They also aggressively hide UI and
+    # block ability to interject. We will
+    # compromise and lean on nodeboot <node> setup
+    # as a means to give someone reasonable shot at
+    # the short timeout
+    with open('/etc/default/grub') as grubin:
+        grublines = grubin.read().split('\n')
+    with open('/etc/default/grub', 'w') as grubout:
+        for grubline in grublines:
+            if grubline.startswith('GRUB_TIMEOUT_STYLE=hidden'):
+                grubline = 'GRUB_TIMEOUT_STYLE=menu'
+            elif grubline.startswith('GRUB_TIMEOUT=0'):
+                grubline = 'GRUB_TIMEOUT=2'
+            elif grubline.startswith('#GRUB_TERMINAL=console'):
+                grubline = grubline.replace('#', '')
+            grubout.write(grubline + '\n')
+    subprocess.check_call(['update-grub'])
+
+def get_serial_config():
+    if not os.path.exists('/sys/firmware/efi'):
+        return None
+    if not os.path.exists('/sys/firmware/acpi/tables/SPCR'):
+        return None
+    spcr = open("/sys/firmware/acpi/tables/SPCR", "rb")
+    spcr = bytearray(spcr.read())
+    if spcr[8] != 2 or spcr[36] != 0 or spcr[40] != 1:
+        return None
+    address = struct.unpack('<Q', spcr[44:52])[0]
+    tty = None
+    try:
+        tty = addrtoname[address]
+    except KeyError:
+        return None
+    retval = { 'tty': tty }
+    try:
+        retval['speed'] = speedmap[spcr[58]]
+    except KeyError:
+        return None
+    if retval['speed']:
+        ttyf = os.open(tty, os.O_RDWR | os.O_NOCTTY)
+        currattr = termios.tcgetattr(ttyf)
+        currattr[4:6] = [0, termiobaud[retval['speed']]]
+        termios.tcsetattr(ttyf, termios.TCSANOW, currattr)
+    retval['connected'] = bool(struct.unpack('<I', fcntl.ioctl(
+        ttyf, termios.TIOCMGET, '\x00\x00\x00\x00'))[0] & termios.TIOCM_CAR)
+    os.close(ttyf)
+    return retval
+
+def main():
+    autoconscfg = get_serial_config()
+    if not autoconscfg or not autoconscfg['connected']:
+        return
+    if os.path.exists('/etc/redhat-release'):  # redhat family
+        deserialize_grub_rh()
+    elif os.path.exists('/etc/os-release'):
+        with open('/etc/os-release') as osr:
+            if 'Ubuntu' in osr.read():
+                fixup_ubuntu_grub_serial()
+
+if __name__ == '__main__':
+    main()

confluent_osdeploy/common/profile/scripts/confignet

@@ -164,6 +164,9 @@ class NetplanManager(object):
                 if curraddr not in currips:
                     needcfgwrite = True
                     currips.append(curraddr)
+            if stgs.get('mtu', None):
+                devdict = self.getcfgarrpath([devname])
+                devdict['mtu'] = int(stgs['mtu'])
             gws = []
             gws.append(stgs.get('ipv4_gateway', None))
             gws.append(stgs.get('ipv6_gateway', None))
@@ -367,7 +370,7 @@ class NetworkManager(object):
                 args.append(bondcfg[parm])
             subprocess.check_call(['nmcli', 'c', 'm', team] + args)
 
-    def apply_configuration(self, cfg):
+    def apply_configuration(self, cfg, lastchance=False):
         cmdargs = {}
         cmdargs['connection.autoconnect'] = 'yes'
         stgs = cfg['settings']
@@ -381,6 +384,8 @@ class NetworkManager(object):
             cmdargs['ipv4.gateway'] = stgs['ipv4_gateway']
         if stgs.get('ipv6_gateway', None):
             cmdargs['ipv6.gateway'] = stgs['ipv6_gateway']
+        if stgs.get('mtu', None):
+            cmdargs['802-3-ethernet.mtu'] = stgs['mtu']
         dnsips = self.deploycfg.get('nameservers', [])
         if not dnsips:
             dnsips = []
@@ -406,10 +411,10 @@ class NetworkManager(object):
             cargs = []
             for arg in cmdargs:
                 cargs.append(arg)
-                cargs.append(cmdargs[arg])
+                cargs.append('{}'.format(cmdargs[arg]))
             if stgs['team_mode'] in self.bondtypes:
                 stgs['team_mode'] = self.bondtypes[stgs['team_mode']]
-            subprocess.check_call(['nmcli', 'c', 'add', 'type', 'bond', 'con-name', cname, 'connection.interface-name', cname, 'bond.options', 'mode={}'.format(stgs['team_mode'])] + cargs)
+            subprocess.check_call(['nmcli', 'c', 'add', 'type', 'bond', 'con-name', cname, 'connection.interface-name', cname, 'bond.options', 'miimon=100,mode={}'.format(stgs['team_mode'])] + cargs)
             for iface in cfg['interfaces']:
                 self.add_team_member(cname, iface)
             subprocess.check_call(['nmcli', 'c', 'u', cname])
@@ -418,8 +423,9 @@ class NetworkManager(object):
             iname = list(cfg['interfaces'])[0]
             ctype = self.devtypes.get(iname, None)
             if not ctype:
-                sys.stderr.write("Warning, no device found for interface_name ({0}), skipping setup\n".format(iname))
-                return
+                if lastchance:
+                    sys.stderr.write("Warning, no device found for interface_name ({0}), skipping setup\n".format(iname))
+                return 1
             if stgs.get('vlan_id', None):
                 vlan = stgs['vlan_id']
                 if ctype == 'infiniband':
@@ -440,7 +446,7 @@ class NetworkManager(object):
             cargs = []
             for arg in cmdargs:
                 cargs.append(arg)
-                cargs.append(cmdargs[arg])
+                cargs.append('{}'.format(cmdargs[arg]))
             if u:
                 subprocess.check_call(['nmcli', 'c', 'm', u, 'connection.interface-name', iname] + cargs)
                 subprocess.check_call(['nmcli', 'c', 'u', u])
@@ -454,6 +460,9 @@ class NetworkManager(object):
 
 
 if __name__ == '__main__':
+    checktarg = None
+    if '-c' in sys.argv:
+        checktarg = sys.argv[sys.argv.index('-c') + 1]
     havefirewall = subprocess.call(['systemctl', 'status', 'firewalld'])
     havefirewall = havefirewall == 0
     if havefirewall:
@@ -539,13 +548,49 @@ if __name__ == '__main__':
     rm_tmp_llas(tmpllas)
     if os.path.exists('/usr/sbin/netplan'):
         nm = NetplanManager(dc)
-    if os.path.exists('/usr/bin/nmcli'):
+    elif os.path.exists('/usr/bin/nmcli'):
         nm = NetworkManager(devtypes, dc)
     elif os.path.exists('/usr/sbin/wicked'):
         nm = WickedManager()
+    retrynics = []
     for netn in netname_to_interfaces:
-        nm.apply_configuration(netname_to_interfaces[netn])
+        redo = nm.apply_configuration(netname_to_interfaces[netn])
+        if redo == 1:
+            retrynics.append(netn)
+    if retrynics:
+        idxmap, devtypes = map_idx_to_name()
+        if os.path.exists('/usr/sbin/netplan'):
+            nm = NetplanManager(dc)
+        if os.path.exists('/usr/bin/nmcli'):
+            nm = NetworkManager(devtypes, dc)
+        elif os.path.exists('/usr/sbin/wicked'):
+            nm = WickedManager()
+        for netn in retrynics:
+            nm.apply_configuration(netname_to_interfaces[netn], lastchance=True)
     if havefirewall:
         subprocess.check_call(['systemctl', 'start', 'firewalld'])
     await_tentative()
+    maxwait = 10
+    while maxwait:
+        try:
+            tclient = apiclient.HTTPSClient(checkonly=True)
+            tclient.check_connections()
+            break
+        except Exception:
+            maxwait -= 1
+            time.sleep(1)
+    maxwait = 10
+    if checktarg:
+        while maxwait:
+            try:
+                addrinf = socket.getaddrinfo(checktarg, 443)[0]
+                psock = socket.socket(addrinf[0], socket.SOCK_STREAM)
+                psock.settimeout(10)
+                psock.connect(addrinf[4])
+                psock.close()
+                break
+            except Exception:
+                maxwait -= 1
+                time.sleep(1)
+
 

confluent_osdeploy/common/profile/scripts/sample/consoleredirect

@@ -1,49 +0,0 @@
-is_suse=false
-is_rhel=false
-
-if test -f /boot/efi/EFI/redhat/grub.cfg; then
-    grubcfg="/boot/efi/EFI/redhat/grub.cfg"
-    grub2-mkconfig -o $grubcfg
-    is_rhel=true
-elif test -f /boot/efi/EFI/sle_hpc/grub.cfg; then
-    grubcfg="/boot/efi/EFI/sle_hpc/grub.cfg"
-    grub2-mkconfig -o $grubcfg
-    is_suse=true
-else
-    echo "Expected File missing: Check if os sle_hpc or redhat"
-    exit
-fi
-
-# working on SUSE
-if $is_suse; then
-    start=false
-    num_line=0
-    lines_to_edit=()
-    while read line; do
-        ((num_line++))
-        if [[ $line == *"grub_platform"* ]]; then
-            start=true
-        fi
-        if $start; then
-            if [[ $line != "#"* ]];then
-                lines_to_edit+=($num_line)
-            fi
-        fi
-        if [[ ${#line} -eq 2 && $line == *"fi" ]]; then
-            if $start; then
-                start=false
-            fi
-        fi
-    done < grub_cnf.cfg
-
-    for line_num in "${lines_to_edit[@]}"; do
-        line_num+="s"
-        sed -i "${line_num},^,#," $grubcfg
-    done
-    sed -i 's,^terminal,#terminal,' $grubcfg
-fi
-
-# Working on Redhat
-if $is_rhel; then
-    sed -i 's,^serial,#serial, ; s,^terminal,#terminal,' $grubcfg
-fi

confluent_osdeploy/common/profile/scripts/setupssh

@@ -7,7 +7,7 @@ for pubkey in /etc/ssh/ssh_host*key.pub; do
         continue
     fi
     certfile=${pubkey/.pub/-cert.pub}
-    rm $certfile
+    echo -n > $certfile
     confluentpython $confapiclient /confluent-api/self/sshcert $pubkey -o $certfile
 done
 if [ -d /etc/ssh/sshd_config.d/ -a ! -e /etc/ssh/sshd_config.d/90-confluent.conf ]; then
@@ -17,6 +17,13 @@ if [ -d /etc/ssh/sshd_config.d/ -a ! -e /etc/ssh/sshd_config.d/90-confluent.conf
 	echo HostbasedAuthentication yes >> /etc/ssh/sshd_config.d/90-confluent.conf
 	echo HostbasedUsesNameFromPacketOnly yes >> /etc/ssh/sshd_config.d/90-confluent.conf
 	echo IgnoreRhosts no >> /etc/ssh/sshd_config.d/90-confluent.conf
+elif [ ! -d /etc/ssh/sshd_config.d/ ] && ! grep HostCertificate /etc/ssh/sshd_config > /dev/null; then
+	for cert in /etc/ssh/ssh*-cert.pub; do
+		echo HostCertificate $cert >> /etc/ssh/sshd_config
+	done
+	echo HostbasedAuthentication yes >> /etc/ssh/sshd_config
+	echo HostbasedUsesNameFromPacketOnly yes >> /etc/ssh/sshd_config
+	echo IgnoreRhosts no >> /etc/ssh/sshd_config
 fi
 
 TMPDIR=$(mktemp -d)
@@ -25,13 +32,20 @@ confluentpython $confapiclient /confluent-public/site/initramfs.tgz -o initramfs
 tar xf initramfs.tgz
 for ca in ssh/*.ca; do
 	LINE=$(cat $ca)
-	cp -af /etc/ssh/ssh_known_hosts /etc/ssh/ssh_known_hosts.new
-	grep -v "$LINE" /etc/ssh/ssh_known_hosts > /etc/ssh/ssh_known_hosts.new
+	if [ -z "$LINE" ]; then continue; fi
+	if [ -f /etc/ssh/ssh_known_hosts ]; then
+		cp -af /etc/ssh/ssh_known_hosts /etc/ssh/ssh_known_hosts.new
+		grep -v "$LINE" /etc/ssh/ssh_known_hosts > /etc/ssh/ssh_known_hosts.new
+	fi
 	echo '@cert-authority *' $LINE >> /etc/ssh/ssh_known_hosts.new
 	mv /etc/ssh/ssh_known_hosts.new /etc/ssh/ssh_known_hosts
 done
+mkdir -p /root/.ssh/
+chmod 700 /root/.ssh/
+touch /root/.ssh/authorized_keys
 for pubkey in ssh/*.*pubkey; do
 	LINE=$(cat $pubkey)
+        if [ -z "$LINE" ]; then continue; fi
 	cp -af /root/.ssh/authorized_keys /root/.ssh/authorized_keys.new
 	grep -v "$LINE" /root/.ssh/authorized_keys > /root/.ssh/authorized_keys.new
 	echo "$LINE" >> /root/.ssh/authorized_keys.new
@@ -41,3 +55,4 @@ confluentpython $confapiclient /confluent-api/self/nodelist | sed -e 's/^- //' >
 cat /etc/ssh/shosts.equiv > /root/.shosts
 cd -
 rm -rf $TMPDIR
+systemctl try-restart sshd

confluent_osdeploy/confluent_osdeploy-aarch64.spec.tmpl

@@ -26,11 +26,19 @@ mkdir -p opt/confluent/bin
 mkdir -p stateless-bin
 cp -a el8bin/* .
 ln -s el8 el9
-for os in rhvh4 el7 genesis el8 suse15 ubuntu20.04 ubuntu22.04 ubuntu24.04 coreos el9; do
+ln -s el8 el10
+mv el10/initramfs/usr el10/initramfs/var
+cp -a debian debian13
+mkdir -p debian13/initramfs/usr
+mv debian13/initramfs/lib debian13/initramfs/usr/
+mv el10/initramfs/usr el10/initramfs/var
+for os in rhvh4 el7 genesis el8 suse15 debian debian13 ubuntu20.04 ubuntu22.04 ubuntu24.04 coreos el9 el10; do
     mkdir ${os}out
     cd ${os}out
     if [ -d ../${os}bin ]; then 
         cp -a ../${os}bin/opt .
+    elif [ $os = el10 ]; then
+        cp -a ../el9bin/opt .
     else
         cp -a ../opt .
     fi
@@ -40,11 +48,13 @@ for os in rhvh4 el7 genesis el8 suse15 ubuntu20.04 ubuntu22.04 ubuntu24.04 coreo
     mv ../addons.cpio .
     cd ..
 done
-for os in el7 el8 suse15 el9 ubuntu20.04; do
+for os in el7 el8 suse15 el9 el10 ubuntu20.04; do
     mkdir ${os}disklessout
     cd ${os}disklessout
     if [ -d ../${os}bin ]; then
         cp -a ../${os}bin/opt .
+    elif [ $os = el10 ]; then
+        cp -a ../el9bin/opt .
     else
         cp -a ../opt .
     fi
@@ -76,7 +86,7 @@ cp -a esxi7 esxi8
 %install
 mkdir -p %{buildroot}/opt/confluent/share/licenses/confluent_osdeploy/
 #cp LICENSE %{buildroot}/opt/confluent/share/licenses/confluent_osdeploy/
-for os in rhvh4 el7 el8 el9 genesis suse15 ubuntu20.04 ubuntu22.04 ubuntu24.04 esxi6 esxi7 esxi8 coreos; do
+for os in rhvh4 el7 el8 el9 el10 debian debian13 genesis suse15 ubuntu20.04 ubuntu22.04 ubuntu24.04 esxi6 esxi7 esxi8 coreos; do
     mkdir -p %{buildroot}/opt/confluent/lib/osdeploy/$os/initramfs/aarch64/
     cp ${os}out/addons.* %{buildroot}/opt/confluent/lib/osdeploy/$os/initramfs/aarch64/
     if [ -d ${os}disklessout ]; then

confluent_osdeploy/confluent_osdeploy.spec.tmpl

@@ -29,8 +29,11 @@ This contains support utilities for enabling deployment of x86_64 architecture s
 #cd ..
 ln -s el8 el9
 cp -a el8 el10
+cp -a debian debian13
+mkdir -p debian13/initramfs/usr
+mv debian13/initramfs/lib debian13/initramfs/usr/
 mv el10/initramfs/usr el10/initramfs/var
-for os in rhvh4 el7 genesis el8 suse15 ubuntu18.04 ubuntu20.04 ubuntu22.04 ubuntu24.04 coreos el9 el10; do
+for os in rhvh4 el7 genesis el8 suse15 debian debian13 ubuntu18.04 ubuntu20.04 ubuntu22.04 ubuntu24.04 coreos el9 el10; do
     mkdir ${os}out
     cd ${os}out
     if [ -d ../${os}bin ]; then 
@@ -46,11 +49,13 @@ for os in rhvh4 el7 genesis el8 suse15 ubuntu18.04 ubuntu20.04 ubuntu22.04 ubunt
     mv ../addons.cpio .
     cd ..
 done
-for os in el7 el8 suse15 el9 ubuntu20.04 ubuntu22.04 ubuntu24.04; do
+for os in el7 el8 suse15 el9 el10 ubuntu20.04 ubuntu22.04 ubuntu24.04; do
     mkdir ${os}disklessout
     cd ${os}disklessout
     if [ -d ../${os}bin ]; then
         cp -a ../${os}bin/opt .
+    elif [ $os = el10 ]; then
+        cp -a ../el9bin/opt .
     else
         cp -a ../el8bin/opt .
     fi
@@ -77,12 +82,14 @@ cd ..
 cp -a esxi7out esxi6out
 cp -a esxi7 esxi6
 cp -a esxi7out esxi8out
+cp -a esxi7out esxi9out
 cp -a esxi7 esxi8
+cp -a esxi7 esxi9
 
 %install
 mkdir -p %{buildroot}/opt/confluent/share/licenses/confluent_osdeploy/
 cp LICENSE %{buildroot}/opt/confluent/share/licenses/confluent_osdeploy/
-for os in rhvh4 el7 el8 el9 el10 genesis suse15 ubuntu20.04 ubuntu18.04 ubuntu22.04 ubuntu24.04 esxi6 esxi7 esxi8 coreos; do
+for os in rhvh4 el7 el8 el9 el10 genesis suse15 ubuntu20.04 debian debian13 ubuntu18.04 ubuntu22.04 ubuntu24.04 esxi6 esxi7 esxi8 esxi9 coreos; do
     mkdir -p %{buildroot}/opt/confluent/lib/osdeploy/$os/initramfs
     mkdir -p %{buildroot}/opt/confluent/lib/osdeploy/$os/profiles
     cp ${os}out/addons.* %{buildroot}/opt/confluent/lib/osdeploy/$os/initramfs

confluent_osdeploy/debian/initramfs/lib/debian-installer-startup.d/S25confluentinit

@@ -14,65 +14,127 @@ setdebopt() {
     echo d-i $1 $3 $2 >> /preseed.cfg
 }
 
-dhuuid=$(reverse_uuid $(cat /sys/devices/virtual/dmi/id/product_uuid))
-dhcpid=$(mktemp)
 mkdir -p /etc/confluent
-cp /tls/* /etc/ssl/certs/
-for nic in $(ip link | grep mtu|grep -v LOOPBACK|cut -d: -f 2|sed -e 's/ //'); do
-     ip link set $nic up
+for i in /sys/class/net/*; do
+    ip link set $(basename $i) up
 done
-for nic in $(ip link | grep mtu|grep -v LOOPBACK|grep LOWER_UP|cut -d: -f 2|sed -e 's/ //'); do
-    if udhcpc -i $nic -p $dhcpid -t 2 -T 2 -n -x 93:0007 -x 97:00$dhuuid -q; then
-         /opt/confluent/bin/copernicus > /etc/confluent/confluent.info
-         if grep ^MANAGER:.*\\. /etc/confluent/confluent.info ; then
-             break
-         fi
+TRIES=5
+while [ ! -e /dev/disk ] && [ $TRIES -gt 0 ]; do
+    sleep 2
+    TRIES=$((TRIES - 1))
+done
+for i in /sys/class/net/*; do
+    ip link set $(basename $i) down
+    udevadm info $i | grep ID_NET_DRIVER=cdc_ether > /dev/null &&  continue
+    ip link set $(basename $i) up
+done
+cp -a /tls/* /etc/ssl/certs/
+mkdir -p /etc/confluent
+if [ -e /dev/disk/by-label/CNFLNT_IDNT ]; then
+    tmnt=$(mktemp -d)
+    tcfg=$(mktemp)
+    mount /dev/disk/by-label/CNFLNT_IDNT $tmnt
+    cd $tmnt
+    deploysrvs=$(sed -n '/^deploy_servers:/,/^[^-]/p' cnflnt.yml |grep ^-|sed -e 's/^- //'|grep -v :)
+    nodename=$(grep ^nodename: cnflnt.yml|cut -f 2 -d ' ')
+    echo NODENAME: $nodename > /etc/confluent/confluent.info
+    sed -n '/^net_cfgs:/,/^[^- ]/{/^[^- ]/!p}' cnflnt.yml |sed -n '/^-/,/^-/{/^-/!p}'| sed -e 's/^[- ]*//'> $tcfg
+    autoconfigmethod=$(grep ^ipv4_method: $tcfg)
+    autoconfigmethod=${autoconfigmethod#ipv4_method: }
+    if [ "$autoconfigmethod" = "static" ]; then
+        setdebopt netcfg/disable_dhcp true boolean
+        v4addr=$(grep ^ipv4_address: $tcfg|cut -d: -f 2|sed -e 's/ //')
+        v4gw=$(grep ^ipv4_gateway: $tcfg|cut -d: -f 2| sed -e 's/ //')
+        if [ "$v4gw" = "null" ]; then
+            v4gw=""
+        fi
+        v4nm=$(grep ^ipv4_netmask: $tcfg|cut -d: -f 2|sed -e 's/ //')
+        setdebopt netcfg/get_netmask $v4nm string
+        setdebopt netcfg/get_ipaddress ${v4addr%/*} string
+        setdebopt netcfg/confirm_static true boolean
+        if [ ! -z "$v4gw" ]; then
+            setdebopt netcfg/get_gateway $v4gw string
+        fi
+        NIC=""
+        while [ -z "$NIC" ]; do
+            for NICGUESS in $(ip link|grep LOWER_UP|grep -v LOOPBACK|cut -d ' ' -f 2 | sed -e 's/:$//'); do
+                ip addr add dev $NICGUESS $v4addr
+                if [ ! -z "$v4gw" ]; then
+                    ip route add default via $v4gw
+                fi
+                for dsrv in $deploysrvs; do
+                    if wget https://$dsrv/confluent-public/ --tries=1 --timeout=1 -O /dev/null > /dev/null 2>&1; then
+                        deploysrvs=$dsrv
+                        NIC=$NICGUESS
+                        setdebopt netcfg/choose_interface $NIC select
+                        break
+                    fi
+                done
+                if [ -z "$NIC" ]; then
+                    ip -4 a flush dev $NICGUESS
+                else
+                    break
+                fi
+            done
+        done
+        #TODO: nameservers
+    elif [ "$v4cfgmeth" = "dhcp" ]; then
+        setdebopt netcfg/disable_dhcp false boolean
+        setdebopt netcfg/confirm_static false boolean
+        for NICGUESS in $(ip link|grep LOWER_UP|grep -v LOOPBACK|cut -d ' ' -f 2 | sed -e 's/:$//'); do
+            udhcpc $NICGUESS
+        done
+        for dsrv in $deploysrvs; do
+            if wget https://$dsrv/confluent-public/ --tries=1 --timeout=1 -O /dev/null > /dev/null 2>&1; then
+                deploysrvs=$dsrv
+            fi
+        done
     fi
-    ip -4 flush dev $nic
-done    
-mgr=$(grep ^MANAGER:.*\\. /etc/confluent/confluent.info|head -n 1|cut -d: -f 2|sed -e 's/ //')
-nodename=$(grep ^NODENAME: /etc/confluent/confluent.info|head -n 1|cut -d: -f 2|sed -e 's/ //')
-/opt/confluent/bin/clortho $nodename $mgr > /etc/confluent/confluent.apikey
+    mgr=$deploysrvs
+    ln -s /opt/confluent/bin/clortho /opt/confluent/bin/genpasshmac
+    hmackeyfile=/tmp/cnflnthmackeytmp
+    passfile=/tmp/cnflnttmppassfile
+    passcrypt=/tmp/cnflntcryptfile
+    hmacfile=/tmp/cnflnthmacfile
+    echo -n $(grep ^apitoken: cnflnt.yml|cut -d ' ' -f 2) > $hmackeyfile
+    /opt/confluent/bin/genpasshmac $passfile $passcrypt $hmacfile $hmackeyfile
+    wget --header="CONFLUENT_NODENAME: $nodename" --header="CONFLUENT_CRYPTHMAC: $(cat $hmacfile)" --post-file=$passcrypt https://$mgr/confluent-api/self/registerapikey -O - --quiet
+    cp $passfile /etc/confluent/confluent.apikey
+    nic=$NIC
+else
+    dhuuid=$(reverse_uuid $(cat /sys/devices/virtual/dmi/id/product_uuid))
+    dhcpid=$(mktemp)
+    mkdir -p /etc/confluent
+    cp /tls/* /etc/ssl/certs/
+    cat /tls/*.pem >> /etc/confluent/ca.pem
+    for nic in $(ip link | grep mtu|grep -v LOOPBACK|cut -d: -f 2|sed -e 's/ //'); do
+        ip link set $nic up
+    done
+    for nic in $(ip link | grep mtu|grep -v LOOPBACK|grep LOWER_UP|cut -d: -f 2|sed -e 's/ //'); do
+        if udhcpc -i $nic -p $dhcpid -t 2 -T 2 -n -x 93:0007 -x 97:00$dhuuid -q; then
+            /opt/confluent/bin/copernicus > /etc/confluent/confluent.info
+            if grep ^MANAGER:.*\\. /etc/confluent/confluent.info ; then
+                break
+            fi
+        fi
+        ip -4 flush dev $nic
+    done
+    mgr=$(grep ^MANAGER:.*\\. /etc/confluent/confluent.info|head -n 1|cut -d: -f 2|sed -e 's/ //')
+    nodename=$(grep ^NODENAME: /etc/confluent/confluent.info|head -n 1|cut -d: -f 2|sed -e 's/ //')
+    /opt/confluent/bin/clortho $nodename $mgr > /etc/confluent/confluent.apikey
+fi
 apikey=$(cat /etc/confluent/confluent.apikey)
 cd /etc/confluent
 wget --header="CONFLUENT_NODENAME: $nodename" --header="CONFLUENT_APIKEY: $apikey" https://$mgr/confluent-api/self/deploycfg
 cd -
 predir=$(mktemp -d)
 cd $predir
+cp /etc/confluent/deploycfg /etc/confluent/confluent.deploycfg
 profile=$(grep ^profile: /etc/confluent/deploycfg|cut -d ' ' -f 2)
-wget https://$mgr/confluent-public/os/$profile/scripts/pre.sh
-chmod u+x pre.sh
-wget https://$mgr/confluent-public/os/$profile/preseed.cfg
-mv preseed.cfg /
-setdebopt auto-install/enable true boolean
-setdebopt partman/early_command  $predir/pre.sh string
-cd -
-ip -4 a flush dev $nic
-setdebopt netcfg/choose_interface $nic select
-setdebopt netcfg/get_hostname $nodename string
-v4cfgmeth=$(grep ipv4_method: /etc/confluent/deploycfg |cut -d: -f 2|sed -e 's/ //')
-if [ "$v4cfgmeth" = "static" ]; then
-    setdebopt netcfg/disable_dhcp true boolean
-    v4addr=$(grep ^ipv4_address: /etc/confluent/deploycfg|cut -d: -f 2|sed -e 's/ //')
-    v4gw=$(grep ^ipv4_gateway: /etc/confluent/deploycfg|cut -d: -f 2| sed -e 's/ //')
-    if [ "$v4gw" = "null" ]; then
-        v4gw=""
-    fi
-    v4nm=$(grep ^ipv4_netmask: /etc/confluent/deploycfg|cut -d: -f 2|sed -e 's/ //')
-    setdebopt netcfg/get_netmask $v4nm string
-    setdebopt netcfg/get_ipaddress $v4addr string
-    setdebopt netcfg/confirm_static true boolean
-    if [ ! -z "$v4gw" ]; then
-        setdebopt netcfg/get_gateway $v4gw string
-    fi
-    namesrvs=$(sed -n '/^nameservers:/,/^[^-]/p' /etc/confluent/deploycfg|grep ^- | cut -d ' ' -f 2|sed -e 's/ //')
-    for namesrv in "$namesrvs"; do
-        setdebopt netcfg/get_nameservers $namesrv string
-    done
-elif [ "$v4cfgmeth" = "dhcp" ]; then
-    setdebopt netcfg/disable_dhcp false boolean
-    setdebopt netcfg/confirm_static false boolean
-fi
+namesrvs=$(sed -n '/^nameservers:/,/^[^-]/p' /etc/confluent/deploycfg|grep ^- | cut -d ' ' -f 2|sed -e 's/ //')
+for namesrv in "$namesrvs"; do
+    setdebopt netcfg/get_nameservers $namesrv string
+done
 rootpass=$(grep ^rootpassword: /etc/confluent/deploycfg|cut -d ' ' -f 2|sed -e 's/ //')
 if [ "$rootpass" = null ] || [ -z "$rootpass" ]; then
     setdebopt passwd/root-login false boolean
@@ -84,9 +146,8 @@ setdebopt time/zone $(grep ^timezone: /etc/confluent/deploycfg|cut -d ' ' -f 2|s
 ntpsrvs=$(sed -n '/^ntpservers:/,/^[^-]/p' /etc/confluent/deploycfg|grep ^- | cut -d ' ' -f 2|sed -e 's/ //')
 for ntpsrv in "$ntpsrvs"; do
     setdebopt clock-setup/ntp true boolean
-    setdebopt clock-setup/ntep-server $ntpsrv string
+    setdebopt clock-setup/ntp-server $ntpsrv string
 done
-#setdebopt console-setup/layoutcode $(grep ^keymap: /etc/confluent/deploycfg|cut -d ' ' -f 2) string
 setdebopt debian-installer/locale $(grep ^locale: /etc/confluent/deploycfg|cut -d ' ' -f 2) select
 domainname=$(grep ^dnsdomain: /etc/confluent/deploycfg|cut -d ' ' -f 2)
 if [ ! -z "$domainname" ] && [ "$domainname" != "null" ]; then
@@ -95,3 +156,54 @@ fi
 
 
 
+
+wget https://$mgr/confluent-public/os/$profile/scripts/pre.sh
+chmod u+x pre.sh
+wget https://$mgr/confluent-public/os/$profile/scripts/prechroot.sh
+chmod u+x prechroot.sh
+wget https://$mgr/confluent-public/os/$profile/scripts/post.sh
+chmod u+x post.sh
+wget https://$mgr/confluent-public/os/$profile/preseed.cfg
+cat preseed.cfg >> /preseed.cfg
+echo $mgr > /etc/confluent/deployer
+setdebopt auto-install/enable true boolean
+setdebopt partman/early_command  $predir/pre.sh string
+setdebopt preseed/late_command $predir/prechroot.sh string
+mv $predir/post.sh /tmp/
+cd -
+ip -4 a flush dev $nic
+setdebopt netcfg/choose_interface $nic select
+setdebopt netcfg/get_hostname $nodename string
+setdebopt netcfg/hostname $nodename string
+setdebopt mirror/protocol https string
+setdebopt mirror/country manual string
+setdebopt mirror/https/hostname deb.debian.org string
+setdebopt mirror/https/directory /debian/ string
+setdebopt mirror/protocol https string
+setdebopt mirror/https/proxy "" string
+#setdebopt apt-setup/security_host $mgr string
+if [ ! -e /dev/disk/by-label/CNFLNT_IDNT ]; then
+    v4cfgmeth=$(grep ipv4_method: /etc/confluent/deploycfg |cut -d: -f 2|sed -e 's/ //')
+    if [ "$v4cfgmeth" = "static" ]; then
+        setdebopt netcfg/disable_dhcp true boolean
+        v4addr=$(grep ^ipv4_address: /etc/confluent/deploycfg|cut -d: -f 2|sed -e 's/ //')
+        v4gw=$(grep ^ipv4_gateway: /etc/confluent/deploycfg|cut -d: -f 2| sed -e 's/ //')
+        if [ "$v4gw" = "null" ]; then
+            v4gw=""
+        fi
+        v4nm=$(grep ^ipv4_netmask: /etc/confluent/deploycfg|cut -d: -f 2|sed -e 's/ //')
+        setdebopt netcfg/get_netmask $v4nm string
+        setdebopt netcfg/get_ipaddress $v4addr string
+        setdebopt netcfg/confirm_static true boolean
+        if [ ! -z "$v4gw" ]; then
+            setdebopt netcfg/get_gateway $v4gw string
+        fi
+        namesrvs=$(sed -n '/^nameservers:/,/^[^-]/p' /etc/confluent/deploycfg|grep ^- | cut -d ' ' -f 2|sed -e 's/ //')
+        for namesrv in "$namesrvs"; do
+            setdebopt netcfg/get_nameservers $namesrv string
+        done
+    elif [ "$vpcfgmeth" = "dhcp" ]; then
+        setdebopt netcfg/disable_dhcp false boolean
+        setdebopt netcfg/confirm_static false boolean
+    fi
+fi

confluent_osdeploy/debian/profiles/default/initprofile.sh

@@ -0,0 +1,8 @@
+#!/bin/sh
+sed -i 's/label: debian/label: Debian/' $2/profile.yaml && \
+ln -s $1/linux $2/boot/kernel && \
+ln -s $1/initrd.gz $2/boot/initramfs/distribution && \
+mkdir -p $2/boot/efi/boot && \
+rm $2/distribution && \
+mcopy -i $1/boot/grub/efi.img ::/efi/boot/* $2/boot/efi/boot
+

confluent_osdeploy/debian/profiles/default/preseed.cfg

@@ -0,0 +1,27 @@
+d-i anna/choose_modules string openssh-server-udeb
+d-i partman-auto/method string regular
+d-i partman-lvm/device_remove_lvm boolean true
+d-i partman-md/device_remove_md boolean true
+d-i partman-auto/expert_recipe_file string /tmp/partitionfile
+d-i partman/confirm_write_new_label boolean true
+d-i partman/choose_partition select finish
+d-i partman/confirm boolean true
+d-i partman/confirm_nooverwrite boolean true
+d-i passwd/make-user boolean false
+d-i clock-setup/utc boolean true
+d-i apt-setup/multiverse boolean false
+d-i apt-setup/universe boolean false
+d-i apt-setup/backports boolean false
+d-i apt-setup/updates boolean false
+d-i grub-installer/only_debian boolean true
+tasksel tasksel/first multiselect standard
+d-i pkgsel/include string openssh-server curl
+d-i pkgsel/update-policy select none
+d-i pkgsel/updatedb boolean false
+d-i finish-install/reboot_in_progress note
+popularity-contest popularity-contest/participate boolean false
+d-i partman-auto/method string lvm
+d-i partman-auto/choose_recipe select atomic
+d-i partman-lvm/confirm boolean true
+d-i partman-lvm/confirm_nooverwrite boolean true
+d-i partman-auto-lvm/guided_size string max

confluent_osdeploy/debian/profiles/default/profile.yaml

@@ -0,0 +1,3 @@
+label: %%DISTRO%% %%VERSION%% %%ARCH%% (Default Profile)
+kernelargs: quiet osprofile=%%PROFILE%%
+#installedargs: example # These arguments would be added to the installed system

confluent_osdeploy/debian/profiles/default/scripts/confignet

@@ -0,0 +1,570 @@
+#!/usr/bin/python
+
+import glob
+import json
+import os
+import socket
+import sys
+import time
+import shlex
+import subprocess
+try:
+    import yaml
+except ImportError:
+    pass
+try:
+    from importlib.machinery import SourceFileLoader
+    def load_source(mod, path):
+        return SourceFileLoader(mod, path).load_module()
+except ImportError:
+    from imp import load_source
+
+try:
+    apiclient = load_source('apiclient', '/opt/confluent/bin/apiclient')
+except IOError:
+    apiclient = load_source('apiclient', '/etc/confluent/apiclient')
+
+def add_lla(iface, mac):
+    pieces = mac.split(':')
+    initbyte = int(pieces[0], 16) ^ 2
+    lla = 'fe80::{0:x}{1}:{2}ff:fe{3}:{4}{5}/64'.format(initbyte, pieces[1], pieces[2], pieces[3], pieces[4], pieces[5])
+    try:
+        with open('/proc/sys/net/ipv6/conf/{0}/disable_ipv6'.format(iface), 'w') as setin:
+            setin.write('0')
+        subprocess.check_call(['ip', 'addr', 'add', 'dev', iface, lla, 'scope', 'link'])
+    except Exception:
+        return None
+    return lla
+
+#cli = apiclient.HTTPSClient(json=True)
+#c = cli.grab_url_with_status('/confluent-api/self/netcfg')
+def add_missing_llas():
+    #NetworkManager goes out of its way to suppress ipv6 lla, so will just add some
+    added = {}
+    linkinfo = subprocess.check_output(['ip', '-br', 'l']).decode('utf8')
+    ifaces = {}
+    for line in linkinfo.split('\n'):
+        line = line.strip().split()
+        if not line or 'LOOPBACK' in line[-1] or 'NO-CARRIER' in line[-1]:
+            continue
+        if 'UP' not in line[-1]:
+            subprocess.call(['ip', 'link', 'set', line[0], 'up'])
+        ifaces[line[0]] = line[2]
+    ips = {}
+    ipinfo = subprocess.check_output(['ip', '-br', '-6', 'a']).decode('utf8')
+    for line in ipinfo.split('\n'):
+        line = line.strip().split(None, 2)
+        if not line:
+            continue
+        ips[line[0]] = line[2]
+    for iface in ifaces:
+        for addr in ips.get(iface, '').split():
+            if addr.startswith('fe80::'):
+                break
+        else:
+            newlla = add_lla(iface, ifaces[iface])
+            if newlla:
+                added[iface] = newlla
+    return added
+
+def rm_tmp_llas(tmpllas):
+    for iface in tmpllas:
+        subprocess.check_call(['ip', 'addr', 'del', 'dev', iface, tmpllas[iface]])
+
+def await_tentative():
+    maxwait = 10
+    while b'tentative' in subprocess.check_output(['ip', 'a']):
+        if maxwait == 0:
+            break
+        maxwait -= 1
+        time.sleep(1)
+
+def map_idx_to_name():
+    map = {}
+    devtype = {}
+    prevdev = None
+    for line in subprocess.check_output(['ip', 'l']).decode('utf8').splitlines():
+        if line.startswith(' ') and 'link/' in line:
+            typ = line.split()[0].split('/')[1]
+            devtype[prevdev] = typ if typ != 'ether' else 'ethernet'
+        if line.startswith(' '):
+            continue
+        idx, iface, rst = line.split(':', 2)
+        prevdev = iface.strip()
+        rst = rst.split()
+        try:
+            midx = rst.index('master')
+            continue
+        except ValueError:
+            pass
+        idx = int(idx)
+        iface = iface.strip()
+        map[idx] = iface
+    return map, devtype
+
+
+def get_interface_name(iname, settings):
+    explicitname = settings.get('interface_names', None)
+    if explicitname:
+        return explicitname
+    if settings.get('current_nic', False):
+        return iname
+    return None
+
+class NetplanManager(object):
+    def __init__(self, deploycfg):
+        self.cfgbydev = {}
+        self.read_connections()
+        self.deploycfg = deploycfg
+
+    def read_connections(self):
+        for plan in glob.glob('/etc/netplan/*.y*ml'):
+            with open(plan) as planfile:
+                planinfo = yaml.safe_load(planfile)
+                if not planinfo:
+                    continue
+            nicinfo = planinfo.get('network', {}).get('ethernets', {})
+            for devname in nicinfo:
+                if devname == 'lo':
+                    continue
+                if 'gateway4' in nicinfo[devname]:
+                    # normalize deprecated syntax on read in
+                    gw4 = nicinfo[devname]['gateway4']
+                    del nicinfo[devname]['gateway4']
+                    routeinfo = nicinfo[devname].get('routes', [])
+                    for ri in routeinfo:
+                        if ri.get('via', None) == gw4 and ri.get('to', None) in ('default', '0.0.0.0/0', '0/0'):
+                            break
+                    else:
+                        routeinfo.append({
+                            'to': 'default',
+                            'via': gw4
+                        })
+                    nicinfo[devname]['routes'] = routeinfo
+                self.cfgbydev[devname] = nicinfo[devname]
+    
+    def apply_configuration(self, cfg):
+        devnames = cfg['interfaces']
+        if len(devnames) != 1:
+            raise Exception('Multi-nic team/bonds not yet supported')
+        stgs = cfg['settings']
+        needcfgapply = False
+        for devname in devnames:
+            needcfgwrite = False
+            # ipv6_method missing at uconn...
+            if stgs.get('ipv6_method', None) == 'static':
+                curraddr = stgs['ipv6_address']
+                currips = self.getcfgarrpath([devname, 'addresses'])
+                if curraddr not in currips:
+                    needcfgwrite = True
+                    currips.append(curraddr)
+            if stgs.get('ipv4_method', None) == 'static':
+                curraddr = stgs['ipv4_address']
+                currips = self.getcfgarrpath([devname, 'addresses'])
+                if curraddr not in currips:
+                    needcfgwrite = True
+                    currips.append(curraddr)
+            if stgs.get('mtu', None):
+                devdict = self.getcfgarrpath([devname])
+                devdict['mtu'] = int(stgs['mtu'])
+            gws = []
+            gws.append(stgs.get('ipv4_gateway', None))
+            gws.append(stgs.get('ipv6_gateway', None))
+            for gwaddr in gws:
+                if gwaddr:
+                    cfgroutes = self.getcfgarrpath([devname, 'routes'])
+                    for rinfo in cfgroutes:
+                        if rinfo.get('via', None) == gwaddr:
+                            break
+                    else:
+                        needcfgwrite = True
+                        cfgroutes.append({'via': gwaddr, 'to': 'default'})
+            dnsips = self.deploycfg.get('nameservers', [])
+            dnsdomain = self.deploycfg.get('dnsdomain', '')
+            if dnsips:
+                currdnsips = self.getcfgarrpath([devname, 'nameservers', 'addresses'])
+                for dnsip in dnsips:
+                    if dnsip and dnsip not in currdnsips:
+                        needcfgwrite = True
+                        currdnsips.append(dnsip)
+            if dnsdomain:
+                currdnsdomain = self.getcfgarrpath([devname, 'nameservers', 'search'])
+                if dnsdomain not in currdnsdomain:
+                    needcfgwrite = True
+                    currdnsdomain.append(dnsdomain)
+            if needcfgwrite:
+                needcfgapply = True
+                newcfg = {'network': {'version': 2, 'ethernets': {devname: self.cfgbydev[devname]}}}
+                oumask = os.umask(0o77)
+                with open('/etc/netplan/{0}-confluentcfg.yaml'.format(devname), 'w') as planout:
+                    planout.write(yaml.dump(newcfg))
+                os.umask(oumask)
+        if needcfgapply:
+            subprocess.call(['netplan', 'apply'])
+
+    def getcfgarrpath(self, devpath):
+        currptr = self.cfgbydev
+        for k in devpath[:-1]:
+            if k not in currptr:
+                currptr[k] = {}
+            currptr = currptr[k]
+        if devpath[-1] not in currptr:
+            currptr[devpath[-1]] = []
+        return currptr[devpath[-1]]
+        
+
+
+class WickedManager(object):
+    def __init__(self):
+        self.teamidx = 0
+        self.read_connections()
+
+    def read_connections(self):
+        self.cfgbydev = {}
+        for ifcfg in glob.glob('/etc/sysconfig/network/ifcfg-*'):
+            devname = ifcfg.replace('/etc/sysconfig/network/ifcfg-', '')
+            if devname == 'lo':
+                continue
+            currcfg = {}
+            self.cfgbydev[devname] = currcfg
+            for cfg in open(ifcfg).read().splitlines():
+                cfg = cfg.split('#', 1)[0]
+                try:
+                    kv = ' '.join(shlex.split(cfg)).split('=', 1)
+                except Exception:
+                    # unparseable line, likely having something we can't handle
+                    del self.cfgbydev[devname]
+                if len(kv) != 2:
+                    continue
+                k, v = kv
+                k = k.strip()
+                v = v.strip()
+                currcfg[k] = v
+
+    def apply_configuration(self, cfg):
+        stgs = cfg['settings']
+        ipcfg = 'STARTMODE=auto\n'
+        routecfg = ''
+        bootproto4 = stgs.get('ipv4_method', 'none')
+        bootproto6 = stgs.get('ipv6_method', 'none')
+        if bootproto4 == 'dhcp' and bootproto6 == 'dhcp':
+            ipcfg += 'BOOTPROTO=dhcp\n'
+        elif bootproto4 == 'dhcp':
+            ipcfg += 'BOOTPROTO=dhcp4\n'
+        elif bootproto6 == 'dhcp':
+            ipcfg += 'BOOTPROTO=dhcp6\n'
+        else:
+            ipcfg += 'BOOTPROTO=static\n'
+        if stgs.get('ipv4_address', None):
+            ipcfg += 'IPADDR=' + stgs['ipv4_address'] + '\n'
+        v4gw = stgs.get('ipv4_gateway', None)
+        if stgs.get('ipv6_address', None):
+            ipcfg += 'IPADDR_V6=' + stgs['ipv6_address'] + '\n'
+        v6gw = stgs.get('ipv6_gateway', None)
+        cname = None
+        if len(cfg['interfaces']) > 1:  # creating new team
+            if not stgs.get('team_mode', None):
+                sys.stderr.write("Warning, multiple interfaces ({0}) without a team_mode, skipping setup\n".format(','.join(cfg['interfaces'])))
+                return
+            if not stgs.get('connection_name', None):
+                stgs['connection_name'] = 'bond{0}'.format(self.teamidx)
+                self.teamidx += 1
+            cname = stgs['connection_name']
+            with open('/etc/sysconfig/network/ifcfg-{0}'.format(cname), 'w') as teamout:
+                teamout.write(ipcfg)
+                if stgs['team_mode'] == 'lacp':
+                    stgs['team_mode'] = '802.3ad'
+                teamout.write("BONDING_MODULE_OPTS='mode={0} miimon=100'\nBONDING_MASTER=yes\n".format(stgs['team_mode']))
+                idx = 1
+                for iface in cfg['interfaces']:
+                    subprocess.call(['wicked', 'ifdown', iface])
+                    try:
+                        os.remove('/etc/sysconfig/network/ifcfg-{0}'.format(iface))
+                        os.remove('/etc/sysconfig/network/ifroute-{0}'.format(iface))
+                    except OSError:
+                        pass
+                    teamout.write('BONDING_SLAVE{0}={1}\n'.format(idx, iface))
+                    idx += 1
+        else:
+            cname = list(cfg['interfaces'])[0]
+            priorcfg = self.cfgbydev.get(cname, {})
+            for cf in priorcfg:
+                if cf.startswith('TEAM_'):
+                    ipcfg += '{0}={1}\n'.format(cf, priorcfg[cf])
+            with open('/etc/sysconfig/network/ifcfg-{0}'.format(cname), 'w') as iout:
+                iout.write(ipcfg)
+        if v4gw:
+            routecfg += 'default {0} - {1}\n'.format(v4gw, cname)
+        if v6gw:
+            routecfg += 'default {0} - {1}\n'.format(v6gw, cname)
+        if routecfg:
+            with open('/etc/sysconfig/network/ifroute-{0}'.format(cname), 'w') as routeout:
+                routeout.write(routecfg)
+        subprocess.call(['wicked', 'ifup', cname])
+
+
+class NetworkManager(object):
+    bondtypes = {
+            'lacp': '802.3ad',
+            'loadbalance': 'balance-alb',
+            'roundrobin': 'balance-rr',
+            'activebackup': 'active-backup',
+    }
+    def __init__(self, devtypes, deploycfg):
+        self.deploycfg = deploycfg
+        self.connections = {}
+        self.uuidbyname = {}
+        self.uuidbydev = {}
+        self.connectiondetail = {}
+        self.read_connections()
+        self.teamidx = 0
+        self.devtypes = devtypes
+
+    def read_connections(self):
+        self.connections = {}
+        self.uuidbyname = {}
+        self.uuidbydev = {}
+        self.connectiondetail = {}
+        ci = subprocess.check_output(['nmcli', '-t', 'c']).decode('utf8')
+        for inf in ci.splitlines():
+            n, u, t, dev = inf.split(':')
+            if n == 'NAME':
+                continue
+            if dev == '--':
+                dev = None
+            self.uuidbyname[n] = u
+            if dev:
+                self.uuidbydev[dev] = u
+            self.connections[u] = {'name': n, 'uuid': u, 'type': t, 'dev': dev}
+            deats = {}
+            for deat in subprocess.check_output(['nmcli', 'c', 's', u]).decode('utf8').splitlines():
+                k, v = deat.split(':', 1)
+                v = v.strip()
+                if v == '--':
+                    continue
+                if '(default)' in v:
+                    continue
+                deats[k] = v
+            self.connectiondetail[u] = deats
+
+
+    def add_team_member(self, team, member):
+        bondcfg = {}
+        if member in self.uuidbydev:
+            myuuid = self.uuidbydev[member]
+            deats = self.connectiondetail[myuuid]
+            currteam = deats.get('connection.master', None)
+            if currteam == team:
+                return
+            for stg in ('ipv4.dhcp-hostname', 'ipv4.dns', 'ipv6.dns', 'ipv6.dhcp-hostname'):
+                if deats.get(stg, None):
+                    bondcfg[stg] = deats[stg]
+        if member in self.uuidbyname:
+            subprocess.check_call(['nmcli', 'c', 'del', self.uuidbyname[member]])
+        devtype = self.devtypes.get(member, 'bond-slave')
+        subprocess.check_call(['nmcli', 'c', 'add', 'type', devtype, 'master', team, 'con-name', member, 'connection.interface-name', member])
+        if bondcfg:
+            args = []
+            for parm in bondcfg:
+                args.append(parm)
+                args.append(bondcfg[parm])
+            subprocess.check_call(['nmcli', 'c', 'm', team] + args)
+
+    def apply_configuration(self, cfg, lastchance=False):
+        cmdargs = {}
+        cmdargs['connection.autoconnect'] = 'yes'
+        stgs = cfg['settings']
+        cmdargs['ipv6.method'] = stgs.get('ipv6_method', 'link-local')
+        if stgs.get('ipv6_address', None):
+            cmdargs['ipv6.addresses'] = stgs['ipv6_address']
+        cmdargs['ipv4.method'] = stgs.get('ipv4_method', 'disabled')
+        if stgs.get('ipv4_address', None):
+            cmdargs['ipv4.addresses'] = stgs['ipv4_address']
+        if stgs.get('ipv4_gateway', None):
+            cmdargs['ipv4.gateway'] = stgs['ipv4_gateway']
+        if stgs.get('ipv6_gateway', None):
+            cmdargs['ipv6.gateway'] = stgs['ipv6_gateway']
+        if stgs.get('mtu', None):
+            cmdargs['802-3-ethernet.mtu'] = stgs['mtu']
+        dnsips = self.deploycfg.get('nameservers', [])
+        if not dnsips:
+            dnsips = []
+        dns4 = []
+        dns6 = []
+        for dnsip in dnsips:
+            if '.' in dnsip:
+                dns4.append(dnsip)
+            elif ':' in dnsip:
+                dns6.append(dnsip)
+        if dns4:
+            cmdargs['ipv4.dns'] = ','.join(dns4)
+        if dns6:
+            cmdargs['ipv6.dns'] = ','.join(dns6)
+        if len(cfg['interfaces']) > 1:  # team time.. should be..
+            if not cfg['settings'].get('team_mode', None):
+                sys.stderr.write("Warning, multiple interfaces ({0}) without a team_mode, skipping setup\n".format(','.join(cfg['interfaces'])))
+                return
+            if not cfg['settings'].get('connection_name', None):
+                cfg['settings']['connection_name'] = 'team{0}'.format(self.teamidx)
+                self.teamidx += 1
+            cname = cfg['settings']['connection_name']
+            cargs = []
+            for arg in cmdargs:
+                cargs.append(arg)
+                cargs.append('{}'.format(cmdargs[arg]))
+            if stgs['team_mode'] in self.bondtypes:
+                stgs['team_mode'] = self.bondtypes[stgs['team_mode']]
+            subprocess.check_call(['nmcli', 'c', 'add', 'type', 'bond', 'con-name', cname, 'connection.interface-name', cname, 'bond.options', 'miimon=100,mode={}'.format(stgs['team_mode'])] + cargs)
+            for iface in cfg['interfaces']:
+                self.add_team_member(cname, iface)
+            subprocess.check_call(['nmcli', 'c', 'u', cname])
+        else:
+            cname = stgs.get('connection_name', None)
+            iname = list(cfg['interfaces'])[0]
+            ctype = self.devtypes.get(iname, None)
+            if not ctype:
+                if lastchance:
+                    sys.stderr.write("Warning, no device found for interface_name ({0}), skipping setup\n".format(iname))
+                return 1
+            if stgs.get('vlan_id', None):
+                vlan = stgs['vlan_id']
+                if ctype == 'infiniband':
+                    vlan = '0x{0}'.format(vlan) if not vlan.startswith('0x') else vlan
+                    cmdargs['infiniband.parent'] = iname
+                    cmdargs['infiniband.p-key'] = vlan
+                    iname = '{0}.{1}'.format(iname, vlan[2:])
+                elif ctype == 'ethernet':
+                    ctype = 'vlan'
+                    cmdargs['vlan.parent'] = iname
+                    cmdargs['vlan.id'] = vlan
+                    iname = '{0}.{1}'.format(iname, vlan)
+                else:
+                    sys.stderr.write("Warning, unknown interface_name ({0}) device type ({1}) for VLAN/PKEY, skipping setup\n".format(iname, ctype))
+                    return
+            cname = iname if not cname else cname
+            u = self.uuidbyname.get(cname, None)
+            cargs = []
+            for arg in cmdargs:
+                cargs.append(arg)
+                cargs.append('{}'.format(cmdargs[arg]))
+            if u:
+                subprocess.check_call(['nmcli', 'c', 'm', u, 'connection.interface-name', iname] + cargs)
+                subprocess.check_call(['nmcli', 'c', 'u', u])
+            else:
+                subprocess.check_call(['nmcli', 'c', 'add', 'type', ctype, 'con-name', cname, 'connection.interface-name', iname] + cargs)
+                self.read_connections()
+                u = self.uuidbyname.get(cname, None)
+                if u:
+                    subprocess.check_call(['nmcli', 'c', 'u', u])
+
+
+
+if __name__ == '__main__':
+    havefirewall = subprocess.call(['systemctl', 'status', 'firewalld'])
+    havefirewall = havefirewall == 0
+    if havefirewall:
+        subprocess.check_call(['systemctl', 'stop', 'firewalld'])
+    tmpllas = add_missing_llas()
+    await_tentative()
+    idxmap, devtypes = map_idx_to_name()
+    netname_to_interfaces = {}
+    myaddrs = apiclient.get_my_addresses()
+    srvs, _ = apiclient.scan_confluents()
+    doneidxs = set([])
+    dc = None
+    if not srvs:  # the multicast scan failed, fallback to deploycfg cfg file
+        with open('/etc/confluent/confluent.deploycfg', 'r') as dci:
+            for cfgline in dci.read().split('\n'):
+                if cfgline.startswith('deploy_server:'):
+                    srvs = [cfgline.split()[1]]
+                    break
+    for srv in srvs:
+        try:
+            s = socket.create_connection((srv, 443))
+        except socket.error:
+            continue
+        myname = s.getsockname()
+        s.close()
+        if len(myname) == 4:
+            curridx = myname[-1]
+        else:
+            myname = myname[0]
+            myname = socket.inet_pton(socket.AF_INET, myname)
+            for addr in myaddrs:
+                if myname == addr[1].tobytes():
+                    curridx = addr[-1]
+        if curridx in doneidxs:
+            continue
+        for tries in (1, 2, 3):
+            try:
+                status, nc = apiclient.HTTPSClient(usejson=True, host=srv).grab_url_with_status('/confluent-api/self/netcfg')
+                break
+            except Exception:
+                if tries == 3:
+                    raise
+                time.sleep(1)
+                continue
+        nc = json.loads(nc)
+        if not dc:
+            for tries in (1, 2, 3):
+                try:
+                    status, dc = apiclient.HTTPSClient(usejson=True, host=srv).grab_url_with_status('/confluent-api/self/deploycfg2')
+                    break
+                except Exception:
+                    if tries == 3:
+                        raise
+                    time.sleep(1)
+                    continue
+            dc = json.loads(dc)
+        iname = get_interface_name(idxmap[curridx], nc.get('default', {}))
+        if iname:
+            for iname in iname.split(','):
+                if 'default' in netname_to_interfaces:
+                    netname_to_interfaces['default']['interfaces'].add(iname)
+                else:
+                    netname_to_interfaces['default'] = {'interfaces': set([iname]), 'settings': nc['default']}
+        for netname in nc.get('extranets', {}):
+            uname = '_' + netname
+            iname = get_interface_name(idxmap[curridx], nc['extranets'][netname])
+            if iname:
+                for iname in iname.split(','):
+                    if uname in netname_to_interfaces:
+                        netname_to_interfaces[uname]['interfaces'].add(iname)
+                    else:
+                        netname_to_interfaces[uname] = {'interfaces': set([iname]), 'settings': nc['extranets'][netname]}
+        doneidxs.add(curridx)
+    if 'default' in netname_to_interfaces:
+        for netn in netname_to_interfaces:
+            if netn == 'default':
+                continue
+            netname_to_interfaces['default']['interfaces'] -= netname_to_interfaces[netn]['interfaces']
+        if not netname_to_interfaces['default']['interfaces']:
+            del netname_to_interfaces['default']
+    # Make sure VLAN/PKEY connections are created last
+    netname_to_interfaces = dict(sorted(netname_to_interfaces.items(), key=lambda item: 'vlan_id' in item[1]['settings']))
+    rm_tmp_llas(tmpllas)
+    if os.path.exists('/usr/sbin/netplan'):
+        nm = NetplanManager(dc)
+    if os.path.exists('/usr/bin/nmcli'):
+        nm = NetworkManager(devtypes, dc)
+    elif os.path.exists('/usr/sbin/wicked'):
+        nm = WickedManager()
+    retrynics = []
+    for netn in netname_to_interfaces:
+        redo = nm.apply_configuration(netname_to_interfaces[netn])
+        if redo == 1:
+            retrynics.append(netn)
+    if retrynics:
+        idxmap, devtypes = map_idx_to_name()
+        if os.path.exists('/usr/sbin/netplan'):
+            nm = NetplanManager(dc)
+        if os.path.exists('/usr/bin/nmcli'):
+            nm = NetworkManager(devtypes, dc)
+        elif os.path.exists('/usr/sbin/wicked'):
+            nm = WickedManager()
+        for netn in retrynics:
+            nm.apply_configuration(netname_to_interfaces[netn], lastchance=True)
+    if havefirewall:
+        subprocess.check_call(['systemctl', 'start', 'firewalld'])
+    await_tentative()
+

confluent_osdeploy/debian/profiles/default/scripts/firstboot.service

@@ -0,0 +1,11 @@
+[Unit]
+Description=First Boot Process
+Requires=network-online.target
+After=network-online.target
+
+[Service]
+ExecStart=/opt/confluent/bin/firstboot.sh
+
+[Install]
+WantedBy=multi-user.target
+

confluent_osdeploy/debian/profiles/default/scripts/firstboot.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+echo "Confluent first boot is running"
+HOME=$(getent passwd $(whoami)|cut -d: -f 6)
+export HOME
+#cp -a /etc/confluent/ssh/* /etc/ssh/
+#systemctl restart sshd
+rootpw=$(grep ^rootpassword: /etc/confluent/confluent.deploycfg |awk '{print $2}')
+if [ ! -z "$rootpw" -a "$rootpw" != "null" ]; then
+	echo root:$rootpw | chpasswd -e
+fi
+nodename=$(grep ^NODENAME: /etc/confluent/confluent.info | awk '{print $2}')
+confluent_apikey=$(cat /etc/confluent/confluent.apikey)
+confluent_mgr=$(grep ^deploy_server: /etc/confluent/confluent.deploycfg |awk '{print $2}')
+while ! ping -c 1 $confluent_mgr >& /dev/null; do
+	sleep 1
+done
+source /etc/confluent/functions
+
+run_remote_parts firstboot.d
+run_remote_config firstboot.d
+systemctl disable firstboot
+curl -f -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $confluent_apikey" -X POST -d "status: complete" https://$confluent_mgr/confluent-api/self/updatestatus

confluent_osdeploy/debian/profiles/default/scripts/functions

@@ -0,0 +1,216 @@
+#!/bin/bash
+function test_mgr() {
+    whost=$1
+    if [[ "$whost" == *:* ]] && [[ "$whost" != *[* ]] ; then
+        whost="[$whost]"
+    fi
+    if curl -gs https://${whost}/confluent-api/ > /dev/null; then
+        return 0
+    fi
+    return 1
+}
+
+function initconfluentscriptstmp() {
+    if [ -z "$confluentscripttmpdir" ]; then
+        mkdir -p /opt/confluent/tmpexec
+        confluentscripttmpdir=$(mktemp -d /opt/confluent/tmpexec/confluentscripts.XXXXXXXXX)
+    fi
+}
+
+function confluentpython() {
+    if [ -x /usr/libexec/platform-python ]; then
+        /usr/libexec/platform-python $*
+    elif [ -x /usr/bin/python3 ]; then
+        /usr/bin/python3 $*
+    elif [ -x /usr/bin/python ]; then
+        /usr/bin/python $*
+    elif [ -x /usr/bin/python2 ]; then
+        /usr/bin/python2 $*
+    fi
+}
+
+function set_confluent_vars() {
+    if [ -z "$nodename" ]; then
+        nodename=$(grep ^NODENAME: /etc/confluent/confluent.info | awk '{print $2}')
+    fi
+    if [[ "$confluent_mgr" == *"%"* ]]; then
+        confluent_mgr=""
+    fi
+    if [ -z "$confluent_mgr" ]; then
+        confluent_mgr=$(grep ^deploy_server: /etc/confluent/confluent.deploycfg | sed -e 's/[^ ]*: //')
+        if ! test_mgr $confluent_mgr; then
+            confluent_mgr=$(grep ^deploy_server_v6: /etc/confluent/confluent.deploycfg | sed -e 's/[^ ]*: //')
+            if [[ "$confluent_mgr" = *":"* ]]; then
+                confluent_mgr="[$confluent_mgr]"
+            fi
+        fi
+        if ! test_mgr $confluent_mgr; then
+            BESTMGRS=$(grep ^EXTMGRINFO: /etc/confluent/confluent.info | grep '|1$' | sed -e 's/EXTMGRINFO: //' -e 's/|.*//')
+            OKMGRS=$(grep ^EXTMGRINFO: /etc/confluent/confluent.info | grep '|0$' | sed -e 's/EXTMGRINFO: //' -e 's/|.*//')
+            for confluent_mgr in $BESTMGRS $OKMGRS; do
+                if [[ $confluent_mgr == *":"* ]]; then
+                    confluent_mgr="[$confluent_mgr]"
+                fi
+                if test_mgr $confluent_mgr; then
+                    break
+                fi
+            done
+        fi
+    fi
+    if [ -z "$confluent_profile" ]; then
+        confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg | sed -e 's/[^ ]*: //')
+    fi
+    export confluent_profile confluent_mgr nodename
+}
+
+fetch_remote() {
+    curlargs=""
+    if [ -f /etc/confluent/ca.pem ]; then
+        curlargs=" --cacert /etc/confluent/ca.pem"
+    fi
+    set_confluent_vars
+    mkdir -p $(dirname $1)
+    whost=$confluent_mgr
+    if [[ "$whost" == *:* ]] && [[ "$whost" != *[* ]] ; then
+        whost="[$whost]"
+    fi
+    curl -gf -sS $curlargs https://$whost/confluent-public/os/$confluent_profile/scripts/$1 > $1
+    if [ $? != 0 ]; then echo $1 failed to download; return 1; fi
+}
+
+source_remote_parts() {
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
+    apiclient=/opt/confluent/bin/apiclient
+    if [ -f /etc/confluent/apiclient ]; then
+        apiclient=/etc/confluent/apiclient
+    fi
+    scriptlist=$(confluentpython $apiclient /confluent-api/self/scriptlist/$1|sed -e 's/^- //')
+    for script in $scriptlist; do
+        source_remote $1/$script
+    done
+    rm -rf $confluentscripttmpdir
+    unset confluentscripttmpdir
+}
+
+run_remote_parts() {
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
+    apiclient=/opt/confluent/bin/apiclient
+    if [ -f /etc/confluent/apiclient ]; then
+        apiclient=/etc/confluent/apiclient
+    fi
+    scriptlist=$(confluentpython $apiclient /confluent-api/self/scriptlist/$1|sed -e 's/^- //')
+    for script in $scriptlist; do
+        run_remote $1/$script
+    done
+    rm -rf $confluentscripttmpdir
+    unset confluentscripttmpdir
+}
+
+source_remote() {
+    set_confluent_vars
+    unsettmpdir=0
+    echo
+    echo '---------------------------------------------------------------------------'
+    echo Sourcing $1 from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
+    initconfluentscriptstmp
+    echo Sourcing from $confluentscripttmpdir
+    cd $confluentscripttmpdir
+    fetch_remote $1
+    if [ $? != 0 ]; then echo $1 failed to download; return 1; fi
+    chmod +x $1
+    cmd=$1
+    shift
+    source ./$cmd
+    cd - > /dev/null
+    if [ "$unsettmpdir" = 1 ]; then
+        rm -rf $confluentscripttmpdir
+        unset confluentscripttmpdir
+        unsettmpdir=0
+    fi
+    rm -rf $confluentscripttmpdir
+    return $retcode
+}
+
+run_remote() {
+    requestedcmd="'$*'"
+    unsettmpdir=0
+    set_confluent_vars
+    echo
+    echo '---------------------------------------------------------------------------'
+    echo Running $requestedcmd from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
+    if [ -z "$confluentscripttmpdir" ]; then
+        unsettmpdir=1
+    fi
+    initconfluentscriptstmp
+    echo Executing in $confluentscripttmpdir
+    cd $confluentscripttmpdir
+    fetch_remote $1
+    if [ $? != 0 ]; then echo $requestedcmd failed to download; return 1; fi
+    chmod +x $1
+    cmd=$1
+    if [ -x /usr/bin/chcon ]; then
+        chcon system_u:object_r:bin_t:s0 $cmd
+    fi
+    shift
+    ./$cmd $*
+    retcode=$?
+    if [ $retcode -ne 0 ]; then
+        echo "$requestedcmd exited with code $retcode"
+    fi
+    cd - > /dev/null
+    if [ "$unsettmpdir" = 1 ]; then
+        rm -rf $confluentscripttmpdir
+        unset confluentscripttmpdir
+        unsettmpdir=0
+    fi
+    return $retcode
+}
+
+run_remote_python() {
+    echo
+    set_confluent_vars
+    if [ -f /etc/confluent/ca.pem ]; then
+        curlargs=" --cacert /etc/confluent/ca.pem"
+    fi
+    echo '---------------------------------------------------------------------------'
+    echo Running python script "'$*'" from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
+    echo Executing in $confluentscripttmpdir
+    cd $confluentscripttmpdir
+    mkdir -p $(dirname $1)
+    whost=$confluent_mgr
+    if [[ "$whost" == *:* ]] && [[ "$whost" != *[* ]] ; then
+        whost="[$whost]"
+    fi
+    curl -gf -sS $curlargs https://$whost/confluent-public/os/$confluent_profile/scripts/$1 > $1
+    if [ $? != 0 ]; then echo "'$*'" failed to download; return 1; fi
+    confluentpython $*
+    retcode=$?
+    echo "'$*' exited with code $retcode"
+    cd - > /dev/null
+    rm -rf $confluentscripttmpdir
+    unset confluentscripttmpdir
+    return $retcode
+}
+
+run_remote_config() {
+    echo
+    set_confluent_vars
+    apiclient=/opt/confluent/bin/apiclient
+    if [ -f /etc/confluent/apiclient ]; then
+        apiclient=/etc/confluent/apiclient
+    fi
+    echo '---------------------------------------------------------------------------'
+    echo Requesting to run remote configuration for "'$*'" from $confluent_mgr under profile $confluent_profile
+    confluentpython $apiclient /confluent-api/self/remoteconfig/"$*" -d {}
+    confluentpython $apiclient /confluent-api/self/remoteconfig/status -w 204
+    echo
+    echo 'Completed remote configuration'
+    echo '---------------------------------------------------------------------------'
+    return
+}
+#If invoked as a command, use the arguments to actually run a function
+(return 0 2>/dev/null) || $1 "${@:2}"

confluent_osdeploy/debian/profiles/default/scripts/post.sh

@@ -0,0 +1,67 @@
+#!/bin/bash
+mkdir -p /run/sshd
+mkdir -p /root/.ssh
+cat /tmp/ssh/*pubkey >> /root/.ssh/authorized_keys
+cat /tmp/ssh/*.ca | sed -e s/^/'@cert-authority * '/ >> /etc/ssh/ssh_known_hosts
+chmod 700 /etc/confluent
+chmod go-rwx /etc/confluent/*
+sshconf=/etc/ssh/ssh_config
+if [ -d /etc/ssh/ssh_config.d/ ]; then
+    sshconf=/etc/ssh/ssh_config.d/01-confluent.conf
+fi
+echo 'Host *' >> $sshconf
+echo '    HostbasedAuthentication yes' >> $sshconf
+echo '    EnableSSHKeysign yes' >> $sshconf
+echo '    HostbasedKeyTypes *ed25519*' >> $sshconf
+/usr/sbin/sshd
+confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg | awk '{print $2}')
+mkdir -p /opt/confluent/bin
+python3 /opt/confluent/bin/apiclient /confluent-public/os/$confluent_profile/scripts/firstboot.sh > /opt/confluent/bin/firstboot.sh
+chmod +x /opt/confluent/bin/firstboot.sh
+python3 /opt/confluent/bin/apiclient /confluent-public/os/$confluent_profile/scripts/firstboot.service > /etc/systemd/system/firstboot.service
+systemctl enable firstboot
+python3 /opt/confluent/bin/apiclient /confluent-public/os/$confluent_profile/scripts/functions > /etc/confluent/functions
+source /etc/confluent/functions
+python3 /opt/confluent/bin/apiclient /confluent-api/self/nodelist | sed -e s/'^- //' > /tmp/allnodes
+cp /tmp/allnodes /root/.shosts
+cp /tmp/allnodes /etc/ssh/shosts.equiv
+if grep ^ntpservers: /etc/confluent/confluent.deploycfg > /dev/null; then
+    ntps=$(sed -n '/^ntpservers:/,/^[^-]/p' /etc/confluent/confluent.deploycfg|sed 1d|sed '$d' | sed -e 's/^- //' | paste -sd ' ')
+    sed -i "s/#NTP=/NTP=$ntps/" /etc/systemd/timesyncd.conf
+fi
+textcons=$(grep ^textconsole: /etc/confluent/confluent.deploycfg |awk '{print $2}')
+updategrub=0
+if [ "$textcons" = "true" ] && ! grep console= /proc/cmdline > /dev/null; then
+    cons=""
+    if [ -f /tmp/autocons.info ]; then
+        cons=$(cat /tmp/autocons.info)
+    fi
+    if [ ! -z "$cons" ]; then
+        sed -i 's/GRUB_CMDLINE_LINUX="\([^"]*\)"/GRUB_CMDLINE_LINUX="\1 console='${cons#/dev/}'"/' /etc/default/grub
+        updategrub=1
+    fi
+fi
+kargs=$(python3 /opt/confluent/bin/apiclient /confluent-public/os/$confluent_profile/profile.yaml | grep ^installedargs: | sed -e 's/#.*//')
+if [ ! -z "$kargs" ]; then
+    sed -i 's/GRUB_CMDLINE_LINUX="\([^"]*\)"/GRUB_CMDLINE_LINUX="\1 '"${kargs}"'"/' /etc/default/grub
+fi
+
+if [ 1 = $updategrub ]; then
+    update-grub
+fi
+
+if [ -e /sys/firmware/efi ]; then
+    bootnum=$(efibootmgr | grep ubuntu | sed -e 's/ .*//' -e 's/\*//' -e s/Boot//)
+    if [ ! -z "$bootnum" ]; then
+        currboot=$(efibootmgr | grep ^BootOrder: | awk '{print $2}')
+        nextboot=$(echo $currboot| awk -F, '{print $1}')
+        [ "$nextboot" = "$bootnum" ] || efibootmgr -o $bootnum,$currboot
+        efibootmgr -D
+    fi
+fi
+run_remote_python syncfileclient
+run_remote_parts post.d
+run_remote_config post
+
+python3 /opt/confluent/bin/apiclient /confluent-api/self/updatestatus  -d 'status: staged'
+

confluent_osdeploy/debian/profiles/default/scripts/pre.sh

@@ -1,18 +1,88 @@
-anna-install openssh-server-udeb
-mkdir -p ~/.ssh/
-cat /ssh/*pubkey > ~/.ssh/authorized_keys
-ssh-keygen -A
-mgr=$(grep ^MANAGER:.*\\. /etc/confluent/confluent.info|head -n 1|cut -d: -f 2|sed -e 's/ //')
-nodename=$(grep ^NODENAME: /etc/confluent/confluent.info|head -n 1|cut -d: -f 2|sed -e 's/ //')
+#!/bin/sh
+## Use the following option to add additional boot parameters for the
+## installed system (if supported by the bootloader installer).
+## Note: options passed to the installer will be added automatically.
+#d-i debian-installer/add-kernel-opts string [from profile.yaml]
+deploycfg=/etc/confluent/confluent.deploycfg
+mgr=$(cat /etc/confluent/deployer)
+
+cryptboot=$(grep encryptboot: $deploycfg|sed -e 's/^encryptboot: //')
+if [ "$cryptboot" != "" ]  && [ "$cryptboot" != "none" ] && [ "$cryptboot" != "null" ]; then
+   echo "****Encrypted boot requested, but not implemented for this OS, halting install" > /dev/console
+   [ -f '/tmp/autoconsdev' ] && (echo "****Encryptod boot requested, but not implemented for this OS,halting install" >> $(cat /tmp/autoconsdev))
+   while :; do sleep 86400; done
+fi
+cat > /usr/lib/live-installer.d/confluent-certs << EOF
+#!/bin/sh
+cp /tls/* /target/etc/ssl/certs/
+cat /tls/*.pem >> /target/etc/ssl/certs/ca-certificates.crt
+EOF
+chmod a+x /usr/lib/live-installer.d/confluent-certs
+mkdir -p /.ssh/
+cat /ssh/*pubkey > /.ssh/authorized_keys
+mkdir -p /etc/ssh
+nodename=$(grep ^NODENAME: /etc/confluent/confluent.info|cut -d ' ' -f 2)
 apikey=$(cat /etc/confluent/confluent.apikey)
+ssh-keygen -A
 for pubkey in /etc/ssh/ssh_host*key.pub; do
-    certfile=${pubkey%.pub}-cert.pub
+    certfile=$(echo $pubkey | sed -e s/.pub/-cert.pub/)
     keyfile=${pubkey%.pub}
-    wget --post-file=$pubkey --header='CONFLUENT_NODENAME: '$nodename --header="CONFLUENT_APIKEY: $apikey" https://$mgr/confluent-api/self/sshcert -O $certfile
+    wget --header="CONFLUENT_NODENAME: $nodename" --header="CONFLUENT_APIKEY: $apikey" --post-file=$pubkey https://$mgr/confluent-api/self/sshcert -O $certfile --quiet
     echo HostKey $keyfile >> /etc/ssh/sshd_config
     echo HostCertificate $certfile >> /etc/ssh/sshd_config
 done
-
-echo sshd:x:939:939::/: >> /etc/passwd
+if [ -e /tmp/installdisk ]; then
+    instdisk=$(cat /tmp/installdisk)
+else
+    for blockdev in $(ls /sys/class/block/); do
+        shortname=$(basename $blockdev)
+        if [ "$shortname" != "${shortname%loop*}" ]; then
+            continue
+        fi
+        udevadm info --query=property /dev/$shortname |grep DEVTYPE=disk > /dev/null || continue # ignore partitions
+        udevadm info --query=property /dev/$shortname |grep DM_NAME > /dev/null && continue # not a real disk
+        sz=$(cat /sys/block/$shortname/size 2> /dev/null)
+        [ -z "$sz" ] && continue
+        [ $sz -lt 1048576 ] && continue # Too small
+        [ -z "$firstdisk" ] && firstdisk=$shortname
+        if udevadm info --query=property /dev/$shortname|grep ID_MODEL=| sed -e s/' '/_/g | grep -iE '(thinksystem_m.2|m.2_nvme_2-bay_raid_kit)' > /dev/null; then
+            instdisk=$shortname
+            break
+        fi
+        if udevadm info --query=property /dev/$shortname|grep MD_CONTAINER=imsm; then
+            sraid=$sortname
+        else
+            drv=$(udevadm info -a /dev/sdb|grep DRIVERS==|grep -Ev '""|"sd"' | sed -e s/.*=// -e s/'"'//g)
+            if [ "ahci" = "$drv" -a -z "$onbdisk" ]; then
+                onbdisk=$shortname
+            elif [ "megaraid" = "$drv" -a -z "$rdisk" ]; then
+                rdisk=$shortname
+            fi
+        fi
+    done
+fi
+if [ -z "$instdisk" ]; then
+    if [ ! -z "$sraid"]; then
+        instdisk=$sraid
+    elif [ ! -z "$onbdisk" ]; then
+        instdisk=$onbdisk
+    elif [ ! -z "$rdisk" ]; then
+        instdisk=$rdisk
+    else
+        instdisk=$firstdisk
+    fi
+fi
+if [ ! -z "$instdisk" ]; then
+    debconf-set partman-auto/disk /dev/$instdisk
+    debconf-set grub-installer/bootdev /dev/$instdisk
+fi
+echo HostbasedAuthentication yes >> /etc/ssh/sshd_config
+echo HostbasedUsesNameFromPacketOnly yes >> /etc/ssh/sshd_config
+echo IgnoreRhosts no >> /etc/ssh/sshd_config
+echo sshd:x:1:1::/run/sshd:/bin/false >> /etc/passwd
 /usr/sbin/sshd
-
+wget --header="CONFLUENT_NODENAME: $nodename" --header="CONFLUENT_APIKEY: $apikey" https://$mgr/confluent-api/self/nodelist -O /tmp/allnodes --quiet
+#kill -HUP $(ps | grep -v grep | grep /usr/sbin/sshd | sed -e 's/^ *//'|cut -d ' ' -f 1)
+#curl -f https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/getinstalldisk > /tmp/getinstalldisk
+#python3 /tmp/getinstalldisk
+#sed -i s!%%INSTALLDISK%%!/dev/$(cat /tmp/installdisk)! /autoinstall.yaml

confluent_osdeploy/debian/profiles/default/scripts/prechroot.sh

@@ -0,0 +1,19 @@
+#!/bin/sh
+mount -o bind /sys /target/sys
+mount -o bind /dev /target/dev
+mount -o bind /dev/pts /target/dev/pts
+mount -o bind /proc /target/proc
+mount -o bind /dev/pts /target/dev/pts
+mount -o bind /run /target/run
+cp -a /etc/confluent /target/etc/confluent
+cp -a /opt/confluent /target/opt/confluent
+mv /tmp/post.sh /target/tmp/
+cp -a /ssh /tls /target/tmp
+cat /tls/*.pem >> /target/etc/confluent/ca.pem
+cp -a /etc/ssh/ssh_host_* /target/etc/ssh/
+grep HostCertificate /etc/ssh/sshd_config  >> /target/etc/ssh/sshd_config
+echo Port 2222 >> /etc/ssh/sshd_config
+kill -HUP $(ps |grep -v grep|grep sshd|grep /usr|sed -e s/' root.*//')
+cp /tls/* /target/etc/ssl/certs/
+cat /tls/*.pem >> /target/etc/ssl/certs/ca-certificates.crt
+chroot /target bash /tmp/post.sh

confluent_osdeploy/debian/profiles/default/scripts/proxmox/proxmoxve.firstboot

@@ -0,0 +1,5 @@
+#!/bin/bash
+# Add this to firstboot.d
+export DEBIAN_FRONTEND=noninteractive
+apt-get -y install proxmox-ve postfix open-iscsi chrony < /dev/null
+

confluent_osdeploy/debian/profiles/default/scripts/proxmox/proxmoxve.post

@@ -0,0 +1,26 @@
+#!/bin/bash
+# This script would run in post.d
+#
+export DEBIAN_FRONTEND=noninteractive
+codename=$(grep ^VERSION_CODENAME /etc/os-release | cut -d= -f2)
+echo "deb [arch=amd64] http://download.proxmox.com/debian/pve $codename pve-no-subscription" > /etc/apt/sources.list.d/pve-install-repo.list
+wget https://enterprise.proxmox.com/debian/proxmox-release-$codename.gpg -O /etc/apt/trusted.gpg.d/proxmox-release-$codename.gpg
+sum=$(sha512sum /etc/apt/trusted.gpg.d/proxmox-release-$codename.gpg)
+if [ $codename == "bookworm" ]; then
+	expectedsum=7da6fe34168adc6e479327ba517796d4702fa2f8b4f0a9833f5ea6e6b48f6507a6da403a274fe201595edc86a84463d50383d07f64bdde2e3658108db7d6dc87
+elif [ $codename == "trixie" ]; then
+	expectedsum=8678f2327c49276615288d7ca11e7d296bc8a2b96946fe565a9c81e533f9b15a5dbbad210a0ad5cd46d361ff1d3c4bac55844bc296beefa4f88b86e44e69fa51
+fi
+if [ "$sum" -ne "$expectedsum" ]; then
+	echo "Mismatch in fingerprint!"
+	rm /etc/apt/trusted.gpg.d/proxmox-release-$codename.gpg
+	exit 1
+fi
+apt-get update && apt-get -y full-upgrade < /dev/null
+
+apt-get -y install proxmox-default-kernel < /dev/null
+apt-get -y remove linux-image-amd64 'linux-image-6.1*' < /dev/null
+update-grub
+apt-get -y remove os-prober < /dev/null
+
+

confluent_osdeploy/debian/profiles/default/scripts/setupssh

@@ -0,0 +1,45 @@
+[ -f /lib/confluent/functions ] && . /lib/confluent/functions
+[ -f /etc/confluent/functions ] && . /etc/confluent/functions
+[ -f /opt/confluent/bin/apiclient ] && confapiclient=/opt/confluent/bin/apiclient
+[ -f /etc/confluent/apiclient ] && confapiclient=/etc/confluent/apiclient
+for pubkey in /etc/ssh/ssh_host*key.pub; do
+    if [ "$pubkey" = /etc/ssh/ssh_host_key.pub ]; then
+        continue
+    fi
+    certfile=${pubkey/.pub/-cert.pub}
+    confluentpython $confapiclient /confluent-api/self/sshcert $pubkey -o $certfile
+done
+if [ -d /etc/ssh/sshd_config.d/ -a ! -e /etc/ssh/sshd_config.d/90-confluent.conf ]; then
+	for cert in /etc/ssh/ssh*-cert.pub; do
+		echo HostCertificate $cert >> /etc/ssh/sshd_config.d/90-confluent.conf
+	done
+	echo HostbasedAuthentication yes >> /etc/ssh/sshd_config.d/90-confluent.conf
+	echo HostbasedUsesNameFromPacketOnly yes >> /etc/ssh/sshd_config.d/90-confluent.conf
+	echo IgnoreRhosts no >> /etc/ssh/sshd_config.d/90-confluent.conf
+fi
+
+TMPDIR=$(mktemp -d)
+cd $TMPDIR
+confluentpython $confapiclient /confluent-public/site/initramfs.tgz -o initramfs.tgz
+tar xf initramfs.tgz
+for ca in ssh/*.ca; do
+	LINE=$(cat $ca)
+        if [ -z "$LINE" ]; then continue; fi
+	cp -af /etc/ssh/ssh_known_hosts /etc/ssh/ssh_known_hosts.new
+	grep -v "$LINE" /etc/ssh/ssh_known_hosts > /etc/ssh/ssh_known_hosts.new
+	echo '@cert-authority *' $LINE >> /etc/ssh/ssh_known_hosts.new
+	mv /etc/ssh/ssh_known_hosts.new /etc/ssh/ssh_known_hosts
+done
+for pubkey in ssh/*.*pubkey; do
+	LINE=$(cat $pubkey)
+        if [ -z "$LINE" ]; then continue; fi
+	cp -af /root/.ssh/authorized_keys /root/.ssh/authorized_keys.new
+	grep -v "$LINE" /root/.ssh/authorized_keys > /root/.ssh/authorized_keys.new
+	echo "$LINE" >> /root/.ssh/authorized_keys.new
+	mv /root/.ssh/authorized_keys.new /root/.ssh/authorized_keys
+done
+confluentpython $confapiclient /confluent-api/self/nodelist | sed -e 's/^- //' > /etc/ssh/shosts.equiv
+cat /etc/ssh/shosts.equiv > /root/.shosts
+cd -
+rm -rf $TMPDIR
+systemctl try-restart sshd

confluent_osdeploy/el10-diskless/initramfs/usr/lib/dracut/hooks/cmdline/10-confluentdiskless.sh

@@ -0,0 +1,327 @@
+get_remote_apikey() {
+    while [ -z "$confluent_apikey" ]; do
+        /opt/confluent/bin/clortho $nodename $confluent_mgr > /etc/confluent/confluent.apikey
+        if grep ^SEALED: /etc/confluent/confluent.apikey > /dev/null; then
+            # we don't support remote sealed api keys anymore
+            echo > /etc/confluent/confluent.apikey
+        fi
+        confluent_apikey=$(cat /etc/confluent/confluent.apikey)
+        if [ -z "$confluent_apikey" ]; then
+            echo "Unable to acquire node api key, set deployment.apiarmed=once on node '$nodename', retrying..."
+            if [ ! -z "$autoconsdev" ]; then echo "Unable to acquire node api key, set deployment.apiarmed=once on node '$nodename', retrying..." > $autoconsdev; fi
+            sleep 10
+        elif [ -c /dev/tpmrm0 ]; then
+            tmpdir=$(mktemp -d)
+            cd $tmpdir
+            tpm2_startauthsession --session=session.ctx
+            tpm2_policypcr -Q --session=session.ctx --pcr-list="sha256:15" --policy=pcr15.sha256.policy
+            tpm2_createprimary -G ecc -Q --key-context=prim.ctx
+            (echo -n "CONFLUENT_APIKEY:";cat /etc/confluent/confluent.apikey) | tpm2_create -Q --policy=pcr15.sha256.policy --public=data.pub --private=data.priv -i - -C prim.ctx
+            tpm2_load -Q --parent-context=prim.ctx --public=data.pub --private=data.priv --name=confluent.apikey --key-context=data.ctx
+            tpm2_evictcontrol -Q -c data.ctx
+            tpm2_flushcontext session.ctx
+            cd - > /dev/null
+            rm -rf $tmpdir
+        fi
+    done
+}
+root=1
+rootok=1
+netroot=confluent
+echo -ne '\033[H\033[2J\033[3J'
+mkdir -p /etc/ssh
+mkdir -p /var/tmp/
+mkdir -p /var/empty/sshd
+mkdir -p /usr/share/empty.sshd
+mkdir -p /etc/confluent
+sed -i '/^root:x/d' /etc/passwd
+echo root:x:0:0::/:/bin/bash >> /etc/passwd
+echo sshd:x:30:30:SSH User:/var/empty/sshd:/sbin/nologin >> /etc/passwd
+
+if ! grep console= /proc/cmdline >& /dev/null; then
+    autocons=$(/opt/confluent/bin/autocons)
+    autoconsdev=${autocons%,*}
+    autocons=${autocons##*/}
+    echo "Automatic console configured for $autocons"
+fi
+echo "Initializing confluent diskless environment"
+echo -n "udevd: "
+/usr/lib/systemd/systemd-udevd --daemon
+echo -n "Loading drivers..."
+udevadm trigger
+udevadm trigger --type=devices --action=add
+udevadm settle
+modprobe ib_ipoib
+modprobe ib_umad
+modprobe hfi1
+modprobe mlx5_ib
+echo "done"
+cat > /etc/ssh/sshd_config << EOF
+Port 2222
+Subsystem       sftp    /usr/libexec/openssh/sftp-server
+PermitRootLogin yes
+AuthorizedKeysFile      .ssh/authorized_keys
+EOF
+mkdir /root/.ssh
+mkdir /.ssh
+cat /ssh/*pubkey > /root/.ssh/authorized_keys 2>/dev/null
+cp /root/.ssh/authorized_keys /.ssh/
+cat /tls/*.pem > /etc/confluent/ca.pem
+mkdir -p /etc/pki/tls/certs
+cat /tls/*.pem > /etc/pki/tls/certs/ca-bundle.crt
+TRIES=0
+oldumask=$(umask)
+umask 0077
+tpmdir=$(mktemp -d)
+cd $tpmdir
+lasthdl=""
+if [ -c /dev/tpmrm0 ]; then
+    for hdl in $(tpm2_getcap handles-persistent|awk '{print $2}'); do
+        tpm2_startauthsession --policy-session --session=session.ctx
+        tpm2_policypcr -Q --session=session.ctx --pcr-list="sha256:15" --policy=pcr15.sha256.policy
+        unsealeddata=$(tpm2_unseal --auth=session:session.ctx -Q -c $hdl 2>/dev/null)
+        tpm2_flushcontext session.ctx
+        if [[ $unsealeddata == "CONFLUENT_APIKEY:"* ]]; then
+            confluent_apikey=${unsealeddata#CONFLUENT_APIKEY:}
+            echo $confluent_apikey > /etc/confluent/confluent.apikey
+            if [ -n "$lasthdl" ]; then
+                tpm2_evictcontrol -c $lasthdl
+            fi
+            lasthdl=$hdl
+        fi
+    done
+fi
+cd - > /dev/null
+rm -rf $tpmdir
+touch /etc/confluent/confluent.info
+cd /sys/class/net
+echo -n "Scanning for network configuration..."
+while ! grep ^EXTMGRINFO: /etc/confluent/confluent.info | awk -F'|' '{print $3}' | grep 1 >& /dev/null && [ "$TRIES" -lt 30 ]; do
+    TRIES=$((TRIES + 1))
+    for i in *; do
+        ip link set $i up
+    done
+    /opt/confluent/bin/copernicus -t > /etc/confluent/confluent.info
+    echo -n .
+done
+TRIES=0
+while ! grep ^NODENAME: /etc/confluent/confluent.info >& /dev/null && [ "$TRIES" -lt 300 ]; do
+    sleep 0.5
+    echo -n .
+    /opt/confluent/bin/copernicus -t > /etc/confluent/confluent.info
+    TRIES=$((TRIES + 1))
+done
+cd /
+nodename=$(grep ^NODENAME /etc/confluent/confluent.info|awk '{print $2}')
+hostname $nodename
+confluent_mgr=$(grep '^EXTMGRINFO:.*1$' /etc/confluent/confluent.info | head -n 1 | awk -F': ' '{print $2}' | awk -F'|' '{print $1}')
+if [ -z "$confluent_mgr" ]; then
+    confluent_mgr=$(grep ^MANAGER: /etc/confluent/confluent.info|head -n 1 | awk '{print $2}')
+fi
+if [[ $confluent_mgr == *%* ]]; then
+    echo $confluent_mgr | awk -F% '{print $2}' > /tmp/confluent.ifidx
+    ifidx=$(cat /tmp/confluent.ifidx)
+    ifname=$(ip link |grep ^$ifidx:|awk '{print $2}')
+    ifname=${ifname%:}
+fi
+
+ready=0
+while [ $ready = "0" ]; do
+    get_remote_apikey
+    if [[ $confluent_mgr == *:* ]] && [[ $confluent_mgr != "["* ]]; then
+        confluent_mgr="[$confluent_mgr]"
+    fi
+    tmperr=$(mktemp)
+    curl -sSf -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $confluent_apikey" https://$confluent_mgr/confluent-api/self/deploycfg2 > /etc/confluent/confluent.deploycfg 2> $tmperr
+    if grep 401 $tmperr > /dev/null; then
+        confluent_apikey=""
+        if [ -n "$lasthdl" ]; then
+            tpm2_evictcontrol -c $lasthdl
+        fi
+        confluent_mgr=${confluent_mgr#[}
+        confluent_mgr=${confluent_mgr%]}
+    elif grep 'SSL' $tmperr > /dev/null; then
+        confluent_mgr=${confluent_mgr#[}
+        confluent_mgr=${confluent_mgr%]}
+    	echo 'Failure establishing TLS conneection to '$confluent_mgr' (try `osdeploy initialize -t` on the deployment server)'
+	if [ ! -z "$autoconsdev" ]; then echo 'Failure establishing TLS conneection to '$confluent_mgr' (try `osdeploy initialize -t` on the deployment server)' > $autoconsdev; fi
+    	sleep 10
+    else
+        ready=1
+    fi
+    rm $tmperr
+done
+if [ ! -z "$autocons" ] && grep "textconsole: true" /etc/confluent/confluent.deploycfg > /dev/null; then /opt/confluent/bin/autocons -c > /dev/null; fi
+if [ -c /dev/tpmrm0 ]; then
+    tpm2_pcrextend 15:sha256=2fbe96c50dde38ce9cd2764ddb79c216cfbcd3499568b1125450e60c45dd19f2
+fi
+umask $oldumask
+mkdir -p /run/NetworkManager/system-connections
+cat > /run/NetworkManager/system-connections/$ifname.nmconnection << EOC
+[connection]
+EOC
+echo id=${ifname} >> /run/NetworkManager/system-connections/$ifname.nmconnection
+echo uuid=$(uuidgen) >> /run/NetworkManager/system-connections/$ifname.nmconnection
+linktype=$(ip link show dev ${ifname}|grep link/|awk '{print $1}')
+if [ "$linktype" = link/infiniband ]; then
+        linktype="infiniband"
+else
+        linktype="ethernet"
+fi
+echo type=$linktype >> /run/NetworkManager/system-connections/$ifname.nmconnection
+cat >> /run/NetworkManager/system-connections/$ifname.nmconnection << EOC
+autoconnect-retries=1
+EOC
+echo interface-name=$ifname >> /run/NetworkManager/system-connections/$ifname.nmconnection
+cat >> /run/NetworkManager/system-connections/$ifname.nmconnection << EOC
+multi-connect=1
+permissions=
+wait-device-timeout=60000
+
+EOC
+if [ "$linktype" = infiniband ]; then
+cat >> /run/NetworkManager/system-connections/$ifname.nmconnection << EOC
+[infiniband]
+transport-mode=datagram
+
+EOC
+fi
+autoconfigmethod=$(grep ^ipv4_method: /etc/confluent/confluent.deploycfg |awk '{print $2}')
+auto6configmethod=$(grep ^ipv6_method: /etc/confluent/confluent.deploycfg |awk '{print $2}')
+if [ "$autoconfigmethod" = "dhcp" ]; then
+    echo -n "Attempting to use dhcp to bring up $ifname..."
+    dhcpcd $ifname
+    echo "Complete:"
+    ip addr show dev $ifname
+    confluent_mgr=$(grep ^deploy_server: /etc/confluent/confluent.deploycfg| awk '{print $2}')
+elif [ "$autoconfigmethod" = "static" ]; then
+    confluent_mgr=$(grep ^deploy_server: /etc/confluent/confluent.deploycfg| awk '{print $2}')
+    v4addr=$(grep ^ipv4_address: /etc/confluent/confluent.deploycfg)
+    v4addr=${v4addr#ipv4_address: }
+    v4gw=$(grep ^ipv4_gateway: /etc/confluent/confluent.deploycfg)
+    v4gw=${v4gw#ipv4_gateway: }
+    if [ "$v4gw" = "null" ]; then
+        v4gw=""
+    fi
+    v4nm=$(grep ^prefix: /etc/confluent/confluent.deploycfg)
+    v4nm=${v4nm#prefix: }
+    echo "Setting up $ifname as static at $v4addr/$v4nm"
+    ip addr add dev $ifname $v4addr/$v4nm
+    if [ ! -z "$v4gw" ]; then
+        ip route add default via $v4gw
+    fi
+    echo '[ipv4]' >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    echo address1=$v4addr/$v4nm >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    if [ ! -z "$v4gw" ]; then
+        echo gateway=$v4gw >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    fi
+    nameserversec=0
+    nameservers=""
+    while read -r entry; do
+        if [ $nameserversec = 1 ]; then
+            if [[ $entry == "-"*.* ]]; then
+                nameservers="$nameservers"${entry#- }";"
+                continue
+            fi
+        fi
+        nameserversec=0
+        if [ "${entry%:*}" = "nameservers" ]; then
+            nameserversec=1
+            continue
+        fi
+    done < /etc/confluent/confluent.deploycfg
+    echo dns=$nameservers >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    dnsdomain=$(grep ^dnsdomain: /etc/confluent/confluent.deploycfg)
+    dnsdomain=${dnsdomain#dnsdomain: }
+    echo dns-search=$dnsdomain >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    cat >> /run/NetworkManager/system-connections/$ifname.nmconnection << EOC
+may-fail=false
+method=manual
+
+[ipv6]
+addr-gen-mode=eui64
+method=auto
+
+EOC
+elif [ "$auto6configmethod" = "static" ]; then
+    confluent_mgr=$(grep ^deploy_server_v6: /etc/confluent/confluent.deploycfg| awk '{print $2}')
+    v6addr=$(grep ^ipv6_address: /etc/confluent/confluent.deploycfg)
+    v6addr=${v6addr#ipv6_address: }
+    v6gw=$(grep ^ipv6_gateway: /etc/confluent/confluent.deploycfg)
+    v6gw=${v6gw#ipv6_gateway: }
+    if [ "$v6gw" = "null" ]; then
+        v6gw=""
+    fi
+    v6nm=$(grep ^ipv6_prefix: /etc/confluent/confluent.deploycfg)
+    v6nm=${v6nm#ipv6_prefix: }
+    echo "Setting up $ifname as static at $v6addr/$v6nm"
+    ip addr add dev $ifname $v6addr/$v6nm
+
+    cat >> /run/NetworkManager/system-connections/$ifname.nmconnection << EOC
+[ipv4]
+dhcp-timeout=90
+dhcp-vendor-class-identifier=anaconda-Linux
+method=disabled
+
+[ipv6]
+addr-gen-mode=eui64
+method=manual
+may-fail=false
+EOC
+    echo address1=$v6addr/$v6nm >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    if [ ! -z "$v6gw" ]; then
+        ip route add default via $v6gw
+        echo gateway=$v6gw >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    fi
+    nameserversec=0
+    nameservers=""
+    while read -r entry; do
+        if [ $nameserversec = 1 ]; then
+            if [[ $entry == "-"*:* ]]; then
+                nameservers="$nameservers"${entry#- }";"
+                continue
+            fi
+        fi
+        nameserversec=0
+        if [ "${entry%:*}" = "nameservers" ]; then
+            nameserversec=1
+            continue
+        fi
+    done < /etc/confluent/confluent.deploycfg
+    echo dns=$nameservers >> /run/NetworkManager/system-connections/$ifname.nmconnection
+    dnsdomain=$(grep ^dnsdomain: /etc/confluent/confluent.deploycfg)
+    dnsdomain=${dnsdomain#dnsdomain: }
+    echo dns-search=$dnsdomain >> /run/NetworkManager/system-connections/$ifname.nmconnection
+fi
+echo '[proxy]' >> /run/NetworkManager/system-connections/$ifname.nmconnection
+chmod 600 /run/NetworkManager/system-connections/*.nmconnection
+confluent_websrv=$confluent_mgr
+if [[ $confluent_websrv == *:* ]] && [[ $confluent_websrv != "["* ]]; then
+    confluent_websrv="[$confluent_websrv]"
+fi
+echo -n "Initializing ssh..."
+ssh-keygen -t ed25519 -f /etc/ssh/ssh_host_ed25519_key -C '' -N ''
+for pubkey in /etc/ssh/ssh_host*key.pub; do
+    certfile=${pubkey/.pub/-cert.pub}
+    privfile=${pubkey%.pub}
+    curl -sf -X POST -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $confluent_apikey" -d @$pubkey  https://$confluent_websrv/confluent-api/self/sshcert > $certfile
+    if [ -s $certfile ]; then
+        echo HostCertificate $certfile >> /etc/ssh/sshd_config
+    fi
+    echo HostKey $privfile >> /etc/ssh/sshd_config
+done
+/usr/sbin/sshd
+confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg| awk '{print $2}')
+confluent_proto=$(grep ^protocol: /etc/confluent/confluent.deploycfg| awk '{print $2}')
+confluent_urls=""
+for addr in $(grep ^MANAGER: /etc/confluent/confluent.info|awk '{print $2}'|sed -e s/%/%25/); do
+    if [[ $addr == *:* ]]; then
+        confluent_urls="$confluent_urls $confluent_proto://[$addr]/confluent-public/os/$confluent_profile/rootimg.sfs"
+    else
+        confluent_urls="$confluent_urls $confluent_proto://$addr/confluent-public/os/$confluent_profile/rootimg.sfs"
+    fi
+done
+mkdir -p /etc/confluent
+curl -sf https://$confluent_websrv/confluent-public/os/$confluent_profile/scripts/functions > /etc/confluent/functions
+. /etc/confluent/functions
+source_remote imageboot.sh

confluent_osdeploy/el10-diskless/profiles/default/scripts/add_local_repositories

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/add_local_repositories

confluent_osdeploy/el10-diskless/profiles/default/scripts/firstboot.custom

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/firstboot.custom

confluent_osdeploy/el10-diskless/profiles/default/scripts/firstboot.service

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/firstboot.service

confluent_osdeploy/el10-diskless/profiles/default/scripts/firstboot.sh

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/firstboot.sh

confluent_osdeploy/el10-diskless/profiles/default/scripts/functions

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/functions

confluent_osdeploy/el10-diskless/profiles/default/scripts/getinstalldisk

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/getinstalldisk

confluent_osdeploy/el10-diskless/profiles/default/scripts/image2disk.py

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/image2disk.py

confluent_osdeploy/el10-diskless/profiles/default/scripts/imageboot.sh

@@ -0,0 +1,138 @@
+. /lib/dracut-lib.sh
+confluent_whost=$confluent_mgr
+if [[ "$confluent_whost" == *:* ]] && [[ "$confluent_whost" != "["* ]]; then
+    confluent_whost="[$confluent_mgr]"
+fi
+mkdir -p /mnt/remoteimg /mnt/remote /mnt/overlay
+if [ "untethered" = "$(getarg confluent_imagemethod)" ]; then
+    mount -t tmpfs untethered /mnt/remoteimg
+    curl https://$confluent_whost/confluent-public/os/$confluent_profile/rootimg.sfs -o /mnt/remoteimg/rootimg.sfs
+else
+    confluent_urls="$confluent_urls https://$confluent_whost/confluent-public/os/$confluent_profile/rootimg.sfs"
+    /opt/confluent/bin/urlmount $confluent_urls /mnt/remoteimg
+fi
+/opt/confluent/bin/confluent_imginfo /mnt/remoteimg/rootimg.sfs > /tmp/rootimg.info
+loopdev=$(losetup -f)
+export mountsrc=$loopdev
+losetup -r $loopdev /mnt/remoteimg/rootimg.sfs
+if grep '^Format: confluent_crypted' /tmp/rootimg.info > /dev/null; then
+    while ! curl -sf -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $(cat /etc/confluent/confluent.apikey)" https://$confluent_whost/confluent-api/self/profileprivate/pending/rootimg.key > /tmp/rootimg.key; do
+        echo "Unable to retrieve private key from $confluent_mgr (verify that confluent can access /var/lib/confluent/private/os/$confluent_profile/pending/rootimg.key)"
+        sleep 1
+    done
+    cipher=$(head -n 1 /tmp/rootimg.key)
+    key=$(tail -n 1 /tmp/rootimg.key)
+    len=$(wc -c /mnt/remoteimg/rootimg.sfs | awk '{print $1}')
+    len=$(((len-4096)/512))
+    dmsetup create cryptimg --table "0 $len crypt $cipher $key 0 $loopdev 8"
+    /opt/confluent/bin/confluent_imginfo /dev/mapper/cryptimg > /tmp/rootimg.info
+    mountsrc=/dev/mapper/cryptimg
+fi
+
+if grep '^Format: squashfs' /tmp/rootimg.info > /dev/null; then
+    mount -o ro $mountsrc /mnt/remote
+elif grep  '^Format: confluent_multisquash' /tmp/rootimg.info; then
+    tail -n +3 /tmp/rootimg.info  | awk '{gsub("/", "_"); print "echo 0 " $4 " linear '$mountsrc' " $3 " | dmsetup create mproot" $7}' > /tmp/setupmount.sh
+    . /tmp/setupmount.sh
+    cat /tmp/setupmount.sh |awk '{printf "mount /dev/mapper/"$NF" "; sub("mproot", ""); gsub("_", "/"); print "/mnt/remote"$NF}' > /tmp/mountparts.sh
+    . /tmp/mountparts.sh
+fi
+
+
+#mount -t tmpfs overlay /mnt/overlay
+modprobe zram
+memtot=$(grep ^MemTotal: /proc/meminfo|awk '{print $2}')
+memtot=$((memtot/2))$(grep ^MemTotal: /proc/meminfo | awk '{print $3'})
+echo $memtot > /sys/block/zram0/disksize
+mkfs.xfs /dev/zram0 > /dev/null
+mount -o discard /dev/zram0 /mnt/overlay
+if [ ! -f /tmp/mountparts.sh ]; then
+    mkdir -p /mnt/overlay/upper /mnt/overlay/work
+    mount -t overlay -o upperdir=/mnt/overlay/upper,workdir=/mnt/overlay/work,lowerdir=/mnt/remote disklessroot /sysroot
+else
+    for srcmount in $(cat /tmp/mountparts.sh | awk '{print $3}'); do
+        mkdir -p /mnt/overlay${srcmount}/upper /mnt/overlay${srcmount}/work
+        mount -t overlay -o upperdir=/mnt/overlay${srcmount}/upper,workdir=/mnt/overlay${srcmount}/work,lowerdir=${srcmount} disklesspart /sysroot${srcmount#/mnt/remote}
+    done
+fi
+mkdir -p /sysroot/etc/ssh
+mkdir -p /sysroot/etc/confluent
+mkdir -p /sysroot/root/.ssh
+cp /root/.ssh/* /sysroot/root/.ssh
+chmod 700 /sysroot/root/.ssh
+cp /etc/confluent/* /sysroot/etc/confluent/
+cp /etc/ssh/*key* /sysroot/etc/ssh/
+for pubkey in /etc/ssh/ssh_host*key.pub; do
+    certfile=${pubkey/.pub/-cert.pub}
+    privfile=${pubkey%.pub}
+    if [ -s $certfile ]; then
+        echo HostCertificate $certfile >> /sysroot/etc/ssh/sshd_config
+    fi
+    echo HostKey $privfile >> /sysroot/etc/ssh/sshd_config
+done
+
+mkdir -p /sysroot/dev /sysroot/sys /sysroot/proc /sysroot/run
+if [ ! -z "$autocons" ]; then
+    autocons=${autocons%,*}
+    mkdir -p /run/systemd/generator/getty.target.wants
+    ln -s /usr/lib/systemd/system/serial-getty@.service /run/systemd/generator/getty.target.wants/serial-getty@${autocons}.service
+fi
+while [ ! -e /sysroot/sbin/init ]; do
+    echo "Failed to access root filesystem or it is missing /sbin/init"
+    echo "System should be accessible through ssh at port 2222 with the appropriate key"
+    while [ ! -e /sysroot/sbin/init ]; do
+        sleep 1
+    done
+done
+rootpassword=$(grep ^rootpassword: /etc/confluent/confluent.deploycfg)
+rootpassword=${rootpassword#rootpassword: }
+if [ "$rootpassword" = "null" ]; then
+    rootpassword=""
+fi
+
+if [ ! -z "$rootpassword" ]; then
+    sed -i "s@root:[^:]*:@root:$rootpassword:@" /sysroot/etc/shadow
+fi
+for i in /ssh/*.ca; do
+    echo '@cert-authority *' $(cat $i) >> /sysroot/etc/ssh/ssh_known_hosts
+done
+echo HostbasedAuthentication yes >> /sysroot/etc/ssh/sshd_config
+echo HostbasedUsesNameFromPacketOnly yes >> /sysroot/etc/ssh/sshd_config
+echo IgnoreRhosts no >> /sysroot/etc/ssh/sshd_config
+sshconf=/sysroot/etc/ssh/ssh_config
+if [ -d /sysroot/etc/ssh/ssh_config.d/ ]; then
+    sshconf=/sysroot/etc/ssh/ssh_config.d/01-confluent.conf
+fi
+echo 'Host *' >> $sshconf
+echo '    HostbasedAuthentication yes' >> $sshconf
+echo '    EnableSSHKeysign yes' >> $sshconf
+echo '    HostbasedKeyTypes *ed25519*' >> $sshconf
+curl -sf -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $(cat /etc/confluent/confluent.apikey)" https://$confluent_whost/confluent-api/self/nodelist > /sysroot/etc/ssh/shosts.equiv
+cp /sysroot/etc/ssh/shosts.equiv /sysroot/root/.shosts
+chmod 640 /sysroot/etc/ssh/*_key
+cp /tls/*.pem /sysroot/etc/pki/ca-trust/source/anchors/
+chroot /sysroot/ update-ca-trust
+curl -sf https://$confluent_whost/confluent-public/os/$confluent_profile/scripts/onboot.service > /sysroot/etc/systemd/system/onboot.service
+mkdir -p /sysroot/opt/confluent/bin
+curl -sf https://$confluent_whost/confluent-public/os/$confluent_profile/scripts/onboot.sh > /sysroot/opt/confluent/bin/onboot.sh
+chmod +x /sysroot/opt/confluent/bin/onboot.sh
+cp /opt/confluent/bin/apiclient /sysroot/opt/confluent/bin
+ln -s /etc/systemd/system/onboot.service /sysroot/etc/systemd/system/multi-user.target.wants/onboot.service
+cp /etc/confluent/functions /sysroot/etc/confluent/functions
+if grep installtodisk /proc/cmdline > /dev/null; then
+    . /etc/confluent/functions
+    run_remote installimage
+    exec reboot -f
+fi
+mv /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs
+ln -s /sysroot/lib/modules/$(uname -r) /lib/modules/
+mv /lib/firmware /lib/firmware-ramfs
+ln -s /sysroot/lib/firmware /lib/firmware
+kill $(grep -l ^/usr/lib/systemd/systemd-udevd  /proc/*/cmdline|cut -d/ -f 3)
+if grep debugssh /proc/cmdline >& /dev/null; then
+    exec /opt/confluent/bin/start_root
+else
+    rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
+    exec /opt/confluent/bin/start_root -s  # share mount namespace, keep kernel callbacks intact
+fi
+

confluent_osdeploy/el10-diskless/profiles/default/scripts/installimage

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/installimage

confluent_osdeploy/el10-diskless/profiles/default/scripts/onboot.custom

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/onboot.custom

confluent_osdeploy/el10-diskless/profiles/default/scripts/onboot.service

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/onboot.service

confluent_osdeploy/el10-diskless/profiles/default/scripts/onboot.sh

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/onboot.sh

confluent_osdeploy/el10-diskless/profiles/default/scripts/post.sh

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/post.sh

confluent_osdeploy/el10-diskless/profiles/default/scripts/syncfileclient

@@ -0,0 +1 @@
+../../../../el9-diskless/profiles/default/scripts/syncfileclient

confluent_osdeploy/el7-diskless/profiles/default/scripts/functions

@@ -10,6 +10,13 @@ function test_mgr() {
     return 1
 }
 
+function initconfluentscriptstmp() {
+    if [ -z "$confluentscripttmpdir" ]; then
+        mkdir -p /opt/confluent/tmpexec
+        confluentscripttmpdir=$(mktemp -d /opt/confluent/tmpexec/confluentscripts.XXXXXXXXX)
+    fi
+}
+
 function confluentpython() {
     if [ -x /usr/libexec/platform-python ]; then
         /usr/libexec/platform-python $*
@@ -72,7 +79,8 @@ fetch_remote() {
 }
 
 source_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -86,7 +94,8 @@ source_remote_parts() {
 }
 
 run_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -105,10 +114,7 @@ source_remote() {
     echo
     echo '---------------------------------------------------------------------------'
     echo Sourcing $1 from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
-        unsettmpdir=1
-    fi
+    initconfluentscriptstmp
     echo Sourcing from $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -135,9 +141,9 @@ run_remote() {
     echo '---------------------------------------------------------------------------'
     echo Running $requestedcmd from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
     if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
         unsettmpdir=1
     fi
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -170,7 +176,8 @@ run_remote_python() {
     fi
     echo '---------------------------------------------------------------------------'
     echo Running python script "'$*'" from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     mkdir -p $(dirname $1)

confluent_osdeploy/el7-diskless/profiles/default/scripts/getinstalldisk

@@ -2,6 +2,9 @@
 import subprocess
 import os
 
+class SilentException(Exception):
+    pass
+
 class DiskInfo(object):
     def __init__(self, devname):
         if devname.startswith('nvme') and 'c' in devname:
@@ -24,9 +27,11 @@ class DiskInfo(object):
                 continue
             k, v = prop.split('=', 1)
             if k == 'DEVTYPE' and v != 'disk':
+                if v == 'partition':
+                    raise SilentException('Partition')
                 raise Exception('Not a disk')
             elif k == 'DM_NAME':
-                raise Exception('Device Mapper')
+                raise SilentException('Device Mapper')
             elif k == 'ID_MODEL':
                 self.model = v
             elif k == 'DEVPATH':
@@ -50,13 +55,17 @@ class DiskInfo(object):
                 self.driver = v.replace('"', '')
             elif k == 'ATTRS{subsystype}':
                 self.subsystype = v.replace('"', '')
+            elif k == 'ATTR{ro}' and v == '"1"':
+                raise Exception("Device is read-only")
         if not self.driver and 'imsm' not in self.mdcontainer and self.subsystype != 'nvm':
             raise Exception("No driver detected")
+        if self.driver == 'sr':
+            raise Exception('cd/dvd')
         if os.path.exists('/sys/block/{0}/size'.format(self.name)):
             with open('/sys/block/{0}/size'.format(self.name), 'r') as sizesrc:
                 self.size = int(sizesrc.read()) * 512
-        if int(self.size) < 536870912:
-            raise Exception("Device too small for install")
+        if int(self.size) < 2147483648:
+            raise Exception("Device too small for install ({}MiB)".format(int(self.size)/1024/1024))
 
     @property
     def priority(self):
@@ -89,9 +98,11 @@ def main():
         try:
             disk = DiskInfo(disk)
             disks.append(disk)
+        except SilentException:
+            pass
         except Exception as e:
             print("Skipping {0}: {1}".format(disk, str(e)))
-    nd = [x.name for x in sorted(disks, key=lambda x: x.priority)]
+    nd = [x.name for x in sorted(disks, key=lambda x: [x.priority, x.size])]
     if nd:
         open('/tmp/installdisk', 'w').write(nd[0])
 

confluent_osdeploy/el7-diskless/profiles/default/scripts/onboot.sh

@@ -5,11 +5,37 @@
 # noted below so custom commands are executed before
 # the script notifies confluent that install is fully complete.
 
+
+ntpsrvs=""
 nodename=$(grep ^NODENAME /etc/confluent/confluent.info|awk '{print $2}')
 confluent_apikey=$(cat /etc/confluent/confluent.apikey)
 confluent_mgr=$(grep ^deploy_server: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 timedatectl set-timezone $(grep ^timezone: /etc/confluent/confluent.deploycfg|awk '{print $2}')
+
+
+if grep ^ntpservers: /etc/confluent/confluent.deploycfg > /dev/null; then
+    for ntpsrv in $(sed -n '/^ntpservers:/,/^[^-]/p' /etc/confluent/confluent.deploycfg|sed 1d|sed '$d' | sed -e 's/^- //'); do
+        echo "server ${ntpsrv} iburst " >> /tmp/timeservers
+    done
+fi
+
+if [ -f /tmp/timeservers ]; then
+
+ntpsrvs=$(cat /tmp/timeservers)
+
+sed -i "1,/^pool * /c\\
+
+${ntpsrvs//$'\n'/\\$'\n'}" /etc/chrony.conf
+
+
+systemctl restart chronyd
+
+rm -f /tmp/timeservers
+fi
+
+
+
 export nodename confluent_mgr confluent_profile
 . /etc/confluent/functions
 mkdir -p /var/log/confluent

confluent_osdeploy/el7/profiles/default/scripts/functions

@@ -10,6 +10,13 @@ function test_mgr() {
     return 1
 }
 
+function initconfluentscriptstmp() {
+    if [ -z "$confluentscripttmpdir" ]; then
+        mkdir -p /opt/confluent/tmpexec
+        confluentscripttmpdir=$(mktemp -d /opt/confluent/tmpexec/confluentscripts.XXXXXXXXX)
+    fi
+}
+
 function confluentpython() {
     if [ -x /usr/libexec/platform-python ]; then
         /usr/libexec/platform-python $*
@@ -72,7 +79,8 @@ fetch_remote() {
 }
 
 source_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -86,7 +94,8 @@ source_remote_parts() {
 }
 
 run_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -105,10 +114,7 @@ source_remote() {
     echo
     echo '---------------------------------------------------------------------------'
     echo Sourcing $1 from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
-        unsettmpdir=1
-    fi
+    initconfluentscriptstmp
     echo Sourcing from $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -135,9 +141,9 @@ run_remote() {
     echo '---------------------------------------------------------------------------'
     echo Running $requestedcmd from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
     if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
         unsettmpdir=1
     fi
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -170,7 +176,8 @@ run_remote_python() {
     fi
     echo '---------------------------------------------------------------------------'
     echo Running python script "'$*'" from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     mkdir -p $(dirname $1)

confluent_osdeploy/el7/profiles/default/scripts/getinstalldisk

@@ -2,6 +2,9 @@
 import subprocess
 import os
 
+class SilentException(Exception):
+    pass
+
 class DiskInfo(object):
     def __init__(self, devname):
         if devname.startswith('nvme') and 'c' in devname:
@@ -24,9 +27,11 @@ class DiskInfo(object):
                 continue
             k, v = prop.split('=', 1)
             if k == 'DEVTYPE' and v != 'disk':
+                if v == 'partition':
+                    raise SilentException('Partition')
                 raise Exception('Not a disk')
             elif k == 'DM_NAME':
-                raise Exception('Device Mapper')
+                raise SilentException('Device Mapper')
             elif k == 'ID_MODEL':
                 self.model = v
             elif k == 'DEVPATH':
@@ -50,13 +55,17 @@ class DiskInfo(object):
                 self.driver = v.replace('"', '')
             elif k == 'ATTRS{subsystype}':
                 self.subsystype = v.replace('"', '')
+            elif k == 'ATTR{ro}' and v == '"1"':
+                raise Exception("Device is read-only")
         if not self.driver and 'imsm' not in self.mdcontainer and self.subsystype != 'nvm':
             raise Exception("No driver detected")
+        if self.driver == 'sr':
+            raise Exception('cd/dvd')
         if os.path.exists('/sys/block/{0}/size'.format(self.name)):
             with open('/sys/block/{0}/size'.format(self.name), 'r') as sizesrc:
                 self.size = int(sizesrc.read()) * 512
-        if int(self.size) < 536870912:
-            raise Exception("Device too small for install")
+        if int(self.size) < 2147483648:
+            raise Exception("Device too small for install ({}MiB)".format(int(self.size)/1024/1024))
 
     @property
     def priority(self):
@@ -89,9 +98,11 @@ def main():
         try:
             disk = DiskInfo(disk)
             disks.append(disk)
+        except SilentException:
+            pass
         except Exception as e:
             print("Skipping {0}: {1}".format(disk, str(e)))
-    nd = [x.name for x in sorted(disks, key=lambda x: x.priority)]
+    nd = [x.name for x in sorted(disks, key=lambda x: [x.priority, x.size])]
     if nd:
         open('/tmp/installdisk', 'w').write(nd[0])
 

confluent_osdeploy/el8-diskless/initramfs/usr/lib/dracut/hooks/cmdline/10-confluentdiskless.sh

@@ -107,7 +107,11 @@ if [ ! -z "$confluentsrv" ]; then
         /usr/libexec/nm-initrd-generator ip=:dhcp6
     else
         confluenthttpsrv=$confluentsrv
-        ifname=$(ip -br link|grep LOWER_UP|grep -v UNKNOWN|head -n 1|awk '{print $1}')
+        ifname=""
+        while [ -z "$ifname" ]; do
+            ifname=$(ip -br link|grep LOWER_UP|grep -v ib|grep -v UNKNOWN|head -n 1|awk '{print $1}')
+            sleep 0.5
+        done
         echo -n "Attempting to use dhcp to bring up $ifname..."
         dhclient $ifname
         while ! ip -br addr show dev $ifname | grep \\. > /dev/null; do

confluent_osdeploy/el8-diskless/profiles/default/scripts/firstboot.sh

@@ -25,7 +25,8 @@ if [ ! -f /etc/confluent/firstboot.ran ]; then
     touch /etc/confluent/firstboot.ran
 
     cat /etc/confluent/tls/*.pem >> /etc/pki/tls/certs/ca-bundle.crt
-
+    confluentpython /root/confignet
+    rm /root/confignet
     run_remote firstboot.custom
     # Firstboot scripts may be placed into firstboot.d, e.g. firstboot.d/01-firstaction.sh, firstboot.d/02-secondaction.sh
     run_remote_parts firstboot.d

confluent_osdeploy/el8-diskless/profiles/default/scripts/functions

@@ -10,6 +10,13 @@ function test_mgr() {
     return 1
 }
 
+function initconfluentscriptstmp() {
+    if [ -z "$confluentscripttmpdir" ]; then
+        mkdir -p /opt/confluent/tmpexec
+        confluentscripttmpdir=$(mktemp -d /opt/confluent/tmpexec/confluentscripts.XXXXXXXXX)
+    fi
+}
+
 function confluentpython() {
     if [ -x /usr/libexec/platform-python ]; then
         /usr/libexec/platform-python $*
@@ -72,7 +79,8 @@ fetch_remote() {
 }
 
 source_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -86,7 +94,8 @@ source_remote_parts() {
 }
 
 run_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -105,10 +114,7 @@ source_remote() {
     echo
     echo '---------------------------------------------------------------------------'
     echo Sourcing $1 from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
-        unsettmpdir=1
-    fi
+    initconfluentscriptstmp
     echo Sourcing from $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -135,9 +141,9 @@ run_remote() {
     echo '---------------------------------------------------------------------------'
     echo Running $requestedcmd from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
     if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
         unsettmpdir=1
     fi
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -170,7 +176,8 @@ run_remote_python() {
     fi
     echo '---------------------------------------------------------------------------'
     echo Running python script "'$*'" from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     mkdir -p $(dirname $1)

confluent_osdeploy/el8-diskless/profiles/default/scripts/getinstalldisk

@@ -2,6 +2,9 @@
 import subprocess
 import os
 
+class SilentException(Exception):
+    pass
+
 class DiskInfo(object):
     def __init__(self, devname):
         if devname.startswith('nvme') and 'c' in devname:
@@ -24,9 +27,11 @@ class DiskInfo(object):
                 continue
             k, v = prop.split('=', 1)
             if k == 'DEVTYPE' and v != 'disk':
+                if v == 'partition':
+                    raise SilentException('Partition')
                 raise Exception('Not a disk')
             elif k == 'DM_NAME':
-                raise Exception('Device Mapper')
+                raise SilentException('Device Mapper')
             elif k == 'ID_MODEL':
                 self.model = v
             elif k == 'DEVPATH':
@@ -50,13 +55,17 @@ class DiskInfo(object):
                 self.driver = v.replace('"', '')
             elif k == 'ATTRS{subsystype}':
                 self.subsystype = v.replace('"', '')
+            elif k == 'ATTR{ro}' and v == '"1"':
+                raise Exception("Device is read-only")
         if not self.driver and 'imsm' not in self.mdcontainer and self.subsystype != 'nvm':
             raise Exception("No driver detected")
+        if self.driver == 'sr':
+            raise Exception('cd/dvd')
         if os.path.exists('/sys/block/{0}/size'.format(self.name)):
             with open('/sys/block/{0}/size'.format(self.name), 'r') as sizesrc:
                 self.size = int(sizesrc.read()) * 512
-        if int(self.size) < 536870912:
-            raise Exception("Device too small for install")
+        if int(self.size) < 2147483648:
+            raise Exception("Device too small for install ({}MiB)".format(int(self.size)/1024/1024))
 
     @property
     def priority(self):
@@ -89,9 +98,11 @@ def main():
         try:
             disk = DiskInfo(disk)
             disks.append(disk)
+        except SilentException:
+            pass
         except Exception as e:
             print("Skipping {0}: {1}".format(disk, str(e)))
-    nd = [x.name for x in sorted(disks, key=lambda x: x.priority)]
+    nd = [x.name for x in sorted(disks, key=lambda x: [x.priority, x.size])]
     if nd:
         open('/tmp/installdisk', 'w').write(nd[0])
 

confluent_osdeploy/el8-diskless/profiles/default/scripts/image2disk.py

@@ -56,7 +56,11 @@ def get_image_metadata(imgpath):
             for md in get_multipart_image_meta(img):
                 yield md
         else:
-            raise Exception('Installation from single part image not supported')
+            yield {'mount': '/', 'filesystem': 'xfs', 'minsize': 4294967296, 'initsize': 954128662528, 'flags': 'rw,seclabel,relatime,attr2,inode64,logbufs=8,logbsize=32k,noquota', 'device': '/dev/mapper/root', 'compressed_size': 27022069760}
+            yield {'mount': '/boot', 'filesystem': 'xfs', 'minsize': 536870912, 'initsize': 1006632960, 'flags': 'rw,seclabel,relatime,attr2,inode64,logbufs=8,logbsize=32k,noquota', 'device': '/dev/nvme1n1p2', 'compressed_size': 171462656}
+            yield {'mount': '/boot/efi', 'filesystem': 'vfat', 'minsize': 33554432, 'initsize': 627900416, 'flags': 'rw,relatime,fmask=0077,dmask=0077,codepage=437,iocharset=ascii,shortname=winnt,errors=remount-ro', 'device': '/dev/nvme1n1p1', 'compressed_size': 1576960}            
+            #raise Exception('Installation from single part image not supported')            
+
 
 class PartedRunner():
     def __init__(self, disk):
@@ -75,8 +79,17 @@ def fixup(rootdir, vols):
     for vol in vols:
         devbymount[vol['mount']] = vol['targetdisk']
     fstabfile = os.path.join(rootdir, 'etc/fstab')
-    with open(fstabfile) as tfile:
-        fstab = tfile.read().split('\n')
+    if os.path.exists(fstabfile):
+        with open(fstabfile) as tfile:
+            fstab = tfile.read().split('\n')
+    else:
+        # fabricate a reference fstab
+        fstab = [
+            "#ORIGFSTAB#/dev/mapper/root# /                       xfs     defaults        0 0",
+            "#ORIGFSTAB#UUID=aaf9e0f9-aa4d-4d74-9e75-3537620cfe23# /boot                   xfs     defaults        0 0",
+            "#ORIGFSTAB#UUID=C21D-B881#          /boot/efi               vfat    umask=0077,shortname=winnt 0 2",
+            "#ORIGFSTAB#/dev/mapper/swap# none                    swap    defaults        0 0",
+        ]       
     while not fstab[0]:
         fstab = fstab[1:]
     if os.path.exists(os.path.join(rootdir, '.autorelabel')):
@@ -126,8 +139,10 @@ def fixup(rootdir, vols):
         newcfg = ifcfg.split('/')[-1]
         newcfg = os.path.join(rootdir, 'etc/NetworkManager/system-connections/{0}'.format(newcfg))
         shutil.copy2(ifcfg, newcfg)
-    shutil.rmtree(os.path.join(rootdir, 'etc/confluent/'))
-    shutil.copytree('/etc/confluent', os.path.join(rootdir, 'etc/confluent'))
+    rootconfluentdir = os.path.join(rootdir, 'etc/confluent/')
+    if os.path.exists(rootconfluentdir):
+        shutil.rmtree(rootconfluentdir)
+    shutil.copytree('/etc/confluent', rootconfluentdir)
     if policy:
         sys.stdout.write('Applying SELinux labeling...')
         sys.stdout.flush()
@@ -142,14 +157,41 @@ def fixup(rootdir, vols):
     grubsyscfg = os.path.join(rootdir, 'etc/sysconfig/grub')
     if not os.path.exists(grubsyscfg):
         grubsyscfg = os.path.join(rootdir, 'etc/default/grub')
-    with open(grubsyscfg) as defgrubin:
-        defgrub = defgrubin.read().split('\n')
+    currcmdline = []
+    with open('/proc/cmdline') as cmdlinein:
+        cmdline = cmdlinein.read().strip()
+        for arg in cmdline.split():
+            if arg.startswith('console='):
+                currcmdline.append(arg)
+            elif arg == 'quiet':
+                currcmdline.append(arg)
+    currcmdlinestr = ' '.join(currcmdline)
+    if os.path.exists(grubsyscfg):
+        with open(grubsyscfg) as defgrubin:
+            defgrub = defgrubin.read().split('\n')
+    else:
+        defgrub = [
+        'GRUB_TIMEOUT=5',
+        'GRUB_DISTRIBUTOR="$(sed ' + "'s, release .*$,,g'" + ' /etc/system-release)"',
+        'GRUB_DEFAULT=saved',
+        'GRUB_DISABLE_SUBMENU=true',
+        'GRUB_TERMINAL=""',
+        'GRUB_SERIAL_COMMAND=""',
+        'GRUB_CMDLINE_LINUX="{} crashkernel=1G-4G:192M,4G-64G:256M,64G-:512M rd.lvm.lv=vg/root rd.lvm.lv=vg/swap"'.format(currcmdlinestr),
+        'GRUB_DISABLE_RECOVERY="true"',
+        'GRUB_ENABLE_BLSCFG=true',
+        ]
+    if not os.path.exists(os.path.join(rootdir, "etc/kernel/cmdline")):
+        with open(os.path.join(rootdir, "etc/kernel/cmdline"), "w") as cmdlineout:
+            cmdlineout.write("{} root=/dev/mapper/localstorage-root rd.lvm.lv=localstorage/root".format(currcmdlinestr))
     with open(grubsyscfg, 'w') as defgrubout:
         for gline in defgrub:
             gline = gline.split()
             newline = []
             for ent in gline:
                 if ent.startswith('resume=') or ent.startswith('rd.lvm.lv'):
+                    if ent.endswith('"'):
+                        newline.append('"')
                     continue
                 newline.append(ent)
             defgrubout.write(' '.join(newline) + '\n')
@@ -159,6 +201,12 @@ def fixup(rootdir, vols):
         grubcfg = grubcfg[:-1]
     if len(grubcfg) == 1:
         grubcfg = grubcfg[0]
+    elif not grubcfg:
+        grubcfg = '/boot/grub2/grub.cfg'
+        paths = glob.glob(os.path.join(rootdir, 'boot/efi/EFI/*'))
+        for path in paths:
+            with open(os.path.join(path, 'grub.cfg'), 'w') as stubgrubout:
+                stubgrubout.write("search --no-floppy --root-dev-only --fs-uuid --set=dev " + bootuuid + "\nset prefix=($dev)/grub2\nexport $prefix\nconfigfile $prefix/grub.cfg\n")
     else:
         for gcfg in grubcfg:
             rgcfg = os.path.join(rootdir, gcfg[1:])  # gcfg has a leading / to get rid of
@@ -214,10 +262,18 @@ def fixup(rootdir, vols):
         shimpath = subprocess.check_output(['find', os.path.join(rootdir, 'boot/efi'), '-name', 'shimx64.efi']).decode('utf8').strip()
         shimpath = shimpath.replace(rootdir, '/').replace('/boot/efi', '').replace('//', '/').replace('/', '\\')
         subprocess.check_call(['efibootmgr', '-c', '-d', targblock, '-l', shimpath, '--part', partnum])
+    try:
+        os.makedirs(os.path.join(rootdir, 'opt/confluent/bin'))
+    except Exception:
+        pass
+    shutil.copy2('/opt/confluent/bin/apiclient', os.path.join(rootdir, 'opt/confluent/bin/apiclient'))
     #other network interfaces
 
 
 def had_swap():
+    if not os.path.exists('/etc/fstab'):
+        # diskless source, assume swap
+        return True
     with open('/etc/fstab') as tabfile:
         tabs = tabfile.read().split('\n')
         for tab in tabs:
@@ -362,6 +418,8 @@ def install_to_disk(imgpath):
             subprocess.check_call(['mount', vol['targetdisk'], '/run/imginst/targ'])
             source = vol['mount'].replace('/', '_')
             source = '/run/imginst/sources/' + source
+            if not os.path.exists(source):
+                source = '/run/imginst/sources/_' + vol['mount']            
             blankfsstat = os.statvfs('/run/imginst/targ')
             blankused = (blankfsstat.f_blocks - blankfsstat.f_bfree) * blankfsstat.f_bsize
             sys.stdout.write('\nWriting {0}: '.format(vol['mount']))
@@ -419,8 +477,14 @@ def install_to_disk(imgpath):
 
 
 
-
-            subprocess.check_call(['umount', '/run/imginst/targ'])
+            while True:
+                try:
+                    subprocess.check_call(['umount', '/run/imginst/targ'])
+                except subprocess.CalledProcessError:
+                    print("Failed to unmount /run/imginst/targ, retrying")
+                    time.sleep(1)
+                else:
+                    break
         for vol in allvols:
             subprocess.check_call(['mount', vol['targetdisk'], '/run/imginst/targ/' + vol['mount']])
         fixup('/run/imginst/targ', allvols)

confluent_osdeploy/el8-diskless/profiles/default/scripts/installimage

@@ -5,6 +5,7 @@
 # and existing mounts of image (to take advantage of caching)
 mount -o bind /sys /sysroot/sys
 mount -o bind /dev /sysroot/dev
+mount -o bind /dev/pts /sysroot/dev/pts
 mount -o bind /proc /sysroot/proc
 mount -o bind /run /sysroot/run
 
@@ -21,8 +22,14 @@ else
     done
 fi
 cd /sysroot/run
+cp /run/sshd.pid /tmp/dbgssh.pid
+chroot /sysroot/ bash -c "/usr/sbin/sshd"
 chroot /sysroot/ bash -c "source /etc/confluent/functions; run_remote_python getinstalldisk"
 chroot /sysroot/ bash -c "source /etc/confluent/functions; run_remote_parts pre.d"
+for nameserver in $(sed -n '/^nameservers:/,/^[^-]/p' /etc/confluent/confluent.deploycfg|grep ^- | cut -d ' ' -f 2|sed -e 's/ //'); do
+    echo "nameserver $nameserver" >> /sysroot/etc/resolv.conf
+done
+#chroot /sysroot/ bash -c "source /etc/confluent/functions; run_remote_python confignet"
 if [ ! -f /sysroot/tmp/installdisk ]; then
     echo 'Unable to find a suitable installation target device, ssh to port 2222 to investigate'
     while [ ! -f /sysroot/tmp/installdisk ]; do
@@ -39,7 +46,10 @@ chroot /sysroot bash -c "source /etc/confluent/functions; run_remote_python imag
 echo "Port 22" >> /etc/ssh/sshd_config
 echo 'Match LocalPort 22' >> /etc/ssh/sshd_config
 echo '    ChrootDirectory /sysroot/run/imginst/targ' >> /etc/ssh/sshd_config
-kill -HUP $(cat /run/sshd.pid)
+kill $(cat /sysroot/var/run/sshd.pid)
+kill -HUP $(cat /tmp/dbgssh.pid)
+cp /sysroot/etc/pki/ca-trust/source/anchors/* /sysroot/run/imginst/targ/etc/pki/ca-trust/source/anchors/
+chroot /sysroot/run/imginst/targ update-ca-trust
 
 chroot /sysroot/run/imginst/targ bash -c "source /etc/confluent/functions; run_remote post.sh"
 chroot /sysroot bash -c "umount \$(tac /proc/mounts|awk '{print \$2}'|grep ^/run/imginst/targ)"

confluent_osdeploy/el8-diskless/profiles/default/scripts/onboot.sh

@@ -5,6 +5,7 @@
 # noted below so custom commands are executed before
 # the script notifies confluent that install is fully complete.
 
+ntpsrvs=""
 nodename=$(grep ^NODENAME /etc/confluent/confluent.info|awk '{print $2}')
 confluent_apikey=$(cat /etc/confluent/confluent.apikey)
 v4meth=$(grep ^ipv4_method: /etc/confluent/confluent.deploycfg|awk '{print $2}')
@@ -17,6 +18,33 @@ fi
 confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 timedatectl set-timezone $(grep ^timezone: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 hostnamectl set-hostname $nodename
+
+
+if grep ^ntpservers: /etc/confluent/confluent.deploycfg > /dev/null; then
+    for ntpsrv in $(sed -n '/^ntpservers:/,/^[^-]/p' /etc/confluent/confluent.deploycfg|sed 1d|sed '$d' | sed -e 's/^- //'); do
+       echo "server ${ntpsrv} iburst " >> /tmp/timeservers
+    done
+fi
+
+if [ -f /tmp/timeservers ]; then
+
+ntpsrvs=$(cat /tmp/timeservers)
+
+sed -i "1,/^pool * /c\\
+${ntpsrvs//$'\n'/\\$'\n'}" /etc/chrony.conf
+
+
+systemctl restart chronyd
+
+rm -f /tmp/timeservers
+
+fi
+
+
+
+
+
+
 export nodename confluent_mgr confluent_profile
 . /etc/confluent/functions
 mkdir -p /var/log/confluent
@@ -31,7 +59,7 @@ rpm --import /etc/pki/rpm-gpg/*
 
 run_remote_python add_local_repositories
 run_remote_python syncfileclient
-run_remote_python confignet
+run_remote_python confignet -c $confluent_mgr
 
 run_remote onboot.custom
 # onboot scripts may be placed into onboot.d, e.g. onboot.d/01-firstaction.sh, onboot.d/02-secondaction.sh

confluent_osdeploy/el8-diskless/profiles/default/scripts/post.sh

@@ -2,13 +2,17 @@
 
 # This script is executed 'chrooted' into a cloned disk target before rebooting
 #
-
+if [ -f /etc/dracut.conf.d/diskless.conf ]; then
+    rm /etc/dracut.conf.d/diskless.conf
+fi
+for kver in /lib/modules/*; do kver=$(basename $kver); kernel-install add $kver /boot/vmlinuz-$kver; done
 nodename=$(grep ^NODENAME /etc/confluent/confluent.info|awk '{print $2}')
 confluent_apikey=$(cat /etc/confluent/confluent.apikey)
 confluent_mgr=$(grep ^deploy_server: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 export nodename confluent_mgr confluent_profile
 . /etc/confluent/functions
+run_remote setupssh
 mkdir -p /var/log/confluent
 chmod 700 /var/log/confluent
 exec >> /var/log/confluent/confluent-post.log
@@ -33,6 +37,8 @@ run_remote_parts post.d
 # Induce execution of remote configuration, e.g. ansible plays in ansible/post.d/
 run_remote_config post.d
 
+cd /root/
+fetch_remote confignet
 curl -sf -X POST -d 'status: staged' -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $confluent_apikey" https://$confluent_mgr/confluent-api/self/updatestatus
 
 kill $logshowpid

confluent_osdeploy/el8/initramfs/usr/lib/dracut/hooks/cmdline/01-confluent.sh

@@ -1,7 +1,15 @@
 #!/bin/bash
 echo -n "" >> /tmp/net.ifaces
 echo -n "" > /tmp/01-autocons.devnode
-cat /tls/*.0 >> /etc/pki/tls/certs/ca-bundle.crt
+BUNDLENAME=/etc/pki/tls/certs/ca-bundle.crt
+if [ ! -e "$BUNDLENAME" ]; then
+    mkdir -p /etc/pki/tls/certs
+    if [ -e /etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem ]; then
+        BUNDLENAME=/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem
+        ln -s $BUNDLENAME /etc/pki/tls/certs/ca-bundle.crt
+    fi
+fi	
+cat /tls/*.0 >> $BUNDLENAME
 if ! grep console= /proc/cmdline >& /dev/null; then
     autocons=$(/opt/confluent/bin/autocons)
     if [ -n "$autocons" ]; then

confluent_osdeploy/el8/initramfs/usr/lib/dracut/hooks/pre-pivot/01-confluent.sh

@@ -1,5 +1,8 @@
 #!/bin/bash
 BUNDLENAME=/sysroot/etc/pki/tls/certs/ca-bundle.crt
+if [ ! -e "$BUNDLENAME" ]; then
+    BUNDLENAME=/sysroot/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem
+fi
 while [ -h $BUNDLENAME ]; do
     BUNDLENAME=/sysroot/$(readlink $BUNDLENAME)
 done

confluent_osdeploy/el8/profiles/default/initprofile.sh

@@ -1,5 +1,12 @@
 #!/bin/sh
-sed -i 's/centos/CentOS/; s/rhel/Red Hat Enterprise Linux/; s/oraclelinux/Oracle Linux/; s/alma/AlmaLinux/' $2/profile.yaml
+sed -i 's/centos/CentOS/; s/rhel/Red Hat Enterprise Linux/; s/oraclelinux/Oracle Linux/; s/alma/AlmaLinux/;s/fedora/Fedora Linux/' $2/profile.yaml
+if grep Fedora $2/profile.yaml > /dev/null; then
+	sed -i 's/@^minimal-environment/#/' $2/packagelist
+fi
+if grep ^label: $2/profile.yaml | grep 10 > /dev/null; then
+	echo 'echo openssh-keysign >> /tmp/addonpackages' > $2/scripts/pre.d/enablekeysign
+	chmod 644 $2/scripts/pre.d/enablekeysign
+fi
 ln -s $1/images/pxeboot/vmlinuz $2/boot/kernel && \
 ln -s $1/images/pxeboot/initrd.img $2/boot/initramfs/distribution
 mkdir -p $2/boot/efi/boot

confluent_osdeploy/el8/profiles/default/scripts/add_local_repositories

@@ -6,6 +6,7 @@ except ImportError:
 import importlib.util
 import importlib.machinery
 import sys
+import glob
 modloader = importlib.machinery.SourceFileLoader('apiclient', '/opt/confluent/bin/apiclient')
 modspec = importlib.util.spec_from_file_location('apiclient', '/opt/confluent/bin/apiclient', loader=modloader)
 apiclient = importlib.util.module_from_spec(modspec)
@@ -41,6 +42,7 @@ try:
 except AttributeError:
     f = cStringIO.StringIO(cfgdata)
     c.readfp(f)
+gpgkeys = glob.glob('/etc/pki/rpm-gpg/RPM-GPG-KEY-*')
 for sec in c.sections():
     if sec.startswith('variant-'):
         try:
@@ -56,3 +58,6 @@ for sec in c.sections():
                 repopath = repopath[1:]
             repout.write('baseurl=https://{}/confluent-public/os/{}/distribution/{}\n'.format(server, profile, repopath))
             repout.write('enabled=1\n')
+            if gpgkeys:
+                gpgkeyvals = ['file://{}'.format(x) for x in gpgkeys]
+                repout.write('gpgkey=' + ' '.join(gpgkeyvals) + '\n')

confluent_osdeploy/el8/profiles/default/scripts/functions

@@ -10,6 +10,13 @@ function test_mgr() {
     return 1
 }
 
+function initconfluentscriptstmp() {
+    if [ -z "$confluentscripttmpdir" ]; then
+        mkdir -p /opt/confluent/tmpexec
+        confluentscripttmpdir=$(mktemp -d /opt/confluent/tmpexec/confluentscripts.XXXXXXXXX)
+    fi
+}
+
 function confluentpython() {
     if [ -x /usr/libexec/platform-python ]; then
         /usr/libexec/platform-python $*
@@ -72,7 +79,8 @@ fetch_remote() {
 }
 
 source_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -86,7 +94,8 @@ source_remote_parts() {
 }
 
 run_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -105,10 +114,7 @@ source_remote() {
     echo
     echo '---------------------------------------------------------------------------'
     echo Sourcing $1 from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
-        unsettmpdir=1
-    fi
+    initconfluentscriptstmp
     echo Sourcing from $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -135,9 +141,9 @@ run_remote() {
     echo '---------------------------------------------------------------------------'
     echo Running $requestedcmd from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
     if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
         unsettmpdir=1
     fi
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -170,7 +176,8 @@ run_remote_python() {
     fi
     echo '---------------------------------------------------------------------------'
     echo Running python script "'$*'" from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     mkdir -p $(dirname $1)

confluent_osdeploy/el8/profiles/default/scripts/getinstalldisk

@@ -2,6 +2,9 @@
 import subprocess
 import os
 
+class SilentException(Exception):
+    pass
+
 class DiskInfo(object):
     def __init__(self, devname):
         if devname.startswith('nvme') and 'c' in devname:
@@ -24,9 +27,11 @@ class DiskInfo(object):
                 continue
             k, v = prop.split('=', 1)
             if k == 'DEVTYPE' and v != 'disk':
+                if v == 'partition':
+                    raise SilentException('Partition')
                 raise Exception('Not a disk')
             elif k == 'DM_NAME':
-                raise Exception('Device Mapper')
+                raise SilentException('Device Mapper')
             elif k == 'ID_MODEL':
                 self.model = v
             elif k == 'DEVPATH':
@@ -50,13 +55,17 @@ class DiskInfo(object):
                 self.driver = v.replace('"', '')
             elif k == 'ATTRS{subsystype}':
                 self.subsystype = v.replace('"', '')
+            elif k == 'ATTR{ro}' and v == '"1"':
+                raise Exception("Device is read-only")
         if not self.driver and 'imsm' not in self.mdcontainer and self.subsystype != 'nvm':
             raise Exception("No driver detected")
+        if self.driver == 'sr':
+            raise Exception('cd/dvd')
         if os.path.exists('/sys/block/{0}/size'.format(self.name)):
             with open('/sys/block/{0}/size'.format(self.name), 'r') as sizesrc:
                 self.size = int(sizesrc.read()) * 512
-        if int(self.size) < 536870912:
-            raise Exception("Device too small for install")
+        if int(self.size) < 2147483648:
+            raise Exception("Device too small for install ({}MiB)".format(int(self.size)/1024/1024))
 
     @property
     def priority(self):
@@ -89,9 +98,11 @@ def main():
         try:
             disk = DiskInfo(disk)
             disks.append(disk)
+        except SilentException:
+            pass
         except Exception as e:
             print("Skipping {0}: {1}".format(disk, str(e)))
-    nd = [x.name for x in sorted(disks, key=lambda x: x.priority)]
+    nd = [x.name for x in sorted(disks, key=lambda x: [x.priority, x.size])]
     if nd:
         open('/tmp/installdisk', 'w').write(nd[0])
 

confluent_osdeploy/el8/profiles/default/scripts/post.sh

@@ -34,6 +34,11 @@ run_remote_python add_local_repositories
 # run_remote_python will use the appropriate python interpreter path to run the specified script
 # A post.custom is provided to more conveniently hold customizations, see the post.custom file.
 
+# This will do some serial console fixup for bad grub configuration when serial is firmware
+# managed. See script for details
+run_remote_python autoconsole
+
+
 # This will induce server side processing of the syncfile contents if
 # present
 run_remote_python syncfileclient

confluent_osdeploy/el9-diskless/initramfs/usr/lib/dracut/hooks/cmdline/10-confluentdiskless.sh

@@ -42,7 +42,9 @@ if ! grep console= /proc/cmdline >& /dev/null; then
     autocons=$(/opt/confluent/bin/autocons)
     autoconsdev=${autocons%,*}
     autocons=${autocons##*/}
-    echo "Automatic console configured for $autocons"
+    if [ ! -z "$autocons" ]; then
+        echo "Automatic console configured for $autocons"
+    fi
 fi
 echo "Initializing confluent diskless environment"
 echo -n "udevd: "
@@ -102,6 +104,14 @@ while ! grep ^EXTMGRINFO: /etc/confluent/confluent.info | awk -F'|' '{print $3}'
         ip link set $i up
     done
     /opt/confluent/bin/copernicus -t > /etc/confluent/confluent.info
+    echo -n .
+done
+TRIES=0
+while ! grep ^NODENAME: /etc/confluent/confluent.info >& /dev/null && [ "$TRIES" -lt 300 ]; do
+    sleep 0.5
+    echo -n .
+    /opt/confluent/bin/copernicus -t > /etc/confluent/confluent.info
+    TRIES=$((TRIES + 1))
 done
 cd /
 nodename=$(grep ^NODENAME /etc/confluent/confluent.info|awk '{print $2}')
@@ -292,7 +302,7 @@ if [[ $confluent_websrv == *:* ]] && [[ $confluent_websrv != "["* ]]; then
     confluent_websrv="[$confluent_websrv]"
 fi
 echo -n "Initializing ssh..."
-ssh-keygen -A
+ssh-keygen -t ed25519 -f /etc/ssh/ssh_host_ed25519_key -C '' -N ''
 for pubkey in /etc/ssh/ssh_host*key.pub; do
     certfile=${pubkey/.pub/-cert.pub}
     privfile=${pubkey%.pub}
@@ -302,7 +312,10 @@ for pubkey in /etc/ssh/ssh_host*key.pub; do
     fi
     echo HostKey $privfile >> /etc/ssh/sshd_config
 done
-/usr/sbin/sshd
+if grep "debugssh" /proc/cmdline > /dev/null; then
+    /usr/sbin/sshd
+fi
+echo "done"
 confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg| awk '{print $2}')
 confluent_proto=$(grep ^protocol: /etc/confluent/confluent.deploycfg| awk '{print $2}')
 confluent_urls=""

confluent_osdeploy/el9-diskless/profiles/default/scripts/firstboot.sh

@@ -31,8 +31,10 @@ done
 if [ ! -f /etc/confluent/firstboot.ran ]; then
     touch /etc/confluent/firstboot.ran
 
-    cat /etc/confluent/tls/*.pem >> /etc/pki/tls/certs/ca-bundle.crt
 
+    cat /etc/confluent/tls/*.pem >> /etc/pki/tls/certs/ca-bundle.crt
+    confluentpython /root/confignet
+    rm /root/confignet
     run_remote firstboot.custom
     # Firstboot scripts may be placed into firstboot.d, e.g. firstboot.d/01-firstaction.sh, firstboot.d/02-secondaction.sh
     run_remote_parts firstboot.d

confluent_osdeploy/el9-diskless/profiles/default/scripts/functions

@@ -10,6 +10,13 @@ function test_mgr() {
     return 1
 }
 
+function initconfluentscriptstmp() {
+    if [ -z "$confluentscripttmpdir" ]; then
+        mkdir -p /opt/confluent/tmpexec
+        confluentscripttmpdir=$(mktemp -d /opt/confluent/tmpexec/confluentscripts.XXXXXXXXX)
+    fi
+}
+
 function confluentpython() {
     if [ -x /usr/libexec/platform-python ]; then
         /usr/libexec/platform-python $*
@@ -72,7 +79,8 @@ fetch_remote() {
 }
 
 source_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -86,7 +94,8 @@ source_remote_parts() {
 }
 
 run_remote_parts() {
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     apiclient=/opt/confluent/bin/apiclient
     if [ -f /etc/confluent/apiclient ]; then
         apiclient=/etc/confluent/apiclient
@@ -105,10 +114,7 @@ source_remote() {
     echo
     echo '---------------------------------------------------------------------------'
     echo Sourcing $1 from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
-        unsettmpdir=1
-    fi
+    initconfluentscriptstmp
     echo Sourcing from $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -135,9 +141,9 @@ run_remote() {
     echo '---------------------------------------------------------------------------'
     echo Running $requestedcmd from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
     if [ -z "$confluentscripttmpdir" ]; then
-        confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
         unsettmpdir=1
     fi
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     fetch_remote $1
@@ -170,7 +176,8 @@ run_remote_python() {
     fi
     echo '---------------------------------------------------------------------------'
     echo Running python script "'$*'" from https://$confluent_mgr/confluent-public/os/$confluent_profile/scripts/
-    confluentscripttmpdir=$(mktemp -d /tmp/confluentscripts.XXXXXXXXX)
+    unset confluentscripttmpdir
+    initconfluentscriptstmp
     echo Executing in $confluentscripttmpdir
     cd $confluentscripttmpdir
     mkdir -p $(dirname $1)

confluent_osdeploy/el9-diskless/profiles/default/scripts/getinstalldisk

@@ -2,6 +2,9 @@
 import subprocess
 import os
 
+class SilentException(Exception):
+    pass
+
 class DiskInfo(object):
     def __init__(self, devname):
         if devname.startswith('nvme') and 'c' in devname:
@@ -24,9 +27,11 @@ class DiskInfo(object):
                 continue
             k, v = prop.split('=', 1)
             if k == 'DEVTYPE' and v != 'disk':
+                if v == 'partition':
+                    raise SilentException('Partition')
                 raise Exception('Not a disk')
             elif k == 'DM_NAME':
-                raise Exception('Device Mapper')
+                raise SilentException('Device Mapper')
             elif k == 'ID_MODEL':
                 self.model = v
             elif k == 'DEVPATH':
@@ -50,13 +55,17 @@ class DiskInfo(object):
                 self.driver = v.replace('"', '')
             elif k == 'ATTRS{subsystype}':
                 self.subsystype = v.replace('"', '')
+            elif k == 'ATTR{ro}' and v == '"1"':
+                raise Exception("Device is read-only")
         if not self.driver and 'imsm' not in self.mdcontainer and self.subsystype != 'nvm':
             raise Exception("No driver detected")
+        if self.driver == 'sr':
+            raise Exception('cd/dvd')
         if os.path.exists('/sys/block/{0}/size'.format(self.name)):
             with open('/sys/block/{0}/size'.format(self.name), 'r') as sizesrc:
                 self.size = int(sizesrc.read()) * 512
-        if int(self.size) < 536870912:
-            raise Exception("Device too small for install")
+        if int(self.size) < 2147483648:
+            raise Exception("Device too small for install ({}MiB)".format(int(self.size)/1024/1024))
 
     @property
     def priority(self):
@@ -89,9 +98,11 @@ def main():
         try:
             disk = DiskInfo(disk)
             disks.append(disk)
+        except SilentException:
+            pass
         except Exception as e:
             print("Skipping {0}: {1}".format(disk, str(e)))
-    nd = [x.name for x in sorted(disks, key=lambda x: x.priority)]
+    nd = [x.name for x in sorted(disks, key=lambda x: [x.priority, x.size])]
     if nd:
         open('/tmp/installdisk', 'w').write(nd[0])
 

confluent_osdeploy/el9-diskless/profiles/default/scripts/image2disk.py

@@ -11,6 +11,10 @@ import struct
 import sys
 import subprocess
 import traceback
+try:
+    import yaml
+except ImportError:
+    yaml = None
 
 bootuuid = None
 vgname = 'localstorage'
@@ -66,9 +70,9 @@ def get_image_metadata(imgpath):
                 yield md
         else:
             # plausible filesystem structure to apply to a nominally "diskless" image
-            yield {'mount': '/', 'filesystem': 'xfs', 'minsize': 39513563136, 'initsize': 954128662528, 'flags': 'rw,seclabel,relatime,attr2,inode64,logbufs=8,logbsize=32k,noquota', 'device': '/dev/mapper/root', 'compressed_size': 27022069760}
-            yield {'mount': '/boot', 'filesystem': 'xfs', 'minsize': 232316928, 'initsize': 1006632960, 'flags': 'rw,seclabel,relatime,attr2,inode64,logbufs=8,logbsize=32k,noquota', 'device': '/dev/nvme1n1p2', 'compressed_size': 171462656}
-            yield {'mount': '/boot/efi', 'filesystem': 'vfat', 'minsize': 7835648, 'initsize': 627900416, 'flags': 'rw,relatime,fmask=0077,dmask=0077,codepage=437,iocharset=ascii,shortname=winnt,errors=remount-ro', 'device': '/dev/nvme1n1p1', 'compressed_size': 1576960}
+            yield {'mount': '/', 'filesystem': 'xfs', 'minsize': 4294967296, 'initsize': 954128662528, 'flags': 'rw,seclabel,relatime,attr2,inode64,logbufs=8,logbsize=32k,noquota', 'device': '/dev/mapper/root', 'compressed_size': 27022069760}
+            yield {'mount': '/boot', 'filesystem': 'xfs', 'minsize': 536870912, 'initsize': 1006632960, 'flags': 'rw,seclabel,relatime,attr2,inode64,logbufs=8,logbsize=32k,noquota', 'device': '/dev/nvme1n1p2', 'compressed_size': 171462656}
+            yield {'mount': '/boot/efi', 'filesystem': 'vfat', 'minsize': 33554432, 'initsize': 627900416, 'flags': 'rw,relatime,fmask=0077,dmask=0077,codepage=437,iocharset=ascii,shortname=winnt,errors=remount-ro', 'device': '/dev/nvme1n1p1', 'compressed_size': 1576960}
             #raise Exception('Installation from single part image not supported')
 
 class PartedRunner():
@@ -166,6 +170,15 @@ def fixup(rootdir, vols):
     grubsyscfg = os.path.join(rootdir, 'etc/sysconfig/grub')
     if not os.path.exists(grubsyscfg):
         grubsyscfg = os.path.join(rootdir, 'etc/default/grub')
+    currcmdline = []
+    with open('/proc/cmdline') as cmdlinein:
+        cmdline = cmdlinein.read().strip()
+        for arg in cmdline.split():
+            if arg.startswith('console='):
+                currcmdline.append(arg)
+            elif arg == 'quiet':
+                currcmdline.append(arg)
+    currcmdlinestr = ' '.join(currcmdline)
     kcmdline = os.path.join(rootdir, 'etc/kernel/cmdline')
     if os.path.exists(kcmdline):
         with open(kcmdline) as kcmdlinein:
@@ -177,8 +190,10 @@ def fixup(rootdir, vols):
             elif ent.startswith('root='):
                 newkcmdlineent.append('root={}'.format(newrootdev))
             elif ent.startswith('rd.lvm.lv='):
-                ent = convert_lv(ent)
-                if ent:
+                nent = convert_lv(ent)
+                if nent:
+                    newkcmdlineent.append(ent)
+                else:
                     newkcmdlineent.append(ent)
             else:
                 newkcmdlineent.append(ent)
@@ -200,8 +215,10 @@ def fixup(rootdir, vols):
                     elif cfgpart.startswith('resume='):
                         newcfgparts.append('resume={}'.format(newswapdev))
                     elif cfgpart.startswith('rd.lvm.lv='):
-                        cfgpart = convert_lv(cfgpart)
-                        if cfgpart:
+                        ncfgpart = convert_lv(cfgpart)
+                        if ncfgpart:
+                            newcfgparts.append(ncfgpart)
+                        else:
                             newcfgparts.append(cfgpart)
                     else:
                         newcfgparts.append(cfgpart)
@@ -217,13 +234,13 @@ def fixup(rootdir, vols):
         'GRUB_DISABLE_SUBMENU=true',
         'GRUB_TERMINAL=""',
         'GRUB_SERIAL_COMMAND=""',
-        'GRUB_CMDLINE_LINUX="crashkernel=1G-4G:192M,4G-64G:256M,64G-:512M rd.lvm.lv=vg/root rd.lvm.lv=vg/swap"',
+        'GRUB_CMDLINE_LINUX="{}crashkernel=1G-4G:192M,4G-64G:256M,64G-:512M rd.lvm.lv=vg/root rd.lvm.lv=vg/swap"'.format(currcmdlinestr),
         'GRUB_DISABLE_RECOVERY="true"',
         'GRUB_ENABLE_BLSCFG=true',
         ]
     if not os.path.exists(os.path.join(rootdir, "etc/kernel/cmdline")):
         with open(os.path.join(rootdir, "etc/kernel/cmdline"), "w") as cmdlineout:
-            cmdlineout.write("root=/dev/mapper/localstorage-root rd.lvm.lv=localstorage/root")
+            cmdlineout.write("{} root=/dev/mapper/localstorage-root rd.lvm.lv=localstorage/root".format(currcmdlinestr))
     with open(grubsyscfg, 'w') as defgrubout:
         for gline in defgrub:
             gline = gline.split()
@@ -234,11 +251,11 @@ def fixup(rootdir, vols):
                 elif ent.startswith('root='):
                     newline.append('root={}'.format(newrootdev))
                 elif ent.startswith('rd.lvm.lv='):
-                    ent = convert_lv(ent)
-                    if ent:
+                    nent = convert_lv(ent)
+                    if nent:
+                        newline.append(nent)
+                    else:
                         newline.append(ent)
-                    elif '""' in ent:
-                        newline.append('""')
                 else:
                     newline.append(ent)
             defgrubout.write(' '.join(newline) + '\n')
@@ -301,8 +318,8 @@ def fixup(rootdir, vols):
     for vol in vols:
         if vol['mount'] == '/boot/efi':
             targdev = vol['targetdisk']
-            partnum = re.search('(\d+)$', targdev).group(1)
-            targblock = re.search('(.*)\d+$', targdev).group(1)
+            partnum = re.search(r'(\d+)$', targdev).group(1)
+            targblock = re.search(r'(.*)\d+$', targdev).group(1)
     if targblock:
         if targblock.endswith('p') and 'nvme' in targblock:
             targblock = targblock[:-1]
@@ -334,13 +351,16 @@ def had_swap():
 
 newrootdev = None
 newswapdev = None
+vgmap = None
 def install_to_disk(imgpath):
+    global vgmap
     global bootuuid
     global newrootdev
     global newswapdev
     global vgname
     global oldvgname
     lvmvols = {}
+    vgmap = {}
     deftotsize = 0
     mintotsize = 0
     deflvmsize = 0
@@ -365,24 +385,30 @@ def install_to_disk(imgpath):
     mintotsize = swapsize
     for fs in get_image_metadata(imgpath):
         allvols.append(fs)
-        deftotsize += fs['initsize']
-        mintotsize += fs['minsize']
-        if fs['initsize'] > biggestsize:
-            biggestfs = fs
-            biggestsize = fs['initsize']
+
         if fs['device'].startswith('/dev/mapper'):
-            oldvgname = fs['device'].rsplit('/', 1)[-1]
+            odevname = fs['device'].rsplit('/', 1)[-1]
             # if node has - then /dev/mapper will double up the hypen
-            if '_' in oldvgname and '-' in oldvgname.split('_')[-1]:
-                oldvgname = oldvgname.rsplit('-', 1)[0].replace('--', '-')
+            if '_' in odevname and '-' in odevname.split('_', 1)[-1]:
+                oldvgname = odevname.rsplit('-', 1)[0].replace('--', '-')
                 osname = oldvgname.split('_')[0]
                 nodename = socket.gethostname().split('.')[0]
                 vgname = '{}_{}'.format(osname, nodename)
-            lvmvols[fs['device'].replace('/dev/mapper/', '')] = fs
+            elif '-' in odevname: # unique one
+                vgmap[odevname] = odevname.split('-')[0]
+                lvmvols[odevname] = fs
+
+                continue
+            lvmvols[odevname] = fs
             deflvmsize += fs['initsize']
             minlvmsize += fs['minsize']
         else:
-            plainvols[int(re.search('(\d+)$', fs['device'])[0])] = fs
+            plainvols[int(re.search(r'(\d+)$', fs['device'])[0])] = fs
+        if fs['initsize'] > biggestsize:
+            biggestfs = fs
+            biggestsize = fs['initsize']
+        deftotsize += fs['initsize']
+        mintotsize += fs['minsize']
     with open('/tmp/installdisk') as diskin:
         instdisk = diskin.read()
     instdisk = '/dev/' + instdisk
@@ -440,6 +466,28 @@ def install_to_disk(imgpath):
         lvmpart = get_partname(instdisk, volidx + 1)
         subprocess.check_call(['pvcreate', '-ff', '-y', lvmpart])
         subprocess.check_call(['vgcreate', vgname, lvmpart])
+        vgroupmap = {}
+        if yaml and vgmap:
+            with open('/tmp/volumegroupmap.yml') as mapin:
+                vgroupmap = yaml.safe_load(mapin)
+        donedisks = {}
+        for morevolname in vgmap:
+            morevg = vgmap[morevolname]
+            if morevg not in vgroupmap:
+                raise Exception("No mapping defined to create volume group {}".format(morevg))
+            targdisk = vgroupmap[morevg]
+            if targdisk not in donedisks:
+                moreparted = PartedRunner(targdisk)
+                moreparted.run('mklabel gpt')
+                moreparted.run('mkpart lvm 0% 100%')
+                morelvmpart = get_partname(targdisk, 1)
+                subprocess.check_call(['pvcreate', '-ff', '-y', morelvmpart])
+                subprocess.check_call(['vgcreate', morevg, morelvmpart])
+                donedisks[targdisk] = 1
+            morelvname = morevolname.split('-', 1)[1]
+            subprocess.check_call(['lvcreate', '-L', '{}b'.format(lvmvols[morevolname]['initsize']), '-y', '-n', morelvname, morevg])
+            lvmvols[morevolname]['targetdisk'] = '/dev/{}/{}'.format(morevg, morelvname)
+
         vginfo = subprocess.check_output(['vgdisplay', vgname, '--units', 'b']).decode('utf8')
         vginfo = vginfo.split('\n')
         pesize = 0
@@ -452,6 +500,9 @@ def install_to_disk(imgpath):
                 pes = int(infline[4])
         takeaway = swapsize // pesize
         for volidx in lvmvols:
+            if volidx in vgmap:
+                # was handled previously
+                continue
             vol = lvmvols[volidx]
             if vol is biggestfs:
                 continue
@@ -460,6 +511,10 @@ def install_to_disk(imgpath):
         biggestextents = pes - takeaway
         for volidx in lvmvols:
             vol = lvmvols[volidx]
+            if volidx in vgmap:
+                # was handled previously
+                continue
+
             if vol is biggestfs:
                 extents = biggestextents
             else:
@@ -546,7 +601,13 @@ def install_to_disk(imgpath):
 
 
 
-            subprocess.check_call(['umount', '/run/imginst/targ'])
+            while True:
+                try:
+                    subprocess.check_call(['umount', '/run/imginst/targ'])
+                    break
+                except subprocess.CalledProcessError:
+                    print("Failed to unmount /run/imginst/targ, retrying")
+                    time.sleep(1)
         for vol in allvols:
             subprocess.check_call(['mount', vol['targetdisk'], '/run/imginst/targ/' + vol['mount']])
         fixup('/run/imginst/targ', allvols)

confluent_osdeploy/el9-diskless/profiles/default/scripts/imageboot.sh

@@ -129,5 +129,11 @@ mv /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs
 ln -s /sysroot/lib/modules/$(uname -r) /lib/modules/
 mv /lib/firmware /lib/firmware-ramfs
 ln -s /sysroot/lib/firmware /lib/firmware
+rm -f /sysroot/etc/dracut.conf.d/diskless.conf # remove diskless dracut from runtime, to make kdump happier
 kill $(grep -l ^/usr/lib/systemd/systemd-udevd  /proc/*/cmdline|cut -d/ -f 3)
-exec /opt/confluent/bin/start_root
+if grep debugssh /proc/cmdline >& /dev/null; then
+    exec /opt/confluent/bin/start_root
+else
+    rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
+    exec /opt/confluent/bin/start_root -s  # share mount namespace, keep kernel callbacks intact
+fi

confluent_osdeploy/el9-diskless/profiles/default/scripts/installimage

@@ -5,8 +5,12 @@
 # and existing mounts of image (to take advantage of caching)
 mount -o bind /sys /sysroot/sys
 mount -o bind /dev /sysroot/dev
+mount -o bind /dev/pts /sysroot/dev/pts
 mount -o bind /proc /sysroot/proc
 mount -o bind /run /sysroot/run
+mount -t efivarfs none /sysroot/sys/firmware/efi/efivars
+
+
 
 
 if [ ! -f /tmp/mountparts.sh ]; then
@@ -21,8 +25,16 @@ else
     done
 fi
 cd /sysroot/run
+[ -f /run/sshd.pid ] && 
+cp /run/sshd.pid /tmp/dbgssh.pid
+chmod 0600 /sysroot/etc/ssh/ssh*key
+chroot /sysroot/ bash -c "/usr/sbin/sshd"
 chroot /sysroot/ bash -c "source /etc/confluent/functions; run_remote_python getinstalldisk"
 chroot /sysroot/ bash -c "source /etc/confluent/functions; run_remote_parts pre.d"
+for nameserver in $(sed -n '/^nameservers:/,/^[^-]/p' /etc/confluent/confluent.deploycfg|grep ^- | cut -d ' ' -f 2|sed -e 's/ //'); do
+    echo "nameserver $nameserver" >> /sysroot/etc/resolv.conf
+done
+
 if [ ! -f /sysroot/tmp/installdisk ]; then
     echo 'Unable to find a suitable installation target device, ssh to port 2222 to investigate'
     while [ ! -f /sysroot/tmp/installdisk ]; do
@@ -40,7 +52,8 @@ chroot /sysroot bash -c "source /etc/confluent/functions; run_remote_python imag
 echo "Port 22" >> /etc/ssh/sshd_config
 echo 'Match LocalPort 22' >> /etc/ssh/sshd_config
 echo '    ChrootDirectory /sysroot/run/imginst/targ' >> /etc/ssh/sshd_config
-kill -HUP $(cat /run/sshd.pid)
+kill $(cat /sysroot/var/run/sshd.pid)
+[ -f /tmp/dbgssh.pid ] && kill -HUP $(cat /tmp/dbgssh.pid)
 cp /sysroot/etc/pki/ca-trust/source/anchors/* /sysroot/run/imginst/targ/etc/pki/ca-trust/source/anchors/
 chroot /sysroot/run/imginst/targ update-ca-trust
 

confluent_osdeploy/el9-diskless/profiles/default/scripts/onboot.sh

@@ -4,7 +4,7 @@
 # completed. It is best to edit the middle of the file as
 # noted below so custom commands are executed before
 # the script notifies confluent that install is fully complete.
-
+ntpsrvs=""
 nodename=$(grep ^NODENAME /etc/confluent/confluent.info|awk '{print $2}')
 confluent_apikey=$(cat /etc/confluent/confluent.apikey)
 v4meth=$(grep ^ipv4_method: /etc/confluent/confluent.deploycfg|awk '{print $2}')
@@ -17,6 +17,28 @@ fi
 confluent_profile=$(grep ^profile: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 timedatectl set-timezone $(grep ^timezone: /etc/confluent/confluent.deploycfg|awk '{print $2}')
 hostnamectl set-hostname $nodename
+
+
+if grep ^ntpservers: /etc/confluent/confluent.deploycfg > /dev/null; then
+    for ntpsrv in $(sed -n '/^ntpservers:/,/^[^-]/p' /etc/confluent/confluent.deploycfg|sed 1d|sed '$d' | sed -e 's/^- //'); do
+        echo "server ${ntpsrv} iburst " >> /tmp/timeservers
+    done
+fi
+
+if [ -f /tmp/timeservers ]; then
+
+ntpsrvs=$(cat /tmp/timeservers)
+
+sed -i "1,/^pool * /c\\
+
+${ntpsrvs//$'\n'/\\$'\n'}" /etc/chrony.conf
+
+
+systemctl restart chronyd
+
+rm -f /tmp/timeservers
+fi
+
 export nodename confluent_mgr confluent_profile
 . /etc/confluent/functions
 mkdir -p /var/log/confluent
@@ -31,7 +53,7 @@ rpm --import /etc/pki/rpm-gpg/*
 
 run_remote_python add_local_repositories
 run_remote_python syncfileclient
-run_remote_python confignet
+run_remote_python confignet -c $confluent_mgr
 
 run_remote onboot.custom
 # onboot scripts may be placed into onboot.d, e.g. onboot.d/01-firstaction.sh, onboot.d/02-secondaction.sh

confluent_osdeploy/el9-diskless/profiles/default/scripts/post.sh

@@ -43,7 +43,8 @@ run_remote_parts post.d
 
 # Induce execution of remote configuration, e.g. ansible plays in ansible/post.d/
 run_remote_config post.d
-
+cd /root/
+fetch_remote confignet
 # rebuild initrd, pick up new drivers if needed
 dracut -f /boot/initramfs-$(uname -r).img $(uname -r)
 

confluent_osdeploy/esxi7/initramfs/bin/dcuiweasel

@@ -6,6 +6,7 @@ if [ ! -f /var/run/vmware/show-esx-shell-login ]; then
     chvt 2
     /etc/init.d/ESXShell start
 fi
+/opt/confluent/bin/apiclient -f
 uuid=$(vsish -e get /hardware/bios/dmiInfo|grep -A15 UUID|sed -e 's/.*://'|sed -e ':a;N;$!ba;s/\n//g' | sed -e 's/ *0x//g')
 uuid=${uuid:0:8}-${uuid:8:4}-${uuid:12:4}-${uuid:16:4}-${uuid:20:12}
 kargs=$(vsish -e get /system/bootCmdLine|grep "command line:")
@@ -61,8 +62,8 @@ if [ -e /tmp/confluentident/cnflnt.yml ]; then
         fi
         v4nm=$(grep ipv4_netmask: $tcfg)
         v4nm=${v4nm#ipv4_netmask: }
-        localcli network ip interface ipv4 set -i vmk0 -I $v4addr -N $v4nm -g $v4gw -t static
-        localcli network ip route ipv4 add -n default -g $v4gw
+        while ! localcli network ip interface ipv4 set -i vmk0 -I $v4addr -N $v4nm -g $v4gw -t static; do echo "Retrying..."; sleep 5; done
+        while ! localcli network ip route ipv4 add -n default -g $v4gw; do sleep 1; done
     fi
     hmackeyfile=$(mktemp)
     echo -n $(grep ^apitoken: /tmp/confluentident/cnflnt.yml|awk '{print $2}') > $hmackeyfile
@@ -72,6 +73,20 @@ if [ -e /tmp/confluentident/cnflnt.yml ]; then
     hmacfile=$(mktemp)
     ln -s /opt/confluent/bin/clortho /opt/confluent/bin/genpasshmac
     /opt/confluent/bin/genpasshmac $passfile $passcrypt $hmacfile $hmackeyfile
+    echo -n 'Checking connectivity to server: '
+    maxwait=30
+    while ! /opt/confluent/bin/apiclient -c >& /dev/null; do
+        echo -n '.'
+        sleep 1
+        maxwait=$((maxwait - 1))
+        if [ $maxwait -le 0 ]; then
+            echo "Unable to contact deployment server, verify network connectivity"
+            echo "A debug session has been made available on Alt-F1"
+            sleep 30
+            maxwait=30
+	fi
+    done
+    echo
     echo -n 'Registering new API key with deployment server: '
     /opt/confluent/bin/apiclient -p $hmacfile /confluent-api/self/registerapikey $passcrypt
     echo

confluent_osdeploy/esxi7/profiles/hypervisor/kickstart

@@ -1,6 +1,5 @@
 accepteula
-clearpart --firstdisk --overwritevmfs
-install --firstdisk --overwritevmfs
+%include /tmp/storagecfg
 %include /tmp/ksnet
 %include /tmp/rootpw
 reboot