Fix type of maxnodes

Start buffer daemon only when needed.

Limit restarts to once every 30 seconds.

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
 

build/arm/el10/Dockerfile

@@ -0,0 +1,8 @@
+FROM almalinux:10
+RUN ["yum", "-y","update"]
+RUN ["yum", "-y","install","gcc","make","rpm-build","python3-devel","python3-setuptools","createrepo","python3", "perl", "perl-DBI", "perl-JSON", "perl-XML-LibXML", "pinentry-tty", "rpm-sign", "git", "golang"]
+ADD rpmmacro /root/.rpmmacros
+ADD buildpackages.sh /bin/
+#VOLUME ["/rpms", "/srpms"]
+CMD ["/bin/bash","/bin/buildpackages.sh"]
+

build/arm/el10/buildpackages.sh

@@ -0,0 +1,6 @@
+for package in /srpms/*; do
+        rpmbuild --rebuild $package
+done
+find ~/rpmbuild/RPMS -type f -exec cp {} /rpms/ \;
+
+

build/arm/el10/rpmmacro

@@ -0,0 +1,3 @@
+%_gpg_digest_algo sha256
+%_gpg_name Lenovo Scalable Infrastructure
+

build/arm/el8/Dockerfile

@@ -0,0 +1,8 @@
+FROM almalinux:8
+RUN ["yum", "-y","update"]
+RUN ["yum", "-y","install","gcc","make","rpm-build","python3-devel","python3-setuptools","createrepo","python3", "perl", "perl-DBI", "perl-JSON", "perl-Net-DNS", "perl-DB_File", "perl-XML-LibXML", "rpm-sign", "git", "fuse-devel","libcurl-devel"]
+ADD rpmmacro /root/.rpmmacros
+ADD buildpackages.sh /bin/
+#VOLUME ["/rpms", "/srpms"]
+CMD ["/bin/bash","/bin/buildpackages.sh"]
+

build/arm/el8/buildpackages.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+for package in /srpms/*; do
+        rpmbuild --rebuild $package
+done
+find ~/rpmbuild/RPMS -type f -exec cp {} /rpms/ \;
+

build/arm/el8/rpmmacro

@@ -0,0 +1,2 @@
+%_gpg_digest_algo sha256
+%_gpg_name Lenovo Scalable Infrastructure

build/arm/el9/Dockerfile

@@ -0,0 +1,10 @@
+FROM almalinux:9
+RUN ["yum", "-y","update"]
+RUN ["yum", "-y","install","gcc","make","rpm-build","python3-devel","python3-setuptools","createrepo","python3", "perl", "perl-DBI", "perl-JSON", "perl-Net-DNS", "perl-DB_File", "perl-XML-LibXML", "pinentry-tty", "rpm-sign", "epel-release", "git"]
+RUN ["crb", "enable"]
+RUN ["yum", "-y","install","fuse-devel","libcurl-devel"]
+ADD rpmmacro /root/.rpmmacros
+ADD buildpackages.sh /bin/
+#VOLUME ["/rpms", "/srpms"]
+CMD ["/bin/bash","/bin/buildpackages.sh"]
+

build/arm/el9/buildpackages.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+for package in /srpms/*; do
+        rpmbuild --rebuild $package
+done
+find ~/rpmbuild/RPMS -type f -exec cp {} /rpms/ \;
+

build/arm/el9/rpmmacro

@@ -0,0 +1,2 @@
+%_gpg_digest_algo sha256
+%_gpg_name Lenovo Scalable Infrastructure

build/arm/noble/Dockerfile

@@ -0,0 +1,12 @@
+FROM ubuntu:noble
+ADD stdeb.patch /tmp/
+ADD buildapt.sh /bin/
+ADD distributions.tmpl /bin/
+RUN ["apt-get", "update"]
+RUN ["apt-get", "install", "-y", "reprepro", "python3-stdeb", "gnupg-agent", "devscripts", "debhelper", "libsoap-lite-perl", "libdbi-perl", "quilt", "git", "python3-pyparsing", "python3-dnspython", "python3-eventlet", "python3-netifaces", "python3-paramiko", "dh-python", "libjson-perl", "ronn", "alien", "gcc", "make"]
+RUN ["mkdir", "-p", "/sources/git/"]
+RUN ["mkdir", "-p", "/debs/"]
+RUN ["mkdir", "-p", "/apt/"]
+RUN ["bash", "-c", "patch -p1 < /tmp/stdeb.patch"]
+CMD ["/bin/bash", "/bin/buildapt.sh"]
+

build/arm/noble/buildapt.sh

@@ -0,0 +1,21 @@
+#cp -a /sources/git /tmp
+for builder in $(find /sources/git -name builddeb); do
+    cd $(dirname $builder)
+    ./builddeb /debs/
+done
+cp /prebuilt/* /debs/
+cp /osd/*.deb /debs/
+mkdir -p /apt/conf/
+CODENAME=$(grep VERSION_CODENAME= /etc/os-release | sed -e 's/.*=//')
+if [ -z "$CODENAME" ]; then
+    CODENAME=$(grep VERSION= /etc/os-release | sed -e 's/.*(//' -e 's/).*//')
+fi
+if ! grep $CODENAME /apt/conf/distributions; then
+    sed -e s/#CODENAME#/$CODENAME/ /bin/distributions.tmpl >> /apt/conf/distributions
+fi
+cd /apt/
+reprepro includedeb $CODENAME /debs/*.deb
+for dsc in /debs/*.dsc; do
+    reprepro includedsc $CODENAME $dsc
+done
+

build/arm/noble/distributions.tmpl

@@ -0,0 +1,7 @@
+Origin: Lenovo HPC Packages
+Label: Lenovo HPC Packages
+Codename: #CODENAME#
+Architectures: amd64 source
+Components: main
+Description: Lenovo HPC Packages
+

build/arm/noble/stdeb.patch

@@ -0,0 +1,34 @@
+diff -urN t/usr/lib/python3/dist-packages/stdeb/cli_runner.py t.patch/usr/lib/python3/dist-packages/stdeb/cli_runner.py
+--- t/usr/lib/python3/dist-packages/stdeb/cli_runner.py 2024-06-11 18:30:13.930328999 +0000
++++ t.patch/usr/lib/python3/dist-packages/stdeb/cli_runner.py   2024-06-11 18:32:05.392731405 +0000
+@@ -8,7 +8,7 @@
+     from ConfigParser import SafeConfigParser  # noqa: F401
+ except ImportError:
+     # python 3.x
+-    from configparser import SafeConfigParser  # noqa: F401
++    from configparser import ConfigParser  # noqa: F401
+ from distutils.util import strtobool
+ from distutils.fancy_getopt import FancyGetopt, translate_longopt
+ from stdeb.util import stdeb_cmdline_opts, stdeb_cmd_bool_opts
+diff -urN t/usr/lib/python3/dist-packages/stdeb/util.py t.patch/usr/lib/python3/dist-packages/stdeb/util.py
+--- t/usr/lib/python3/dist-packages/stdeb/util.py       2024-06-11 18:32:53.864776149 +0000
++++ t.patch/usr/lib/python3/dist-packages/stdeb/util.py 2024-06-11 18:33:02.063952870 +0000
+@@ -730,7 +730,7 @@
+     example.
+     """
+ 
+-    cfg = ConfigParser.SafeConfigParser()
++    cfg = ConfigParser.ConfigParser()
+     cfg.read(cfg_files)
+     if cfg.has_section(module_name):
+         section_items = cfg.items(module_name)
+@@ -801,7 +801,7 @@
+         if len(cfg_files):
+             check_cfg_files(cfg_files, module_name)
+ 
+-        cfg = ConfigParser.SafeConfigParser(cfg_defaults)
++        cfg = ConfigParser.ConfigParser(cfg_defaults)
+         for cfg_file in cfg_files:
+             with codecs.open(cfg_file, mode='r', encoding='utf-8') as fd:
+                 cfg.readfp(fd)
+

build/buildarm

@@ -0,0 +1,9 @@
+cd ~/confluent
+git pull
+rm ~/rpmbuild/RPMS/noarch/*osdeploy*
+rm ~/rpmbuild/SRPMS/*osdeploy*
+sh confluent_osdeploy/buildrpm-aarch64
+mkdir -p $HOME/el9/
+mkdir -p $HOME/el10/
+podman run --rm -it -v $HOME:/build el9build bash /build/confluent/confluent_vtbufferd/buildrpm /build/el9/
+

confluent_client/bin/confetty

@@ -45,6 +45,7 @@ import math
 import getpass
 import optparse
 import os
+import re
 import select
 import shlex
 import signal
@@ -132,6 +133,8 @@ def print_help():
 
 def updatestatus(stateinfo={}):
     global powerstate, powertime, clearpowermessage
+    if opts.headless:
+        return
     status = consolename
     info = []
     for statekey in stateinfo:
@@ -454,8 +457,10 @@ def do_command(command, server):
         currconsole = targpath
         startrequest = {'operation': 'start', 'path': targpath,
                         'parameters': {}}
-        height, width = struct.unpack(
-            'hh', fcntl.ioctl(sys.stdout, termios.TIOCGWINSZ, b'....'))[:2]
+        height, width = 31, 100
+        if not opts.headless:
+            height, width = struct.unpack(
+                'hh', fcntl.ioctl(sys.stdout, termios.TIOCGWINSZ, b'....'))[:2]
         startrequest['parameters']['width'] = width
         startrequest['parameters']['height'] = height
         for param in argv[2:]:
@@ -623,17 +628,19 @@ def startconsole(nodename):
     signal.signal(signal.SIGWINCH, do_resize)
     didconsole = True
     consolename = nodename
-    tty.setraw(sys.stdin.fileno())
-    currfl = fcntl.fcntl(sys.stdin.fileno(), fcntl.F_GETFL)
-    fcntl.fcntl(sys.stdin.fileno(), fcntl.F_SETFL, currfl | os.O_NONBLOCK)
+    if not opts.headless:
+        tty.setraw(sys.stdin.fileno())
+        currfl = fcntl.fcntl(sys.stdin.fileno(), fcntl.F_GETFL)
+        fcntl.fcntl(sys.stdin.fileno(), fcntl.F_SETFL, currfl | os.O_NONBLOCK)
     inconsole = True
+    check_automation('')  # give any leading 'sends' a chance
 
 
 def quitconfetty(code=0, fullexit=False, fixterm=True):
     global inconsole
     global currconsole
     global didconsole
-    if fixterm or didconsole:
+    if (fixterm or didconsole) and not opts.headless:
         currfl = fcntl.fcntl(sys.stdin.fileno(), fcntl.F_GETFL)
         fcntl.fcntl(sys.stdin.fileno(), fcntl.F_SETFL, currfl & ~os.O_NONBLOCK)
         if oldtcattr is not None:
@@ -843,6 +850,20 @@ def check_escape_seq(currinput, filehandle):
         currinput += filehandle.read()
     return currinput
 
+automation_directives = []
+current_automation_directive = None
+automation_map = {
+    '<up>': '\x1b[A',
+    '<down>': '\x1b[B',
+    '<right>': '\x1b[C',
+    '<left>': '\x1b[D',
+    '<enter>': '\r',
+    '<esc>': '\x1b',
+    '<tab>': '\t',
+}
+
+
+
 parser = optparse.OptionParser()
 parser.add_option("-s", "--server", dest="netserver",
                   help="Confluent instance to connect to",
@@ -850,12 +871,64 @@ parser.add_option("-s", "--server", dest="netserver",
 parser.add_option("-c", "--control", dest="controlpath",
                   help="Path to offer terminal control",
                   metavar="PATH")
+parser.add_option('-a', '--automation', type='string', default=None,
+                  help='Specify an automation script to run', metavar='SCRIPT')
+parser.add_option('-e', '--headless', action='store_true', default=False,
+                  help='Run in headless mode, which is designed for use with '
+                       'automation scripts and disables interactive features')
 parser.add_option(
     '-m', '--mintime', default=0,
     help='Minimum time to run or else pause for input (used to keep a '
          'terminal from closing quickly on error)')
 opts, shellargs = parser.parse_args()
 
+def parse_automation_script(script, session_node):
+    global current_automation_directive
+    for line in script.splitlines():
+        line = line.strip()
+        if not line or line.startswith('#'):
+            continue
+        if line.startswith('exit'):
+            automation_directives.append(('exit', None))
+            continue
+        if ' ' not in line:
+            sys.stderr.write("Invalid line in automation script: %s\n" % line)
+            continue
+        directive, arg = line.split(' ', 1)
+        directive = directive.strip().lower()
+        if directive not in ('expect', 'send', 'forget'):
+            sys.stderr.write("Unknown directive in automation script: %s\n" % directive)
+            continue
+        arg = arg.strip()
+        origarg = arg
+        if arg[0] not in ('"', "'"):
+            arg = '"' + arg + '"'
+        if arg[0] == "'" and arg[-1] == "'":
+            # do not process '<>' sequences in single quotes
+            arg = arg[1:-1]
+            arg = bytes(arg, "utf-8").decode("unicode_escape")
+        elif arg[0] == '"' and arg[-1] == '"':
+            arg = bytes(arg[1:-1], "utf-8").decode("unicode_escape")
+            for key, value in automation_map.items():
+                arg = arg.replace(key, value)
+            arg = re.sub(r'<env:(\w+)>', lambda m: os.environ[m.group(1)], arg)
+        if '{' in arg:  # support confluent expressions
+            for res in session.create('/nodes/{0}/attributes/expression'.format(session_node),
+                            {'expression': arg}):
+                if 'error' in res:
+                    sys.stderr.write(res['error'] + '\n')
+                    sys.exit(1)
+                if 'value' in res:
+                    arg = res['value']
+        automation_directives.append((directive, arg, origarg))
+    if automation_directives:
+        current_automation_directive = automation_directives.pop(0)
+        if opts.headless and current_automation_directive[0] == 'expect':
+            sys.stdout.write(f'Expecting {repr(current_automation_directive[2])}\r\n')
+            sys.stdout.flush()
+
+
+
 username = None
 passphrase = None
 def server_connect():
@@ -908,6 +981,7 @@ def main():
     # sys.stdout.write('\x1b[H\x1b[J')
     # sys.stdout.flush()
     global powerstate, powertime, clearpowermessage
+    
 
 
     if sys.stdout.isatty():
@@ -924,6 +998,9 @@ def main():
     targ, session_node = get_session_node(shellargs)
     if session_node is not None:
         consoleonly = True
+        if opts.automation:
+            with open(opts.automation) as f:
+                parse_automation_script(f.read(), session_node)
         do_command("start %s" % targ, netserver)
         doexit = True
     elif shellargs:
@@ -935,9 +1012,13 @@ def main():
 
     while inconsole or not doexit:
         if inconsole:
+            if opts.headless:
+                handles = [session.connection]
+            else:
+                handles = (sys.stdin, session.connection)
             try:
                 rdylist, _, _ = select.select(
-                    (sys.stdin, session.connection), (), (), 10)
+                    handles, (), (), 10)
             except select.error:
                 rdylist = ()
             for fh in rdylist:
@@ -969,8 +1050,81 @@ 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 = ''
+automation_check = ''
+
+def check_automation(data):
+    global automation_check
+    global current_automation_directive
+    if type(data) != str:
+        data = data.decode('utf-8', errors='ignore')
+    while current_automation_directive:
+        if current_automation_directive[0] == 'forget' and data:
+            current_automation_directive = None
+            automation_check = ''
+            if automation_directives:
+                current_automation_directive = automation_directives.pop(0)
+                if opts.headless and current_automation_directive[0] == 'expect':
+                    sys.stdout.write(f'Expecting {repr(current_automation_directive[2])}\n')
+                    sys.stdout.flush()
+            return
+        if current_automation_directive[0] == 'expect':
+            expected = current_automation_directive[1]
+            combined = automation_check + data
+            if expected and expected in combined:
+                data = data[combined.rindex(expected) + len(expected):]
+
+                if opts.headless:
+                    sys.stdout.write(f'Detected {repr(current_automation_directive[2])}\r\n')
+                    sys.stdout.flush()
+                current_automation_directive = None
+                automation_check = ''
+                if automation_directives:
+                    current_automation_directive = automation_directives.pop(0)
+                    if opts.headless and current_automation_directive[0] == 'expect':
+                        sys.stdout.write(f'Expecting {repr(current_automation_directive[2])}\r\n')
+                        sys.stdout.flush()
+            else:
+                # Check if there's potential start of expected data in the incoming data
+                combined = automation_check + data
+                automation_check = ''
+                for i in range(1, min(len(expected), len(combined)) + 1):
+                    if expected.startswith(combined[-i:]):
+                        automation_check = combined[-i:]
+                return # wait for next check
+        elif current_automation_directive[0] == 'send':
+            data = ''
+            automation_check = ''
+            if opts.headless:
+                sys.stdout.write(f'Sending {repr(current_automation_directive[2])}\r\n')
+                sys.stdout.flush()
+            if current_automation_directive[1]:
+                tlvdata.send(session.connection, current_automation_directive[1])
+            current_automation_directive = None
+            if automation_directives:
+                current_automation_directive = automation_directives.pop(0)
+                if opts.headless and current_automation_directive[0] == 'expect':
+                    sys.stdout.write(f'Expecting {repr(current_automation_directive[2])}\r\n')
+                    sys.stdout.flush()
+        elif current_automation_directive[0] == 'exit':
+            if opts.headless:
+                sys.stdout.write('Automation completed\r\n')
+                sys.stdout.flush()
+            return True
+        if opts.headless and not current_automation_directive:
+            sys.stdout.write('Automation completed\r\n')
+            sys.stdout.flush()
+            return True
+    return False
+
 def consume_termdata(fh, bufferonly=False):
     global clearpowermessage
+    global fgcolor, bgcolor, fgshifted, pendseq
     try:
         data = tlvdata.recv(fh)
     except Exception:
@@ -979,7 +1133,64 @@ def consume_termdata(fh, bufferonly=False):
         updatestatus(data)
         return ''
     if data is not None:
-        data = client.stringify(data)
+        shouldexit = check_automation(data)
+        if opts.headless:
+            if shouldexit:
+                quitconfetty(fullexit=True)
+            return ''
+        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
@@ -1006,6 +1217,8 @@ def consume_termdata(fh, bufferonly=False):
             # this scenario comfortable that it
             # will come out soon enough
             pass
+        if shouldexit:
+            quitconfetty(fullexit=True)
     else:
         deadline = 5
         connected = 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/nodeapply

@@ -36,6 +36,36 @@ import confluent.client as client
 import confluent.sortutil as sortutil
 devnull = None
 
+def run_automation(noderange, category, c):
+    automationbynode = {}
+    for res in c.update('/noderange/{0}/deployment/remote_config/run'.format(noderange), {
+            'category': category,
+        }):
+            if 'error' in res:
+                sys.stderr.write(res['error'] + '\n')
+                exitcode |= res.get('errorcode', 1)
+            if 'created' in res:
+                nodename = res['created'].split('/')[2]
+                automationbynode[nodename] = res['created']
+    while automationbynode:
+        for node in list(automationbynode):
+            for res in c.read(automationbynode[node]):
+                if 'error' in res:
+                    sys.stderr.write(res['error'] + '\n')
+                    exitcode |= res.get('errorcode', 1)
+                for result in res.get('results', []):
+                    sys.stdout.write('{0}: Task [{1}] {2}\n'.format(
+                        node, result['task_name'], result['state']))
+                    for warning in result.get('warnings', []):
+                        sys.stderr.write('{0}: [WARNING] {1}\n'.format(node, warning))
+                    if 'errorinfo' in result:
+                        for errorline in result['errorinfo'].splitlines():
+                            sys.stderr.write('{0}: [ERROR] {1}\n'.format(node, errorline))                
+                if res.get('complete', False):
+                    del automationbynode[node]
+                    sys.stdout.write('{0}: Automation complete\n'.format(node))
+                    
+
 def run():
     global devnull
     devnull = open(os.devnull, 'rb')
@@ -51,6 +81,8 @@ def run():
                          help='Run the syncfiles associated with the currently completed OS profile on the noderange')
     argparser.add_option('-P', '--scripts',
                          help='Re-run specified scripts, with full path under scripts, e.g. post.d/first,firstboot.d/second')
+    argparser.add_option('-A', '--automation',
+                         help='Run the automation scripts associated with the current OS profile on the noderange, specifying category (onboot.d/firstboot.d/post.d)')
     argparser.add_option('-m', '--maxnodes', type='int',
                     help='Specify a maximum number of '
                         'nodes to run remote ssh command to, '
@@ -74,16 +106,13 @@ def run():
     exitcode = 0
 
     c.stop_if_noderange_over(args[0], options.maxnodes)
+    if options.automation:
+        run_automation(args[0], options.automation, c)
+        
     nodemap = {}
     cmdparms = []
     nodes = []
-    for res in c.read('/noderange/{0}/nodes/'.format(args[0])):
-        if 'error' in res:
-            sys.stderr.write(res['error'] + '\n')
-            exitcode |= res.get('errorcode', 1)
-            break
-        node = res['item']['href'][:-1]
-        nodes.append(node)
+
     
     cmdstorun = []
     if options.security:
@@ -94,8 +123,17 @@ def run():
         for script in options.scripts.split(','):
             cmdstorun.append(['run_remote', script])
     if not cmdstorun:
+        if options.automation:
+            sys.exit(0)
         argparser.print_help()
         sys.exit(1)
+    for res in c.read('/noderange/{0}/nodes/'.format(args[0])):
+        if 'error' in res:
+            sys.stderr.write(res['error'] + '\n')
+            exitcode |= res.get('errorcode', 1)
+            break
+        node = res['item']['href'][:-1]
+        nodes.append(node)
     idxbynode = {}
     cmdvbase = ['bash', '/etc/confluent/functions']
     for sshnode in nodes:
@@ -145,7 +183,7 @@ def run():
                                 run_cmdv(node, cmdv, all, poller, pipedesc)
                             elif pendingexecs:
                                 node, cmdv = pendingexecs.popleft()
-                                run_cmdv(node, cmdv, all, poller. pipedesc)
+                                run_cmdv(node, cmdv, all, poller, pipedesc)
             singlepoller.close()
         for node in sortutil.natural_sort(pernodeout):
             for line in pernodeout[node]:

confluent_client/bin/nodecertutil

@@ -0,0 +1,104 @@
+#!/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')
+
+    sign_bmc_parser = subparsers.add_parser('signbmccert', help='Sign BMC certificate')
+    sign_bmc_parser.add_argument('--days', type=int, help='Number of days the certificate is valid for')
+    sign_bmc_parser.add_argument('--added-names', type=str, help='Additional names to include in the certificate')
+
+    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)
+    elif args.command == 'signbmccert':
+        payload = {}
+        if args.days is not None:
+            payload['days'] = args.days
+        else:
+            print("Error: --days is required for signbmccert", file=sys.stderr)
+            sys.exit(1)
+        if args.added_names:
+            payload['added_names'] = args.added_names
+        for res in c.update(f'/noderange/{args.noderange}/configuration/management_controller/certificate/sign', payload):
+            print(repr(res))
+    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.
@@ -62,7 +62,7 @@ argparser.add_option('-e', '--extra', dest='extra',
                            'to be extra configuration')
 argparser.add_option('-x', '--exclude', dest='exclude',
                      action='store_true', default=False,
-                     help='Treat positional arguments as items to not '
+                     help='Treat named settings as items to not '
                           'examine, compare, or restore default')
 argparser.add_option('-a', '--advanced', dest='advanced',
                      action='store_true', default=False,
@@ -73,7 +73,7 @@ argparser.add_option('-r', '--restoredefault', default=False,
                      dest='restoredefault', metavar="COMPONENT",
                      help='Restore the configuration of the node '
                           'to factory default for given component. '
-                          'Currently only uefi is supported')
+                          'Currently "uefi" and "bmc" are supported components')
 argparser.add_option('-m', '--maxnodes', type='int',
                      help='Specify a maximum number of '
                           'nodes to configure, '
@@ -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

@@ -18,6 +18,7 @@
 import base64
 import optparse
 import os
+import shlex
 import subprocess
 import sys
 path = os.path.dirname(os.path.realpath(__file__))
@@ -38,7 +39,7 @@ import termios
 import fcntl
 import confluent.screensqueeze as sq
 try:
-    from PIL import Image, ImageDraw
+    from PIL import Image, ImageDraw, ImageFont
 except ImportError:
     Image = None
 
@@ -62,8 +63,13 @@ argparser = optparse.OptionParser(
            "ctrl-'e', then release ctrl, then 'c', then '?' for a full list. "
            "For example, ctrl-'e', then 'c', then '.' will exit the current "
            "console")
+argparser.add_option('-a', '--automation', type='string', default=None,
+                     help='Specify an automation script')
 argparser.add_option('-t', '--tile', action='store_true', default=False,
                      help='Tile console windows in the terminal')
+argparser.add_option('-e', '--headless', action='store_true', default=False,
+                     help='Run in headless mode, which is designed for use with '
+                          'automation scripts and disables interactive features')
 argparser.add_option('-l', '--log', action='store_true', default=False,
                      help='Enter log replay mode instead of showing a live console')
 
@@ -98,6 +104,12 @@ argparser.add_option('-w','--windowed', action='store_true', default=False,
 
 (options, args) = argparser.parse_args()
 
+automation_args = []
+if options.automation:
+    automation_args = ['-a', options.automation]
+if options.headless:
+    automation_args += ['--headless']
+
 oldtcattr = None
 oldfl = None
 
@@ -227,22 +239,24 @@ def draw_text(text, width, height):
         nd = ImageDraw.Draw(nerr)
         for txtpiece in text.split('\n'):
             fntsize = 8
-            while nd.textlength(txtpiece, font_size=fntsize) < int(imgwidth * 0.90):
+            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_size=maxfntsize)
-        nd.rectangle((0, 0, nerr.width - 1, nerr.height -1), outline='white', width=1)
+        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(txt))
+        cursor_left(len(text))
 
 def draw_image(data, width, height, doscale=True):
     imageformat = os.environ.get('CONFLUENT_IMAGE_PROTOCOL', 'kitty')
@@ -270,8 +284,8 @@ def draw_image(data, width, height, doscale=True):
             rzheight = imgheight
         img = img.resize((rzwidth, rzheight))
         nd = ImageDraw.Draw(nimg)
-        nd.rectangle((1, 1, rzwidth + 2, rzheight + 2), outline='black', width=1)
-        nd.rectangle((0, 0, rzwidth + 3, rzheight + 3), outline='white', width=1)
+        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')
@@ -653,7 +667,7 @@ if options.windowed:
     firstnode=nodes[0]
     nodes.pop(0)
     with open(os.devnull, 'wb') as devnull:
-        xopen=subprocess.Popen(envlist + [confettypath, '-c', '/tmp/controlpath-{0}'.format(firstnode), '-m', '5', 'start', '/nodes/{0}/console/session'.format(firstnode) ] , stdin=devnull)
+        xopen=subprocess.Popen(envlist + [confettypath, '-c', '/tmp/controlpath-{0}'.format(firstnode)] + automation_args + ['-m', '5', 'start', '/nodes/{0}/console/session'.format(firstnode)] , stdin=devnull)
         time.sleep(2)
         s=socket.socket(socket.AF_UNIX)
         winid=''
@@ -725,7 +739,7 @@ if options.windowed:
         else:
             pass 
         with open(os.devnull, 'wb') as devnull:
-            xopen=subprocess.Popen(envlist + [confettypath, '-c', '/tmp/controlpath-{0}'.format(node), '-m', '5', 'start', '/nodes/{0}/console/session'.format(node)] , stdin=devnull)
+            xopen=subprocess.Popen(envlist + [confettypath, '-c', '/tmp/controlpath-{0}'.format(node)] + automation_args + ['-m', '5', 'start', '/nodes/{0}/console/session'.format(node)] , stdin=devnull)
     sys.exit(0)
 #end of wcons
 if options.tile:
@@ -750,18 +764,18 @@ if options.tile:
         panename = '{0}:{1}'.format(sessname, pane)
         if initial:
             initial = False
+            confetty_cmd = [confettypath] + automation_args + ['-m', '5', 'start', '/nodes/{0}/console/session'.format(node)]
             subprocess.call(
                 ['tmux', 'new-session', '-d', '-s',
                  sessname, '-x', '800', '-y',
-                 '800', '{0} -m 5 start /nodes/{1}/console/session'.format(
-                    confettypath, node)])
+                 '800', ' '.join(shlex.quote(arg) for arg in confetty_cmd)])
         else:
             subprocess.call(['tmux', 'select-pane', '-t', sessname])
             subprocess.call(['tmux', 'set-option', '-t', panename, 'pane-border-status', 'top'], stderr=null)
+            confetty_cmd = [confettypath] + automation_args + ['-m', '5', 'start', '/nodes/{0}/console/session'.format(node)]
             subprocess.call(
                     ['tmux', 'split', '-h', '-t', sessname,
-                 '{0} -m 5 start /nodes/{1}/console/session'.format(
-                     confettypath, node)])
+                 ' '.join(shlex.quote(arg) for arg in confetty_cmd)])
             subprocess.call(['tmux', 'select-layout', '-t', sessname, 'tiled'], stdout=null)
             pane += 1
     subprocess.call(['tmux', 'select-pane', '-t', sessname])
@@ -769,5 +783,6 @@ if options.tile:
     if not in_tmux:
         os.execlp('tmux', 'tmux', 'attach', '-t', sessname)
 else:
-    os.execl(confettypath, confettypath, 'start',
-             '/nodes/{0}/console/session'.format(args[0]))
+    execl_args = [confettypath] + automation_args + ['start',
+             '/nodes/{0}/console/session'.format(args[0])]
+    os.execl(confettypath, *execl_args)

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'] = ''
@@ -67,8 +78,10 @@ def main(args):
     ap = argparse.ArgumentParser(description='Deploy OS to nodes')
     ap.add_argument('-c', '--clear', help='Clear any pending deployment action', action='store_true')
     ap.add_argument('-n', '--network', help='Initiate deployment over PXE/HTTP', action='store_true')
+    ap.add_argument('-b', '--bootmethod', help='Specify network boot method (e.g., network, http)', default='network')
     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 +91,7 @@ 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:
@@ -96,27 +109,38 @@ 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')
-            sys.exit(1)
+    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', {}):
@@ -127,8 +151,26 @@ def main(args):
             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)):
@@ -172,7 +214,7 @@ def main(args):
                 print('{0}: {1}{2}'.format(node, profile, armed))
         sys.exit(0)
     if not args.clear and args.network and not args.prepareonly:
-        rc = c.simple_noderange_command(args.noderange, '/boot/nextdevice', 'network',
+        rc = c.simple_noderange_command(args.noderange, '/boot/nextdevice', args.bootmethod,
                                         bootmode='uefi',
                                         persistent=False,
                                         errnodes=errnodes)

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
@@ -59,6 +59,8 @@ argparser = optparse.OptionParser(
           "%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('-p', '--parameterfile', type='string',
+                     help='When updating, use the specified parameter file')
 argparser.add_option('-m', '--maxnodes', type='int',
                      help='When updating, prompt if more than the specified '
                           'number of servers will be affected')
@@ -112,13 +114,30 @@ def update_firmware(session, filename):
     upargs = {'filename': filename}
     if options.backup:
         upargs['bank'] = 'backup'
+    if options.parameterfile:
+        with open(options.parameterfile, 'rb') as pf:
+            pfdata = pf.read()
+        upargs['parameterdata'] = pfdata
     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', ()):
@@ -153,9 +172,13 @@ 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:

confluent_client/bin/nodehealth

@@ -84,13 +84,13 @@ def main():
                 healthexplanations[node] = []
                 for sensor in health[node]['sensors']:
                     explanation = sensor['name'] + ':'
-                    if sensor['value'] is not None:
+                    if sensor.get('value', None) is not None:
                         explanation += str(sensor['value'])
-                        if sensor['units'] is not None:
+                        if sensor.get('units', None) is not None:
                             explanation += sensor['units']
-                        if sensor['states']:
+                        if sensor.get('states', None):
                             explanation += ','
-                    if sensor['states']:
+                    if sensor.get('states', None):
                         explanation += ','.join(sensor['states'])
                     healthexplanations[node].append(explanation)
             if node in healthbynode and node in healthexplanations:

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} '
@@ -58,16 +60,21 @@ def print_mem_info(node, prefix, meminfo):
     elif 'DCPMM' in meminfo['memory_type']:
         memdescfmt = '{0}GB {1} '
         meminfo['module_type'] = 'DCPMM'
+    elif meminfo['memory_type'] == 'HBM':
+        memdescfmt = '{0}GB HBM '
     else:
         print('{0}: {1}: Unrecognized Memory'.format(node, prefix))
         return
     if meminfo.get('ecc', False):
         memdescfmt += 'ECC '
-    capacity = meminfo['capacity_mb'] / 1024
     modtype = meminfo.get('module_type', None)
     if modtype:
         memdescfmt += modtype
-    memdesc = memdescfmt.format(capacity, meminfo['speed'])
+    if meminfo.get('capacity_mb', None):
+        capacity = meminfo['capacity_mb'] // 1024
+        memdesc = memdescfmt.format(capacity, meminfo['speed'])
+    else:
+        memdesc = 'Unspecified Module'
     print('{0}: {1} description: {2}'.format(node, prefix, memdesc))
     print('{0}: {1} manufacturer: {2}'.format(
         node, prefix, meminfo['manufacturer']))

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/nodesensors

@@ -123,7 +123,7 @@ def sensorpass(showout=True, appendtime=False):
                 if 'sensors' not in reading[node]:
                     continue
                 for sensedata in reading[node]['sensors']:
-                    if sensedata['value'] is None and options.skipnumberless:
+                    if sensedata.get('value', None) is None and options.skipnumberless:
                         continue
                     for redundant_state in ('Non-Critical', 'Critical'):
                         try:
@@ -134,17 +134,17 @@ def sensorpass(showout=True, appendtime=False):
                     resultdata[node][sensedata['name']] = sensedata
                     sensorname = sensedata['name']
                     sensorheaders[sensorname] = sensorname
-                    if sensedata['units'] not in (None, u''):
+                    if sensedata.get('units', None) not in (None, u''):
                         sensorheaders[sensorname] += u' ({0})'.format(
                             sensedata['units'])
                     if showout:
-                        if sensedata['value'] is None:
+                        if sensedata.get('value', None) is None:
                             showval = ''
                         elif isinstance(sensedata['value'], float):
                             showval = u' {0} '.format(floatformat(sensedata['value']))
                         else:
-                            showval = u' {0} '.format(sensedata['value'])
-                        if sensedata['units'] not in (None, u''):
+                            showval = u' {0} '.format(sensedata.get('value', ''))
+                        if sensedata.get('units', None) not in (None, u''):
                             showval += sensedata['units']
                         if sensedata.get('health', 'ok') != 'ok':
                             datadescription = [sensedata['health']]

confluent_client/bin/nodesetboot

@@ -32,7 +32,7 @@ if path.startswith('/opt'):
 import confluent.client as client
 
 argparser = optparse.OptionParser(
-    usage='Usage: %prog [options] <noderange> [default|cd|network|setup|hd|usb|floppy]')
+    usage='Usage: %prog [options] <noderange> [default|cd|network|http|setup|hd|usb|floppy]')
 argparser.add_option('-b', '--bios', dest='biosmode',
                      action='store_true', default=False,
                      help='Request BIOS style boot (rather than UEFI)')

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
@@ -281,6 +281,7 @@ class Command(object):
         if maxnodes is None:
             return
         nsize = self.get_noderange_size(noderange)
+        maxnodes = int(maxnodes)
         if nsize > maxnodes:
             if nsize == 1:
                 nodename = list(self.read(
@@ -391,8 +392,13 @@ class Command(object):
             cacert = None
             certreqs = ssl.CERT_NONE
             knownhosts = True
-        self.connection = ssl.wrap_socket(self.connection, ca_certs=cacert,
-                                          cert_reqs=certreqs)
+        tlsctx = ssl.create_default_context()
+        if certreqs == ssl.CERT_NONE:
+            tlsctx.check_hostname = False
+        tlsctx.verify_mode = certreqs
+        if cacert:
+            tlsctx.load_verify_locations(cacert)
+        self.connection = tlsctx.wrap_socket(self.connection, server_hostname=server)
         if knownhosts:
             certdata = self.connection.getpeercert(binary_form=True)
             fingerprint = 'sha512$' + hashlib.sha512(certdata).hexdigest()
@@ -464,8 +470,8 @@ 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/doc/man/imgutil.ronn

@@ -0,0 +1,139 @@
+# imgutil(1) -- Work with confluent OS cloning and diskless images
+
+## SYNOPSIS
+
+`imgutil` `build` [<options>] <scratchdir>  
+`imgutil` `exec` [<options>] <scratchdir> [<cmd>...]  
+`imgutil` `unpack` <profilename> <scratchdir>  
+`imgutil` `pack` [<options>] <scratchdir> <profilename>  
+`imgutil` `capture` <node> <profilename>
+
+## DESCRIPTION
+
+**imgutil** is a utility for creating, managing, and deploying OS images for diskless boot and system cloning in a Confluent environment. It supports building images from scratch, capturing images from running systems, and packing/unpacking diskless profiles.
+
+## COMMANDS
+
+* `build`:
+  Build a new diskless image from scratch in the specified scratch directory.
+
+* `exec`:
+  Start the specified scratch directory as a container and optionally run a command inside it.
+
+* `unpack`:
+  Unpack a diskless image profile to a scratch directory for modification.
+
+* `pack`:
+  Pack a scratch directory into a diskless profile that can be deployed.
+
+* `capture`:
+  Capture an image for cloning from a running system.
+
+## BUILD OPTIONS
+
+* `-r`, `--addrepos` <repository>:
+  Repositories to add in addition to the main source. May be specified multiple times.
+
+* `-p`, `--packagelist` <file>:
+  Filename of package list to replace default pkglist.
+
+* `-a`, `--addpackagelist` <file>:
+  A list of additional packages to include. May be specified multiple times.
+
+* `-s`, `--source` <directory>:
+  Directory to pull installation from, typically a subdirectory of `/var/lib/confluent/distributions`. By default, the repositories for the build system are used. For Ubuntu, this is not supported; the build system repositories are always used.
+
+* `-y`, `--non-interactive`:
+  Avoid prompting for confirmation.
+
+* `-v`, `--volume` <mount>:
+  Directory to make available in the build environment. `-v /` will cause it to be mounted in image as `/run/external/`. `-v /:/run/root` will override the target to be `/run/root`. Something like `/var/lib/repository:-` will cause it to mount to the identical path inside the image. May be specified multiple times.
+
+* <scratchdir>:
+  Directory to build new diskless root in.
+
+## EXEC OPTIONS
+
+* `-v`, `--volume` <mount>:
+  Directory to make available in the build environment. `-v /` will cause it to be mounted in image as `/run/external/`. `-v /:/run/root` will override the target to be `/run/root`. May be specified multiple times.
+
+* <scratchdir>:
+  Directory of an unpacked diskless root.
+
+* <cmd>:
+  Optional command to run (defaults to a shell).
+
+## UNPACK OPTIONS
+
+* <profilename>:
+  The diskless OS profile to unpack.
+
+* <scratchdir>:
+  Directory to extract diskless root to.
+
+## PACK OPTIONS
+
+* `-b`, `--baseprofile` <profile>:
+  Profile to copy extra info from. For example, to make a new version of an existing profile, reference the previous one as baseprofile.
+
+* `-u`, `--unencrypted`:
+  Pack an unencrypted image rather than encrypting.
+
+* <scratchdir>:
+  Directory containing diskless root.
+
+* <profilename>:
+  The desired diskless OS profile name to pack the root into.
+
+## CAPTURE OPTIONS
+
+* <node>:
+  Node to capture image from.
+
+* <profilename>:
+  Profile name for captured image.
+
+## EXAMPLES
+
+Build a diskless image from a distribution:
+
+    imgutil build -s  alma-9.6-x86_64 /tmp/myimage
+
+Execute a shell in an unpacked image:
+
+    imgutil exec /tmp/myimage
+
+Execute a specific command in an image:
+
+    imgutil exec /tmp/myimage /bin/rpm -qa
+
+Unpack an existing profile for modification:
+
+    imgutil unpack myprofile /tmp/myimage
+
+Pack a modified image into a new profile:
+
+    imgutil pack /tmp/myimage myprofile-v2
+
+Capture an image from a running node:
+
+    imgutil capture node01 production-image
+
+## FILES
+
+* `/var/lib/confluent/public/os/`:
+  Default location for OS profiles.
+
+* `/var/lib/confluent/private/os/`:
+  Location for encrypted image keys and private data.
+
+* `/var/lib/confluent/distributions/`:
+  Default location for installation sources.
+
+## SEE ALSO
+
+osdeploy(8)
+
+## AUTHOR
+
+Written for the Confluent project.

confluent_client/doc/man/nodeattrib.ronn.tmpl

@@ -24,6 +24,8 @@ For a full list of attributes, run `nodeattrib <node> all` against a node.
 If `-c` is specified, this will set the nodeattribute to a null value.
 This is different from setting the value to an empty string.
 
+Arbitrary custom attributes can also be created with the `custom.` prefix.
+
 Attributes may be specified by wildcard, for example `net.*switch` will report
 all attributes that begin with `net.` and end with `switch`.
 

confluent_client/doc/man/nodeconfig.ronn

@@ -49,8 +49,7 @@ actually be in effect until a reboot.
 
 * `-r COMPONENT`, `--restoredefault=COMPONENT`:
   Request that the specified component of the targeted nodes will have its
-  configuration reset to default.  Currently the only component implemented
-  is uefi.
+  configuration reset to default.  The component may be "uefi" or "bmc".
 
 * `-m MAXNODES`, `--maxnodes=MAXNODES`:
   Specify a maximum number of nodes to configure, prompting if over 

confluent_client/doc/man/nodefirmware.ronn

@@ -35,7 +35,10 @@ the out of band facilities.  Firmware updates can end in one of three states:
 * `-m MAXNODES`, `--maxnodes=MAXNODES`:
   When updating, prompt if more than the specified number of servers will 
   be affected  
-  
+
+* `-p PARAMETERFILE`, `--paramaterfile=PARAMETERFILE`:
+  For updating, a parameter file to provife along with the update payload
+
 * `-h`, `--help`:
   Show help message and exit    
 

confluent_client/doc/man/nodegroupattrib.ronn.tmpl

@@ -11,7 +11,7 @@ nodegroupattrib(8) -- List or change confluent nodegroup attributes
 
 ## DESCRIPTION
 
-`nodegroupattrip` queries the confluent server to get information about nodes.
+`nodegroupattrib` queries the confluent server to get information about nodes.
 In the simplest form, it simply takes the given group and lists the attributes of that group.
 
 Contrasted with nodeattrib(8), settings managed by nodegroupattrib will be added

confluent_client/doc/man/nodemedia.ronn

@@ -32,6 +32,13 @@ BMCs map a virtual USB device to that url.  Content is loaded on demand, and
 as such that URL is referenced potentially once for every IO operation that
 the host platform attempts.
 
+## NOTES
+
+When doing an attach of an https:// url, you may hit an error if you have not enrolled your certificate authority.
+In a general confluent environment, you can usually address it by:
+  `# for cert in /var/lib/confluent/public/site/tls/*.pem; do nodecertutil s1-s4 installbmccacert $cert; done`  
+  
+
 ## OPTIONS
 
 * `-h`, `--help`:

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/bluefield/bfb-autoinstall

@@ -0,0 +1,74 @@
+#!/usr/bin/python3
+import glob
+import gzip
+import base64
+import os
+import subprocess
+import sys
+import tempfile
+
+def collect_certificates(tmpdir):
+    certdata = ''
+    for cacert in glob.glob(f'{tmpdir}/*.pem'):
+        with open(cacert, 'r') as f:
+            certdata += f.read()
+    return certdata
+
+def embed_certificates(incfg, certdata):
+    if not certdata:
+        raise Exception('No certificates found to embed')
+    incfg = incfg.replace('%CONFLUENTCERTCOLL%', certdata)
+    return incfg
+
+def embed_identity(incfg, identityjson):
+    incfg = incfg.replace('%IDENTJSON%', identityjson)
+    return incfg
+
+def embed_apiclient(incfg, apiclient):
+    with open(apiclient, 'r') as f:
+        apiclientdata = f.read()
+    compressed = gzip.compress(apiclientdata.encode())
+    encoded = base64.b64encode(compressed).decode()
+    incfg = incfg.replace('%APICLIENTZ64%', encoded)
+    return incfg
+
+def embed_data(tmpdir, outfile):
+    templatefile = f'{tmpdir}/bfb.cfg.template'
+    with open(templatefile, 'r') as f:
+        incfg = f.read()
+    
+    certdata = collect_certificates(tmpdir)
+    incfg = embed_certificates(incfg, certdata)
+    
+    with open(f'{tmpdir}/identity.json', 'r') as f:
+        identityjson = f.read()
+    
+    incfg = embed_identity(incfg, identityjson)
+    
+    incfg = embed_apiclient(incfg, f'{tmpdir}/../apiclient')
+    
+    with open(outfile, 'w') as f:
+        f.write(incfg)
+
+def get_identity_json(node):
+    identity_file = f'/var/lib/confluent/private/site/identity_files/{node}.json'
+    try:
+        with open(identity_file, 'r') as f:
+            return f.read()
+    except FileNotFoundError:
+        return None
+
+if __name__ == '__main__':
+    if len(sys.argv) != 4:
+        print("Usage: bfb-autoinstall <node> <bfbfile> <rshim>")
+        sys.exit(1)
+    
+    node = sys.argv[1]
+    bfbfile = sys.argv[2]
+    rshim = sys.argv[3]
+
+    os.chdir(os.path.dirname(os.path.abspath(__file__)))
+    currdir = os.getcwd()
+    tempdir = tempfile.mkdtemp(prefix=f'bfb-autoinstall-{node}-')
+    embed_data(f'{currdir}/{node}', f'{tempdir}/bfb.cfg')
+    subprocess.check_call(['bfb-install', '-b', bfbfile, '-c', f'{tempdir}/bfb.cfg', '-r', rshim])

confluent_osdeploy/bluefield/hostscripts/bfb-autoinstall

@@ -0,0 +1,74 @@
+#!/usr/bin/python3
+import glob
+import gzip
+import base64
+import os
+import subprocess
+import sys
+import tempfile
+
+def collect_certificates(tmpdir):
+    certdata = ''
+    for cacert in glob.glob(f'{tmpdir}/*.pem'):
+        with open(cacert, 'r') as f:
+            certdata += f.read()
+    return certdata
+
+def embed_certificates(incfg, certdata):
+    if not certdata:
+        raise Exception('No certificates found to embed')
+    incfg = incfg.replace('%CONFLUENTCERTCOLL%', certdata)
+    return incfg
+
+def embed_identity(incfg, identityjson):
+    incfg = incfg.replace('%IDENTJSON%', identityjson)
+    return incfg
+
+def embed_apiclient(incfg, apiclient):
+    with open(apiclient, 'r') as f:
+        apiclientdata = f.read()
+    compressed = gzip.compress(apiclientdata.encode())
+    encoded = base64.b64encode(compressed).decode()
+    incfg = incfg.replace('%APICLIENTZ64%', encoded)
+    return incfg
+
+def embed_data(tmpdir, outfile):
+    templatefile = f'{tmpdir}/bfb.cfg.template'
+    with open(templatefile, 'r') as f:
+        incfg = f.read()
+    
+    certdata = collect_certificates(tmpdir)
+    incfg = embed_certificates(incfg, certdata)
+    
+    with open(f'{tmpdir}/identity.json', 'r') as f:
+        identityjson = f.read()
+    
+    incfg = embed_identity(incfg, identityjson)
+    
+    incfg = embed_apiclient(incfg, f'{tmpdir}/../apiclient')
+    
+    with open(outfile, 'w') as f:
+        f.write(incfg)
+
+def get_identity_json(node):
+    identity_file = f'/var/lib/confluent/private/site/identity_files/{node}.json'
+    try:
+        with open(identity_file, 'r') as f:
+            return f.read()
+    except FileNotFoundError:
+        return None
+
+if __name__ == '__main__':
+    if len(sys.argv) != 4:
+        print("Usage: bfb-autoinstall <node> <bfbfile> <rshim>")
+        sys.exit(1)
+    
+    node = sys.argv[1]
+    bfbfile = sys.argv[2]
+    rshim = sys.argv[3]
+
+    os.chdir(os.path.dirname(os.path.abspath(__file__)))
+    currdir = os.getcwd()
+    tempdir = tempfile.mkdtemp(prefix=f'bfb-autoinstall-{node}-')
+    embed_data(f'{currdir}/{node}', f'{tempdir}/bfb.cfg')
+    subprocess.check_call(['bfb-install', '-b', bfbfile, '-c', f'{tempdir}/bfb.cfg', '-r', rshim])

confluent_osdeploy/bluefield/profiles/default/bluefield.cfg.template

@@ -0,0 +1,76 @@
+function bfb_modify_os() {
+     echo 'ubuntu:!' | chpasswd -e
+     mkdir -p /mnt/opt/confluent/bin/
+     cat > /mnt/opt/confluent/bin/confluentbootstrap.sh << 'END_OF_EMBED'
+#!/bin/bash
+     cat > /usr/local/share/ca-certificates/confluent.crt << 'END_OF_CERTS'
+%CONFLUENTCERTCOLL%
+END_OF_CERTS
+     update-ca-certificates
+     mkdir -p /opt/confluent/bin /etc/confluent/
+     cp /usr/local/share/ca-certificates/confluent.crt /etc/confluent/ca.pem
+     cat > /opt/confluent/bin/apiclient.gz.b64 << 'END_OF_CLIENT'
+%APICLIENTZ64%
+END_OF_CLIENT
+     base64 -d /opt/confluent/bin/apiclient.gz.b64 | gunzip > /opt/confluent/bin/apiclient
+     cat > /etc/confluent/ident.json << 'END_OF_IDENT'
+%IDENTJSON%
+END_OF_IDENT
+     python3 /opt/confluent/bin/apiclient -i /etc/confluent/ident.json /confluent-api/self/deploycfg2 > /etc/confluent/confluent.deploycfg
+     PROFILE=$(grep ^profile: /etc/confluent/confluent.deploycfg |awk '{print $2}')
+     ROOTPASS=$(grep ^rootpassword: /etc/confluent/confluent.deploycfg | awk '{print $2}'|grep -v null)
+     if [ -n "$ROOTPASS" ]; then
+          echo root:$ROOTPASS | chpasswd -e
+          echo "ubuntu:$ROOTPASS" | chpasswd -e
+     else
+          echo 'ubuntu:!' | chpasswd -e
+     fi
+     cntmp=$(mktemp -d)
+     cd "$cntmp" || { echo "Failed to cd to temporary directory $cntmp"; exit 1; }
+     touch /etc/confluent/confluent.deploycfg
+     python3 /opt/confluent/bin/apiclient /confluent-public/os/$PROFILE/scripts/confignet > confignet
+     python3 confignet
+     cd -
+     rm -rf "$cntmp"
+     python3 /opt/confluent/bin/apiclient /confluent-public/os/$PROFILE/scripts/functions > /etc/confluent/functions
+     bash /etc/confluent/functions run_remote setupssh
+     for cert in /etc/ssh/ssh*-cert.pub; do
+         if [ -s $cert ]; then
+             echo HostCertificate $cert >> /etc/ssh/sshd_config.d/90-confluent.conf
+         fi
+     done
+     mkdir -p /var/log/confluent
+     chmod 700 /var/log/confluent
+     touch /var/log/confluent/confluent-firstboot.log
+     touch /var/log/confluent/confluent-post.log
+     chmod 600 /var/log/confluent/confluent-post.log
+     chmod 600 /var/log/confluent/confluent-firstboot.log
+     exec >> /var/log/confluent/confluent-post.log
+     exec 2>> /var/log/confluent/confluent-post.log
+     bash /etc/confluent/functions run_remote_python syncfileclient
+     bash /etc/confluent/functions run_remote_parts post.d
+     bash /etc/confluent/functions run_remote_config post.d
+     exec >> /var/log/confluent/confluent-firstboot.log
+     exec 2>> /var/log/confluent/confluent-firstboot.log
+     bash /etc/confluent/functions run_remote_parts firstboot.d
+     bash /etc/confluent/functions run_remote_config firstboot.d
+     python3 /opt/confluent/bin/apiclient /confluent-api/self/updatestatus  -d 'status: staged'
+     python3 /opt/confluent/bin/apiclient /confluent-api/self/updatestatus  -d 'status: complete'
+     systemctl disable confluentbootstrap
+     rm /etc/systemd/system/confluentbootstrap.service
+END_OF_EMBED
+    chmod +x /mnt/opt/confluent/bin/confluentbootstrap.sh
+    cat > /mnt/etc/systemd/system/confluentbootstrap.service << EOS
+[Unit]
+Description=First Boot Process
+Requires=network-online.target
+After=network-online.target
+
+[Service]
+ExecStart=/opt/confluent/bin/confluentbootstrap.sh
+
+[Install]
+WantedBy=multi-user.target
+EOS
+    chroot /mnt systemctl enable confluentbootstrap
+}

confluent_osdeploy/bluefield/profiles/default/nodedeploy-bfb

@@ -0,0 +1,125 @@
+#!/usr/bin/python3
+
+import os
+import sys
+import tempfile
+import glob
+import shutil
+import shlex
+import subprocess
+import select
+
+sys.path.append('/opt/lib/confluent/python')
+
+import confluent.sortutil as sortutil
+import confluent.client as client
+
+
+def prep_outdir(node):
+    tmpdir = tempfile.mkdtemp()
+    for certfile in glob.glob('/var/lib/confluent/public/site/tls/*.pem'):
+        basename = os.path.basename(certfile)
+        destfile = os.path.join(tmpdir, basename)
+        shutil.copy2(certfile, destfile)
+    subprocess.check_call(shlex.split(f'confetty set /nodes/{node}/deployment/ident_image=create'))
+    shutil.copy2(f'/var/lib/confluent/private/identity_files/{node}.json', os.path.join(tmpdir, 'identity.json'))
+    return tmpdir
+
+def exec_bfb_install(host, nodetorshim, bfbfile, installprocs, pipedesc, all, poller):
+    remotedir = subprocess.check_output(shlex.split(f'ssh {host} mktemp -d /tmp/bfb.XXXXXX')).decode().strip()
+    bfbbasename = os.path.basename(bfbfile)
+    subprocess.check_call(shlex.split(f'rsync -avz --info=progress2 {bfbfile} {host}:{remotedir}/{bfbbasename}'))
+    subprocess.check_call(shlex.split(f'rsync -avc --info=progress2 /opt/lib/confluent/osdeploy/bluefield/hostscripts/ {host}:{remotedir}/'))
+    for node in nodetorshim:
+        rshim = nodetorshim[node]
+        nodeoutdir = prep_outdir(node)
+        nodeprofile = subprocess.check_output(shlex.split(f'nodeattrib {node} deployment.pendingprofile')).decode().strip().split(':', 2)[2].strip()
+        shutil.copy2(f'/var/lib/confluent/public/os/{nodeprofile}/bfb.cfg.template', os.path.join(nodeoutdir, 'bfb.cfg.template'))
+        subprocess.check_call(shlex.split(f'rsync -avz {nodeoutdir}/ {host}:{remotedir}/{node}/'))
+        shutil.rmtree(nodeoutdir)
+        run_cmdv(node, shlex.split(f'ssh {host} sh /etc/confluent/functions confluentpython {remotedir}/bfb-autoinstall {node} {remotedir}/{bfbbasename} {rshim}'), all, poller, pipedesc)
+
+
+def run_cmdv(node, cmdv, all, poller, pipedesc):
+    try:
+        nopen = subprocess.Popen(
+            cmdv, stdin=subprocess.DEVNULL, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+    except OSError as e:
+        if e.errno == 2:
+            sys.stderr.write('{0}: Unable to find local executable file "{1}"\n'.format(node, cmdv[0]))
+            return
+        raise
+    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__':
+
+
+    if len(sys.argv) < 3:
+        print(f'Usage: {sys.argv[0]} <host> <bfbfile> <node1:rshim1> [<node2:rshim2> ...]')
+        sys.exit(1)
+
+    host = sys.argv[1]
+    bfbfile = sys.argv[2]
+    nodetorshim = {}
+    for arg in sys.argv[3:]:
+        node, rshim = arg.split(':')
+        nodetorshim[node] = rshim
+
+    installprocs = {}
+    pipedesc = {}
+    all = set()
+    poller = select.epoll()
+
+    exec_bfb_install(host, nodetorshim, bfbfile, installprocs, pipedesc, all, poller)
+    rdy = poller.poll(10)
+    pendingexecs = []
+    exitcode = 0
+    while all:
+        pernodeout = {}
+        for r in rdy:
+            r = r[0]
+            desc = pipedesc[r]
+            r = desc['file']
+            node = desc['node']
+            data = True
+            singlepoller = select.epoll()
+            singlepoller.register(r, select.EPOLLIN)
+            while data and singlepoller.poll(0):
+                data = r.readline()
+                if data:
+                    if desc['type'] == 'stdout':
+                        if node not in pernodeout:
+                            pernodeout[node] = []
+                        pernodeout[node].append(data)
+                    else:
+                        data = client.stringify(data)
+                        sys.stderr.write('{0}: {1}'.format(node, data))
+                        sys.stderr.flush()
+                else:
+                    pop = desc['popen']
+                    ret = pop.poll()
+                    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, poller, pipedesc)
+            singlepoller.close()
+        for node in sortutil.natural_sort(pernodeout):
+            for line in pernodeout[node]:
+                line = client.stringify(line)
+                sys.stdout.write('{0}: {1}'.format(node, line))
+        sys.stdout.flush()
+        if all:
+            rdy = poller.poll(10)
+
+

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
@@ -15,7 +16,13 @@ import sys
 import struct
 import time
 import re
-import json
+import hashlib
+try:
+    import json
+    import hmac
+except ImportError:
+    json = None
+    hmac = None
 
 class InvalidApiKey(Exception):
     pass
@@ -40,40 +47,115 @@ c_crypt.restype = ctypes.c_char_p
 
 
 def get_my_addresses():
-    nlhdrsz = struct.calcsize('IHHII')
-    ifaddrsz = struct.calcsize('BBBBI')
-    # RTM_GETADDR = 22
-    # nlmsghdr struct: u32 len, u16 type, u16 flags, u32 seq, u32 pid
-    nlhdr = struct.pack('IHHII', nlhdrsz + ifaddrsz, 22, 0x301, 0, 0)
-    # ifaddrmsg struct: u8 family, u8 prefixlen, u8 flags, u8 scope, u32 index
-    ifaddrmsg = struct.pack('BBBBI', 0, 0, 0, 0, 0)
-    s = socket.socket(socket.AF_NETLINK, socket.SOCK_RAW, socket.NETLINK_ROUTE)
-    s.bind((0, 0))
-    s.sendall(nlhdr + ifaddrmsg)
     addrs = []
-    while True:
-        pdata = s.recv(65536)
-        v = memoryview(pdata)
-        if struct.unpack('H', v[4:6])[0] == 3:  # netlink done message
-            break
-        while len(v):
-            length, typ = struct.unpack('IH', v[:6])
-            if typ == 20:
-                fam, plen, _, scope, ridx = struct.unpack('BBBBI', v[nlhdrsz:nlhdrsz+ifaddrsz])
-                if scope in (253, 0):
-                    rta = v[nlhdrsz+ifaddrsz:length]
-                    while len(rta):
-                        rtalen, rtatyp = struct.unpack('HH', rta[:4])
-                        if rtalen < 4:
-                            break
-                        if rtatyp == 1:
-                            addrs.append((fam, rta[4:rtalen], plen, ridx))
-                        rta = rta[msg_align(rtalen):]
-            v = v[msg_align(length):]
+    for ifa in get_ifaddrs():
+        if ifa[0] == 'ip':
+            addrs.append((ifa[1], ifa[2], ifa[3]))
     return addrs
 
+def get_mac_addresses():
+    macs = []
+    for ifa in get_ifaddrs():
+        if ifa[0] == 'ETHER':
+            macs.append((ifa[1], ifa[2]))
+    return macs
 
-def scan_confluents():
+def get_ifaddrs():
+    class sockaddr(ctypes.Structure):
+        _fields_ = [
+            ('sa_family', ctypes.c_uint16),
+            ('sa_data', ctypes.c_ubyte * 14),
+        ]
+    
+    class sockaddr_in(ctypes.Structure):
+        _fields_ = [
+            ('sin_family', ctypes.c_uint16),
+            ('sin_port', ctypes.c_uint16),
+            ('sin_addr', ctypes.c_ubyte * 4),
+            ('sin_zero', ctypes.c_ubyte * 8),
+        ]
+    
+    class sockaddr_in6(ctypes.Structure):
+        _fields_ = [
+            ('sin6_family', ctypes.c_uint16),
+            ('sin6_port', ctypes.c_uint16),
+            ('sin6_flowinfo', ctypes.c_uint32),
+            ('sin6_addr', ctypes.c_ubyte * 16),
+            ('sin6_scope_id', ctypes.c_uint32),
+        ]
+    
+    class sockaddr_ll(ctypes.Structure):
+        _fields_ = [
+            ('sll_family', ctypes.c_uint16),
+            ('sll_protocol', ctypes.c_uint16),
+            ('sll_ifindex', ctypes.c_int32),
+            ('sll_hatype', ctypes.c_uint16),
+            ('sll_pkttype', ctypes.c_uint8),
+            ('sll_halen', ctypes.c_uint8),
+            ('sll_addr', ctypes.c_ubyte * 8),
+        ]
+    
+    class ifaddrs(ctypes.Structure):
+        pass
+
+    ifaddrs._fields_ = [
+            ('ifa_next', ctypes.POINTER(ifaddrs)),
+            ('ifa_name', ctypes.c_char_p),
+            ('ifa_flags', ctypes.c_uint),
+            ('ifa_addr', ctypes.POINTER(sockaddr)),
+            ('ifa_netmask', ctypes.POINTER(sockaddr)),
+            ('ifa_ifu', ctypes.POINTER(sockaddr)),
+            ('ifa_data', ctypes.c_void_p),
+        ]
+    
+    libc = ctypes.CDLL(ctypes.util.find_library('c'))
+    libc.getifaddrs.argtypes = [ctypes.POINTER(ctypes.POINTER(ifaddrs))]
+    libc.getifaddrs.restype = ctypes.c_int
+    libc.freeifaddrs.argtypes = [ctypes.POINTER(ifaddrs)]
+    libc.freeifaddrs.restype = None
+    ifap = ctypes.POINTER(ifaddrs)()
+    result = libc.getifaddrs(ctypes.pointer(ifap))
+    if result != 0:
+        return []
+    addresses = []
+    ifa = ifap
+    try:
+        while ifa:
+            if ifa.contents.ifa_addr:
+                family = ifa.contents.ifa_addr.contents.sa_family
+                name = ifa.contents.ifa_name.decode('utf-8') if ifa.contents.ifa_name else None
+                if family in (socket.AF_INET, socket.AF_INET6):
+                    # skip loopback and non-multicast interfaces
+                    if ifa.contents.ifa_flags & 8 or not ifa.contents.ifa_flags & 0x1000:
+                        ifa = ifa.contents.ifa_next
+                        continue
+                if family == socket.AF_INET:
+                    addr_ptr = ctypes.cast(ifa.contents.ifa_addr, ctypes.POINTER(sockaddr_in))
+                    addr_bytes = bytes(addr_ptr.contents.sin_addr)
+                    if_index = socket.if_nametoindex(name) if name else 0
+                    addresses.append(('ip', family, addr_bytes, if_index))
+                elif family == socket.AF_INET6:
+                    addr_ptr = ctypes.cast(ifa.contents.ifa_addr, ctypes.POINTER(sockaddr_in6))
+                    addr_bytes = bytes(addr_ptr.contents.sin6_addr)
+                    scope_id = addr_ptr.contents.sin6_scope_id
+                    addresses.append(('ip', family, addr_bytes, scope_id))          
+                elif family == socket.AF_PACKET:
+                    addr_ptr = ctypes.cast(ifa.contents.ifa_addr, ctypes.POINTER(sockaddr_ll))
+                    halen = addr_ptr.contents.sll_halen
+                    if addr_ptr.contents.sll_hatype in (1, 32) and halen > 0:  # ARPHRD_ETHER or ARPHRD_INFINIBAND
+                        if addr_ptr.contents.sll_hatype == 1 and addr_ptr.contents.sll_addr[0] & 2:  # skip locally administered MACs
+                            ifa = ifa.contents.ifa_next
+                            continue
+                        mac_bytes = bytes(addr_ptr.contents.sll_addr[:halen])
+                        macaddr = ':'.join('{:02x}'.format(b) for b in mac_bytes)
+                        addresses.append(('ETHER', name, macaddr))                
+            ifa = ifa.contents.ifa_next
+    finally:
+        libc.freeifaddrs(ifap)
+    
+    return addresses
+
+def scan_confluents(confuuid=None):
     srvs = {}
     s6 = socket.socket(socket.AF_INET6, socket.SOCK_DGRAM)
     s6.setsockopt(socket.IPPROTO_IPV6, socket.IPV6_V6ONLY, 1)
@@ -85,21 +167,24 @@ 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
+    if not confuuid and os.path.exists('/etc/confluent/confluent.deploycfg'):
+        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
+    if not confuuid and os.path.exists('/confluent_uuid'):
+        with open('/confluent_uuid') as cuuidin:
+            confluentuuid = cuuidin.read().strip()
+            msg += '/confluentuuid=' + confluentuuid
     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()
-            msg += '/mac=' + hwaddr
+    for iface, hwaddr in get_mac_addresses():
+        msg += '/mac=' + hwaddr
     msg = msg.encode('utf8')
     for addr in get_my_addresses():
         if addr[0] == socket.AF_INET6:
@@ -127,6 +212,7 @@ def scan_confluents():
     srvlist = []
     if r:
         r = r[0]
+    nodename = None
     while r:
         for s in r:
             (rsp, peer) = s.recvfrom(9000)
@@ -134,6 +220,7 @@ 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
@@ -145,20 +232,37 @@ def scan_confluents():
             if currip.startswith('fe80::') and '%' not in currip:
                 currip = '{0}%{1}'.format(currip, peer[-1])
             srvs[currip] = current
-            srvlist.append(currip)
+            if currip not in srvlist:
+                srvlist.append(currip)
         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])
@@ -196,7 +300,7 @@ 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()
@@ -205,16 +309,16 @@ 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:
@@ -232,35 +336,43 @@ 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)
@@ -295,15 +407,14 @@ 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()
@@ -343,7 +454,7 @@ 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):
@@ -417,7 +528,7 @@ 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)
@@ -464,7 +575,7 @@ def get_available_nics():
         parts = re.split(r'\s{2,}', line.strip())
         if len(parts) >= 5:
             nic_name = parts[0]
-            nic_status = parts[4]  # "Link Status" este al 5-lea câmp
+            nic_status = parts[4]  # "Link Status" is the 5th field
             available_nics[nic_name] = nic_status
 
     return available_nics
@@ -509,7 +620,7 @@ if __name__ == '__main__':
         try:
             fix_vswitch()
         except Exception as e:
-            print(f"fix_vswitch() error: {e}")
+            print("fix_vswitch() error: {}".format(e))
         sys.argv.remove('-f')
         sys.exit(0)
     usejson = False
@@ -546,8 +657,24 @@ 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')
@@ -559,7 +686,7 @@ 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:
@@ -567,7 +694,7 @@ if __name__ == '__main__':
                 chunk = reader.read(16384)
         sys.exit(0)
 
-    mclient = 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:
@@ -576,4 +703,7 @@ if __name__ == '__main__':
     elif checkonly:
         sys.stdout.write(mclient.check_connections())
     else:
-        sys.stdout.write(mclient.grab_url(sys.argv[1], data).decode())
+        try:
+            sys.stdout.buffer.write(mclient.grab_url(sys.argv[1], data))
+        except AttributeError:
+            sys.stdout.write(mclient.grab_url(sys.argv[1], data))

confluent_osdeploy/common/profile/scripts/autoconsole

@@ -10,11 +10,13 @@
 #   serial port is not reporting DCD, then it doesn't look like a comfortable enough scenario
 
 import fcntl
+import glob
 import os
 import os.path
 import struct
 import subprocess
 import termios
+import platform
 
 
 addrtoname = {
@@ -74,9 +76,8 @@ def fixup_ubuntu_grub_serial():
             grubout.write(grubline + '\n')
     subprocess.check_call(['update-grub'])
 
-def get_serial_config():
-    if not os.path.exists('/sys/firmware/efi'):
-        return None
+
+def get_spcr_config():
     if not os.path.exists('/sys/firmware/acpi/tables/SPCR'):
         return None
     spcr = open("/sys/firmware/acpi/tables/SPCR", "rb")
@@ -99,14 +100,39 @@ def get_serial_config():
         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 get_serial_config():
+    if platform.machine() != 'x86_64':
+        return None  # Trust non-x86 to do the right thing
+    if not os.path.exists('/sys/firmware/efi'):
+        return None  # BIOS might fail at grub output, defer to stock OS behavior
+    retval = get_spcr_config()
+    if retval:
+        return retval
+    firstfound = None
+    numpossible = 0
+    numconnected = 0
+    for serdev in glob.glob('/dev/ttyS*'):
+        ttyf = os.open(serdev, os.O_RDWR | os.O_NOCTTY)
+        try:
+            statusreg = fcntl.ioctl(ttyf, termios.TIOCMGET, '\x00\x00\x00\x00')
+            numpossible += 1
+            if not firstfound:
+                firstfound = serdev
+        except Exception:
+            continue
+        finally:
+            os.close(ttyf)
+        if struct.unpack('<I', statusreg)[0] & termios.TIOCM_CAR:
+            numconnected += 1
+            firstfound = serdev
+    if numpossible == 1 or numconnected == 1: 
+        return { 'tty': firstfound, 'speed': 115200 }
+
 def main():
     autoconscfg = get_serial_config()
-    if not autoconscfg or not autoconscfg['connected']:
+    if not autoconscfg:
         return
     if os.path.exists('/etc/redhat-release'):  # redhat family
         deserialize_grub_rh()

confluent_osdeploy/common/profile/scripts/confignet

@@ -80,14 +80,18 @@ def await_tentative():
         time.sleep(1)
 
 def map_idx_to_name():
-    map = {}
+    map_dict = {}
     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(' '):
+            if 'link/' in line and prevdev and prevdev not in devtype:
+                for word in line.split():
+                    if word.startswith('link/'):
+                        typ = word.split('/')[1]
+                        devtype[prevdev] = typ if typ != 'ether' else 'ethernet'
+                        break # Stop detection after the first type hit
             continue
         idx, iface, rst = line.split(':', 2)
         prevdev = iface.strip()
@@ -99,9 +103,8 @@ def map_idx_to_name():
             pass
         idx = int(idx)
         iface = iface.strip()
-        map[idx] = iface
-    return map, devtype
-
+        map_dict[idx] = iface
+    return map_dict, devtype
 
 def get_interface_name(iname, settings):
     explicitname = settings.get('interface_names', None)
@@ -114,8 +117,10 @@ def get_interface_name(iname, settings):
 class NetplanManager(object):
     def __init__(self, deploycfg):
         self.cfgbydev = {}
+        self.cfgbybond = {}
         self.read_connections()
         self.deploycfg = deploycfg
+        self.teamidx = 0
 
     def read_connections(self):
         for plan in glob.glob('/etc/netplan/*.y*ml'):
@@ -124,29 +129,61 @@ class NetplanManager(object):
                 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]
+            bondinfo = planinfo.get('network', {}).get('bonds', {})
+            for currinfo in (nicinfo, bondinfo):
+                currcfg = self.cfgbydev if currinfo is nicinfo else self.cfgbybond
+                for devname in currinfo:
+                    if devname == 'lo':
+                        continue
+                    if 'gateway4' in currinfo[devname]:
+                        # normalize deprecated syntax on read in
+                        gw4 = currinfo[devname]['gateway4']
+                        del currinfo[devname]['gateway4']
+                        routeinfo = currinfo[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
+                            })
+                        currinfo[devname]['routes'] = routeinfo
+                    currcfg[devname] = currinfo[devname]
     
     def apply_configuration(self, cfg):
         devnames = cfg['interfaces']
-        if len(devnames) != 1:
-            raise Exception('Multi-nic team/bonds not yet supported')
+        if len(devnames) > 1:
+            teammode = cfg['settings'].get('team_mode', None)
+            if not teammode:
+                sys.stderr.write("Warning, multiple interfaces ({0}) without a team_mode, skipping setup\n".format(','.join(devnames)))
+                return
+            if teammode == 'lacp':
+                teammode = '802.3ad'
+            elif teammode == 'activebackup':
+                teammode = 'active-backup'
+            for currdev in self.cfgbybond:
+                for iface in self.cfgbybond[currdev].get('interfaces', []):
+                    if iface in devnames:
+                        break
+                    else:
+                        continue
+                else:
+                    continue
+                # this bond is identified as matching
+                self.cfgbybond[currdev]['interfaces'] = list(devnames)
+                self.cfgbybond[currdev]['parameters']['mode'] = teammode
+                devnames = [currdev]
+                break
+            # no current bond, make a new one
+            connname = cfg['settings'].get('connection_name', None)
+            if not connname:
+                connname = 'bond{0}'.format(self.teamidx)
+                while connname in self.cfgbybond:
+                    self.teamidx += 1
+                    connname = 'bond{0}'.format(self.teamidx)
+            self.cfgbybond[connname] = {'interfaces': list(devnames), 'parameters': {'mode': teammode, 'mii-monitor-interval': 100}}
+            devnames = [connname]
         stgs = cfg['settings']
         needcfgapply = False
         for devname in devnames:
@@ -164,11 +201,14 @@ 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))
             for gwaddr in gws:
-                if gwaddr:
+                if gwaddr and gwaddr != '0.0.0.0':
                     cfgroutes = self.getcfgarrpath([devname, 'routes'])
                     for rinfo in cfgroutes:
                         if rinfo.get('via', None) == gwaddr:
@@ -189,18 +229,56 @@ class NetplanManager(object):
                 if dnsdomain not in currdnsdomain:
                     needcfgwrite = True
                     currdnsdomain.append(dnsdomain)
+            prune_from_cloudinit = []
             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:
+                if devname in self.cfgbydev:
+                    prune_from_cloudinit.append(devname)
+                    newcfg = {'network': {'version': 2, 'ethernets': {devname: self.cfgbydev[devname]}}}
+                    cfgfile = '/etc/netplan/10-{0}-confluentcfg.yaml'.format(devname)
+                elif devname in self.cfgbybond:
+                    newcfg = {'network': {'version': 2, 'bonds': {devname: self.cfgbybond[devname]}}}
+                    for iface in newcfg['network']['bonds'][devname]['interfaces']:
+                        prune_from_cloudinit.append(iface)
+                        with open('/etc/netplan/10-{0}-confluentcfg.yaml'.format(iface), 'w') as planout:
+                            planout.write(yaml.dump({'network': {'version': 2, 'ethernets': {iface: {'dhcp4': False}}}}))
+                    cfgfile = '/etc/netplan/30-{0}-confluentcfg.yaml'.format(devname)
+                with open(cfgfile, 'w') as planout:
                     planout.write(yaml.dump(newcfg))
                 os.umask(oumask)
+        if prune_from_cloudinit:
+            prunecfgs = ['/etc/netplan/00-installer-config.yaml',
+                         '/etc/netplan/50-cloud-init.yaml.dist-subiquity',
+                         '/etc/netplan/50-cloud-init.yaml']
+            prunecfgs.extend(glob.glob('/etc/cloud/cloud.cfg.d/*.cfg'))
+            for defcfg in prunecfgs:
+                if not os.path.exists(defcfg):
+                    continue
+                with open(defcfg, 'r') as cloudinit:
+                    cloudinfo = yaml.safe_load(cloudinit)
+                    if 'network' not in cloudinfo:
+                        continue
+                    for clouddev in list(cloudinfo.get('network', {}).get('ethernets', {})):
+                        if clouddev in prune_from_cloudinit:
+                            del cloudinfo['network']['ethernets'][clouddev]
+                    if not cloudinfo['network'].get('ethernets', {}):
+                        os.remove(defcfg)
+                        if '/etc/cloud/cloud.cfg.d/' in defcfg:
+                            # need to also change datasource, probably
+                            if os.path.exists('/var/lib/cloud/instances/iid-datasource-none/network-config.json'):
+                                os.remove('/var/lib/cloud/instances/iid-datasource-none/network-config.json')
+                    else:
+                        oumask = os.umask(0o77)
+                        with open(defcfg, 'w') as cloudinit:
+                            cloudinit.write(yaml.dump(cloudinfo))
+                        os.umask(oumask)
         if needcfgapply:
+            subprocess.check_call(['netplan', 'generate'])
             subprocess.call(['netplan', 'apply'])
 
     def getcfgarrpath(self, devpath):
-        currptr = self.cfgbydev
+        currptr = self.cfgbybond if devpath[0] in self.cfgbybond else self.cfgbydev
         for k in devpath[:-1]:
             if k not in currptr:
                 currptr[k] = {}
@@ -353,7 +431,7 @@ class NetworkManager(object):
             currteam = deats.get('connection.master', None)
             if currteam == team:
                 return
-            for stg in ('ipv4.dhcp-hostname', 'ipv4.dns', 'ipv6.dns', 'ipv6.dhcp-hostname'):
+            for stg in ('ipv4.dhcp-hostname', 'ipv4.dns', 'ipv6.dns', 'ipv6.dhcp-hostname', 'ipv4.dns-search', 'ipv6.dns-search'):
                 if deats.get(stg, None):
                     bondcfg[stg] = deats[stg]
         if member in self.uuidbyname:
@@ -381,6 +459,8 @@ 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 = []
@@ -406,7 +486,7 @@ 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', 'miimon=100,mode={}'.format(stgs['team_mode'])] + cargs)
@@ -441,7 +521,7 @@ 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])
@@ -455,6 +535,9 @@ 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:
@@ -480,15 +563,16 @@ if __name__ == '__main__':
             continue
         myname = s.getsockname()
         s.close()
+        curridx = None
         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():
+                if myname == addr[1]:
                     curridx = addr[-1]
-        if curridx in doneidxs:
+        if curridx is not None and curridx in doneidxs:
             continue
         for tries in (1, 2, 3):
             try:
@@ -540,7 +624,7 @@ 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()
@@ -562,4 +646,27 @@ if __name__ == '__main__':
     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/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 +55,9 @@ confluentpython $confapiclient /confluent-api/self/nodelist | sed -e 's/^- //' >
 cat /etc/ssh/shosts.equiv > /root/.shosts
 cd -
 rm -rf $TMPDIR
+# ssh may be sshd or ssh, depending
+if systemctl list-unit-files | grep -q '^sshd\.service'; then
+    systemctl try-restart sshd
+elif systemctl list-unit-files | grep -q '^ssh\.service'; then
+    systemctl try-restart ssh
+fi

confluent_osdeploy/confluent_osdeploy-aarch64.spec.tmpl

@@ -26,8 +26,12 @@ mkdir -p opt/confluent/bin
 mkdir -p stateless-bin
 cp -a el8bin/* .
 ln -s el8 el9
-ln -s el8 el10
-for os in rhvh4 el7 genesis el8 suse15 ubuntu20.04 ubuntu22.04 ubuntu24.04 coreos el9 el10; do
+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 debian debian13 ubuntu20.04 ubuntu22.04 ubuntu24.04 ubuntu26.04 coreos el9 el10; do
     mkdir ${os}out
     cd ${os}out
     if [ -d ../${os}bin ]; then 
@@ -43,7 +47,7 @@ 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 el10 ubuntu20.04; do
+for os in el7 el8 suse15 el9 el10 ubuntu20.04 ubuntu22.04 ubuntu24.04 ubuntu26.04; do
     mkdir ${os}disklessout
     cd ${os}disklessout
     if [ -d ../${os}bin ]; then
@@ -81,7 +85,10 @@ 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 el10 genesis suse15 ubuntu20.04 ubuntu22.04 ubuntu24.04 esxi6 esxi7 esxi8 coreos; do
+mkdir -p %{buildroot}/opt/confluent/lib/osdeploy/common
+cp common/initramfs/opt/confluent/bin/apiclient %{buildroot}/opt/confluent/lib/osdeploy/common
+
+for os in rhvh4 el7 el8 el9 el10 debian debian13 genesis suse15 ubuntu20.04 ubuntu22.04 ubuntu24.04 ubuntu26.04 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 debian 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 ubuntu26.04 coreos el9 el10; do
     mkdir ${os}out
     cd ${os}out
     if [ -d ../${os}bin ]; then 
@@ -46,7 +49,7 @@ for os in rhvh4 el7 genesis el8 suse15 debian ubuntu18.04 ubuntu20.04 ubuntu22.0
     mv ../addons.cpio .
     cd ..
 done
-for os in el7 el8 suse15 el9 el10 ubuntu20.04 ubuntu22.04 ubuntu24.04; do
+for os in el7 el8 suse15 el9 el10 ubuntu20.04 ubuntu22.04 ubuntu24.04 ubuntu26.04; do
     mkdir ${os}disklessout
     cd ${os}disklessout
     if [ -d ../${os}bin ]; then
@@ -86,7 +89,7 @@ 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 debian ubuntu18.04 ubuntu22.04 ubuntu24.04 esxi6 esxi7 esxi8 esxi9 coreos; do
+for os in rhvh4 el7 el8 el9 el10 genesis suse15 ubuntu20.04 debian debian13 ubuntu18.04 ubuntu22.04 ubuntu24.04 ubuntu26.04 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

@@ -110,14 +110,22 @@ else
     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
+    nic=
+    while [ -z "$nic" ]; do
+        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 address flush dev $nic
+            nic=""
+        done
+        if [ -z "$nic" ]; then
+            echo "No network interface could be detected, retrying...."
+            sleep 2
         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/ //')

confluent_osdeploy/debian/profiles/default/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))
@@ -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])
@@ -441,7 +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])

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

@@ -2,12 +2,18 @@
 # This script would run in post.d
 #
 export DEBIAN_FRONTEND=noninteractive
-echo "deb [arch=amd64] http://download.proxmox.com/debian/pve bookworm pve-no-subscription" > /etc/apt/sources.list.d/pve-install-repo.list
-wget https://enterprise.proxmox.com/debian/proxmox-release-bookworm.gpg -O /etc/apt/trusted.gpg.d/proxmox-release-bookworm.gpg
-sum=$(sha512sum /etc/apt/trusted.gpg.d/proxmox-release-bookworm.gpg)
-if [ "$sum" -ne "7da6fe34168adc6e479327ba517796d4702fa2f8b4f0a9833f5ea6e6b48f6507a6da403a274fe201595edc86a84463d50383d07f64bdde2e3658108db7d6dc87" ]; then
+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 "$expectedsum" ]; then
 	echo "Mismatch in fingerprint!"
-	rm /etc/apt/trusted.gpg.d/proxmox-release-bookworm.gpg
+	rm /etc/apt/trusted.gpg.d/proxmox-release-$codename.gpg
 	exit 1
 fi
 apt-get update && apt-get -y full-upgrade < /dev/null

confluent_osdeploy/debian/profiles/default/scripts/setupssh

@@ -7,7 +7,6 @@ for pubkey in /etc/ssh/ssh_host*key.pub; do
         continue
     fi
     certfile=${pubkey/.pub/-cert.pub}
-    rm $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
@@ -25,6 +24,7 @@ confluentpython $confapiclient /confluent-public/site/initramfs.tgz -o initramfs
 tar xf initramfs.tgz
 for ca in ssh/*.ca; do
 	LINE=$(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 *' $LINE >> /etc/ssh/ssh_known_hosts.new
@@ -32,6 +32,7 @@ for ca in ssh/*.ca; do
 done
 for pubkey in ssh/*.*pubkey; do
 	LINE=$(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
@@ -41,3 +42,4 @@ 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/el10-diskless/profiles/default/scripts/imageboot.sh

@@ -3,8 +3,8 @@ 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" = "$(getarg confluent_imagemethod)" ]; then
+mkdir -p /mnt/remoteimg /mnt/remote /mnt/overlay /sysroot
+if [ "untethered" = "$(getarg confluent_imagemethod)" -o "uncompressed" = "$(getarg confluent_imagemethod)" ]; then
     mount -t tmpfs untethered /mnt/remoteimg
     curl https://$confluent_whost/confluent-public/os/$confluent_profile/rootimg.sfs -o /mnt/remoteimg/rootimg.sfs
 else
@@ -40,20 +40,53 @@ 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 $memtot > /sys/block/zram0/disksize
-mkfs.xfs /dev/zram0 > /dev/null
-mount -o discard /dev/zram0 /mnt/overlay
-if [ ! -f /tmp/mountparts.sh ]; then
-    mkdir -p /mnt/overlay/upper /mnt/overlay/work
-    mount -t overlay -o upperdir=/mnt/overlay/upper,workdir=/mnt/overlay/work,lowerdir=/mnt/remote disklessroot /sysroot
+if [ ! "uncompressed" = "$(getarg confluent_imagemethod)" ]; then
+    modprobe zram
+    memtot=$(grep ^MemTotal: /proc/meminfo|awk '{print $2}')
+    memtot=$((memtot/2))$(grep ^MemTotal: /proc/meminfo | awk '{print $3'})
+    echo $memtot > /sys/block/zram0/disksize
+    mkfs.xfs /dev/zram0 > /dev/null
+fi
+TETHERED=0
+if [ "untethered" = "$(getarg confluent_imagemethod)" -o "uncompressed" = "$(getarg confluent_imagemethod)" ]; then
+    if [ "untethered" = "$(getarg confluent_imagemethod)" ]; then
+        mount -o discard /dev/zram0 /sysroot
+    else
+        mount -t tmpfs disklessroot /sysroot
+    fi
+    echo -en "Decrypting and extracting root filesystem: 0%\r"
+    srcsz=$(du -sk /mnt/remote | awk '{print $1}')
+    while [ -f /mnt/remoteimg/rootimg.sfs ]; do
+        dstsz=$(du -sk /sysroot | awk '{print $1}')
+        pct=$((dstsz * 100 / srcsz))
+        if [ $pct -gt 99 ]; then
+            pct=99
+        fi
+        echo -en "Decrypting and extracting root filesystem: $pct%\r"
+        sleep 0.25
+    done &
+    cp -ax /mnt/remote/* /sysroot/
+    umount /mnt/remote
+    if [ -e /dev/mapper/cryptimg ]; then
+        dmsetup remove cryptimg
+    fi
+    losetup -d $loopdev
+    rm /mnt/remoteimg/rootimg.sfs
+    umount /mnt/remoteimg
+    wait
+    echo -e "Decrypting and extracting root filesystem: 100%"
 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
+    TETHERED=1
+    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
 fi
 mkdir -p /sysroot/etc/ssh
 mkdir -p /sysroot/etc/confluent
@@ -109,7 +142,7 @@ 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
+chmod 600 /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
@@ -129,4 +162,25 @@ 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)
-exec /opt/confluent/bin/start_root
+if [ $TETHERED -eq 1 ]; then
+    (
+        sleep 86400 &
+        ONBOOTPID=$!
+        mkdir -p /run/confluent
+        echo $ONBOOTPID > /run/confluent/onboot_sleep.pid
+        wait $ONBOOTPID
+        dd if=/mnt/remoteimg/rootimg.sfs iflag=nocache count=0 >& /dev/null
+        rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
+    ) &
+    while [ ! -f /run/confluent/onboot_sleep.pid ]; do
+        sleep 0.1
+    done
+else
+    rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
+fi
+if grep debugssh /proc/cmdline >& /dev/null; then
+    exec /opt/confluent/bin/start_root
+else
+    exec /opt/confluent/bin/start_root -s  # share mount namespace, keep kernel callbacks intact
+fi
+

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

@@ -21,17 +21,10 @@ if grep ^ntpservers: /etc/confluent/confluent.deploycfg > /dev/null; then
 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
+    sed -i '/^[[:space:]]*\(pool\|server\)[[:space:]]/d' /etc/chrony.conf
+    cat /tmp/timeservers >> /etc/chrony.conf
+    systemctl restart chronyd
+    rm -f /tmp/timeservers
 fi
 
 

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/imageboot.sh

@@ -4,10 +4,12 @@ if [[ "$confluent_whost" == *:* ]] && [[ "$confluent_whost" != "["* ]]; then
     confluent_whost="[$confluent_mgr]"
 fi
 mkdir -p /mnt/remoteimg /mnt/remote /mnt/overlay
+TETHERED=0
 if [ "untethered" = "$(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
+    TETHERED=1
     confluent_urls="$confluent_urls https://$confluent_whost/confluent-public/os/$confluent_profile/rootimg.sfs"
     /opt/confluent/bin/urlmount $confluent_urls /mnt/remoteimg
 fi
@@ -130,4 +132,17 @@ ln -s /sysroot/lib/modules/$(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 [ $TETHERED -eq 1 ]; then
+    (
+        sleep 86400 &
+        ONBOOTPID=$!
+        mkdir -p /run/confluent
+        echo $ONBOOTPID > /run/confluent/onboot_sleep.pid
+        wait $ONBOOTPID
+        dd if=/mnt/remoteimg/rootimg.sfs iflag=nocache count=0 >& /dev/null
+    ) &
+    while [ ! -f /run/confluent/onboot_sleep.pid ]; do
+        sleep 0.1
+    done
+fi
 exec /opt/confluent/bin/start_root

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

@@ -27,17 +27,10 @@ if grep ^ntpservers: /etc/confluent/confluent.deploycfg > /dev/null; then
 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
-
+    sed -i '/^[[:space:]]*\(pool\|server\)[[:space:]]/d' /etc/chrony.conf
+    cat /tmp/timeservers >> /etc/chrony.conf
+    systemctl restart chronyd
+    rm -f /tmp/timeservers
 fi
 
 
@@ -59,7 +52,7 @@ 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
@@ -68,5 +61,14 @@ run_remote_parts onboot.d
 # Induce execution of remote configuration, e.g. ansible plays in ansible/onboot.d/
 run_remote_config onboot.d
 
+if [ -f /run/confluent/onboot_sleep.pid ]; then
+    loopdev=$(losetup -j /mnt/remoteimg/rootimg.sfs|cut -d: -f 1)
+    if [ -n "$loopdev" ]; then
+        losetup "$loopdev" --direct-io=on
+    fi
+    sleeppid=$(cat /run/confluent/onboot_sleep.pid)
+    kill "$sleeppid"
+    rm -f /run/confluent/onboot_sleep.pid
+fi
 #curl -X POST -d 'status: booted' -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $confluent_apikey" https://$confluent_mgr/confluent-api/self/updatestatus
 kill $logshowpid

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-diskless/profiles/default/scripts/syncfileclient

@@ -277,7 +277,10 @@ def synchronize():
                             try:
                                 uid = pwd.getpwnam(opts[fname][opt]['name']).pw_uid
                             except KeyError:
-                                uid = opts[fname][opt]['id']
+                                try:
+                                    uid = opts[fname][opt]['id']
+                                except KeyError:
+                                    raise Exception(f"Unable to map owner of {fname}") 
                         elif opt == 'group':
                             try:
                                 gid = grp.getgrnam(opts[fname][opt]['name']).gr_gid

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

@@ -3,10 +3,12 @@ echo -n "" >> /tmp/net.ifaces
 echo -n "" > /tmp/01-autocons.devnode
 BUNDLENAME=/etc/pki/tls/certs/ca-bundle.crt
 if [ ! -e "$BUNDLENAME" ]; then
-    BUNDLENAME=/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem
     mkdir -p /etc/pki/tls/certs
-    ln -s $BUNDLENAME /etc/pki/tls/certs/ca-bundle.crt
-fi
+    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)

confluent_osdeploy/el8/profiles/default/initprofile.sh

@@ -3,7 +3,7 @@ sed -i 's/centos/CentOS/; s/rhel/Red Hat Enterprise Linux/; s/oraclelinux/Oracle
 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
+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

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/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)[-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 and os.path.exists('/tmp/volumegroupmap.yml'):
+            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

@@ -3,8 +3,8 @@ 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" = "$(getarg confluent_imagemethod)" ]; then
+mkdir -p /mnt/remoteimg /mnt/remote /mnt/overlay /sysroot
+if [ "untethered" = "$(getarg confluent_imagemethod)" -o "uncompressed" = "$(getarg confluent_imagemethod)" ]; then
     mount -t tmpfs untethered /mnt/remoteimg
     curl https://$confluent_whost/confluent-public/os/$confluent_profile/rootimg.sfs -o /mnt/remoteimg/rootimg.sfs
 else
@@ -40,20 +40,53 @@ 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 $memtot > /sys/block/zram0/disksize
-mkfs.xfs /dev/zram0 > /dev/null
-mount -o discard /dev/zram0 /mnt/overlay
-if [ ! -f /tmp/mountparts.sh ]; then
-    mkdir -p /mnt/overlay/upper /mnt/overlay/work
-    mount -t overlay -o upperdir=/mnt/overlay/upper,workdir=/mnt/overlay/work,lowerdir=/mnt/remote disklessroot /sysroot
+if [ ! "uncompressed" = "$(getarg confluent_imagemethod)" ]; then
+    modprobe zram
+    memtot=$(grep ^MemTotal: /proc/meminfo|awk '{print $2}')
+    memtot=$((memtot/2))$(grep ^MemTotal: /proc/meminfo | awk '{print $3'})
+    echo $memtot > /sys/block/zram0/disksize
+    mkfs.xfs /dev/zram0 > /dev/null
+fi
+TETHERED=0
+if [ "untethered" = "$(getarg confluent_imagemethod)" -o "uncompressed" = "$(getarg confluent_imagemethod)" ]; then
+    if [ "untethered" = "$(getarg confluent_imagemethod)" ]; then
+        mount -o discard /dev/zram0 /sysroot
+    else
+        mount -t tmpfs disklessroot /sysroot
+    fi
+    echo -en "Decrypting and extracting root filesystem: 0%\r"
+    srcsz=$(du -sk /mnt/remote | awk '{print $1}')
+    while [ -f /mnt/remoteimg/rootimg.sfs ]; do
+        dstsz=$(du -sk /sysroot | awk '{print $1}')
+        pct=$((dstsz * 100 / srcsz))
+        if [ $pct -gt 99 ]; then
+            pct=99
+        fi
+        echo -en "Decrypting and extracting root filesystem: $pct%\r"
+        sleep 0.25
+    done &
+    cp -ax /mnt/remote/* /sysroot/
+    umount /mnt/remote
+    if [ -e /dev/mapper/cryptimg ]; then
+        dmsetup remove cryptimg
+    fi
+    losetup -d $loopdev
+    rm /mnt/remoteimg/rootimg.sfs
+    umount /mnt/remoteimg
+    wait
+    echo -e "Decrypting and extracting root filesystem: 100%"
 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
+    TETHERED=1
+    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
 fi
 mkdir -p /sysroot/etc/ssh
 mkdir -p /sysroot/etc/confluent
@@ -109,8 +142,10 @@ 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
-chroot /sysroot chgrp ssh_keys /etc/ssh/*_key
+if grep ^ssh_keys: /etc/group > /dev/null; then
+    chmod 640 /sysroot/etc/ssh/*_key
+    chroot /sysroot chgrp ssh_keys /etc/ssh/*_key
+fi
 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
@@ -129,5 +164,37 @@ 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
+    debugssh=1
+else
+    debugssh=0
+fi
+if [ $TETHERED -eq 1 ]; then
+    # In tethered mode, the double-caching is useful to get through tricky part of
+    # onboot with confignet. After that, it's excessive cache usage.
+    # Give the onboot script a hook to have us come in and enable directio to the
+    # squashfs and drop the cache of the rootimg so far
+    (
+        sleep 86400 &
+        ONBOOTPID=$!
+        mkdir -p /run/confluent
+        echo $ONBOOTPID > /run/confluent/onboot_sleep.pid
+        wait $ONBOOTPID
+        dd if=/mnt/remoteimg/rootimg.sfs iflag=nocache count=0 >& /dev/null
+        if [ $debugssh -eq 0 ]; then
+            rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
+        fi
+    ) &
+    while [ ! -f /run/confluent/onboot_sleep.pid ]; do
+        sleep 0.1
+    done
+elif [ $debugssh -eq 0 ]; then
+    rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
+fi
+if grep debugssh /proc/cmdline >& /dev/null; then
+    exec /opt/confluent/bin/start_root
+else
+    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

@@ -26,17 +26,10 @@ if grep ^ntpservers: /etc/confluent/confluent.deploycfg > /dev/null; then
 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
+    sed -i '/^[[:space:]]*\(pool\|server\)[[:space:]]/d' /etc/chrony.conf
+    cat /tmp/timeservers >> /etc/chrony.conf
+    systemctl restart chronyd
+    rm -f /tmp/timeservers
 fi
 
 export nodename confluent_mgr confluent_profile
@@ -53,7 +46,7 @@ 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
@@ -62,5 +55,15 @@ run_remote_parts onboot.d
 # Induce execution of remote configuration, e.g. ansible plays in ansible/onboot.d/
 run_remote_config onboot.d
 
+if [ -f /run/confluent/onboot_sleep.pid ]; then
+    loopdev=$(losetup -j /mnt/remoteimg/rootimg.sfs|cut -d: -f 1)
+    if [ -n "$loopdev" ]; then
+        losetup "$loopdev" --direct-io=on
+    fi
+    sleeppid=$(cat /run/confluent/onboot_sleep.pid)
+    kill "$sleeppid"
+    rm -f /run/confluent/onboot_sleep.pid
+fi
+
 #curl -X POST -d 'status: booted' -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $confluent_apikey" https://$confluent_mgr/confluent-api/self/updatestatus
 kill $logshowpid

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

@@ -62,8 +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}') > $hmackeyfile
@@ -73,6 +73,20 @@ if [ -e /tmp/confluentident/cnflnt.yml ]; then
     hmacfile=$(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

confluent_osdeploy/esxi7/profiles/hypervisor/scripts/getinstalldisk

@@ -0,0 +1,148 @@
+#!/usr/bin/python3
+import subprocess
+import os
+
+class SilentException(Exception):
+    pass
+
+class DiskInfo(object):
+    def __init__(self, devname, devinfo):
+        self.name = devname
+        self.path = '/dev/' + devname
+        self.wwn = None
+        self.model = devinfo.get('model', 'Unknown')
+        self.driver = devinfo.get('adapter_driver', 'Unknown')
+        self.size = devinfo.get('size', 0)  # in MiB
+        if not devinfo.get('is_local', False):
+            raise SilentException("Not local")
+        if devinfo.get('is_removable', False):
+            raise SilentException("Removable")
+        if devinfo.get('is_usb', False):
+            raise SilentException("USB device")
+        if devinfo.get('type', '').lower() in ('cd-rom',):
+            raise SilentException("CD-ROM device")
+        if self.size < 2048:
+            raise SilentException("Too small")
+
+        
+
+
+    @property
+    def priority(self):
+        if self.model.lower() in ('m.2 nvme 2-bay raid kit', 'thinksystem_m.2_vd', 'thinksystem m.2', 'thinksystem_m.2'):
+            return 0
+        if self.driver == 'vmw_ahci':
+            return 2
+        if self.driver == 'nvme_pcie':
+            return 3
+        return 99
+
+    def __repr__(self):
+        return repr({
+            'name': self.name,
+            'path': self.path,
+            'wwn': self.wwn,
+            'driver': self.driver,
+            'size': self.size,
+            'model': self.model,
+        })
+
+def list_disks():
+    current_dev = None
+    disks = {}
+    devlist = subprocess.check_output(['localcli', 'storage', 'core', 'device', 'list'])
+    if not isinstance(devlist, str):
+        devlist = devlist.decode('utf8')
+    devbyadp = {}
+    for line in devlist.split('\n'):
+        if not line.strip():
+            continue
+        if not line.startswith(' '):
+            current_dev = line.rsplit(':', 1)[0]
+            if current_dev not in disks:
+                disks[current_dev] = {}
+        elif current_dev:
+            if '   Model:' in line:
+                disks[current_dev]['model'] = ' '.join(line.split()[1:])
+            elif '   Driver:' in line:
+                disks[current_dev]['driver'] = ' '.join(line.split()[1:])
+            elif '   Is Local:' in line:
+                disks[current_dev]['is_local'] = ' '.join(line.split()[2:]).lower() == 'true'
+            elif '   Is Removable:' in line:
+                disks[current_dev]['is_removable'] = ' '.join(line.split()[2:]).lower() == 'true'
+            elif '   Size:' in line:  # in MiB
+                disks[current_dev]['size'] = int(line.split()[1])
+            elif '   Is SSD:' in line:
+                disks[current_dev]['is_ssd'] = ' '.join(line.split()[2:]).lower() == 'true'
+            elif '   Is USB:' in line:
+                disks[current_dev]['is_usb'] = ' '.join(line.split()[2:]).lower() == 'true'
+            elif '   Is Removable:' in line:
+                disks[current_dev]['is_removable'] = ' '.join(line.split()[2:]).lower() == 'true'
+            elif 'Device Type:' in line:
+                disks[current_dev]['type'] = ' '.join(line.split()[2:])
+    for dev in disks:
+        pathlist = subprocess.check_output(['localcli', 'storage', 'core', 'path', 'list', '--device', dev])
+        if not isinstance(pathlist, str):
+            pathlist = pathlist.decode('utf8')
+        for line in pathlist.split('\n'):
+            if not line.strip():
+                continue
+            if not line.startswith(' '):
+                continue
+            if '   Adapter Identifier:' in line:
+                adpname = ' '.join(line.split()[2:])
+                disks[dev]['adapter_id'] = adpname
+            elif '   Adapter:' in line:
+                adp = ' '.join(line.split()[1:])
+                disks[dev]['adapter'] = adp
+                devbyadp.setdefault(adp, []).append(dev)
+    adapterlist = subprocess.check_output(['localcli', 'storage', 'core', 'adapter', 'list'])
+    if not isinstance(adapterlist, str):
+        adapterlist = adapterlist.decode('utf8')
+    driverbyadp = {}
+    linenum = 0
+    for line in adapterlist.split('\n'):
+        linenum += 1
+        if not line.strip():
+            continue
+        if linenum < 3:
+            continue
+        parts = line.split()
+        if len(parts) < 2:
+            continue
+        adp = parts[0]
+        driver = parts[1]
+        driverbyadp[adp] = driver
+    for adp in devbyadp:
+        driver = driverbyadp.get(adp, 'Unknown')
+        for dev in devbyadp[adp]:
+            disks[dev]['adapter_driver'] = driver
+    return disks
+        
+def main():
+    disks = []
+    try:
+        alldisks = list_disks()
+    except Exception as e:
+        print("Error listing disks: {0}".format(str(e)))
+        alldisks = {}
+    for disk in alldisks:
+        try:
+            disks.append(DiskInfo(disk, alldisks[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, x.size])]
+    if nd:
+        with open('/tmp/storagecfg', 'w') as sc:
+            sc.write(f'clearpart --all --drives={nd[0]} --overwritevmfs\n')
+            sc.write(f'install --drive={nd[0]} --overwritevmfs\n')
+    else:
+        with open('/tmp/storagecfg', 'w') as sc:
+            sc.write(f'clearpart --firstdisk --overwritevmfs\n')
+            sc.write(f'install --firstdisk --overwritevmfs\n')
+
+
+if __name__ == '__main__':
+    main()

confluent_osdeploy/esxi7/profiles/hypervisor/scripts/makeksnet

@@ -45,7 +45,10 @@ try:
         cfg['ipv4_gateway'] = ncfg['ipv4_gateway']
 except Exception:
     pass
-netline = 'network --hostname={0} --bootproto={1}'.format(nodename, cfg['ipv4_method'])
+if cfg['ipv4_method'] == 'static':
+    netline = 'network --hostname={0} --bootproto={1}'.format(nodename, cfg['ipv4_method'])
+else:
+    netline = 'network --bootproto=dhcp'
 if vmnic:
     netline += ' --device={0}'.format(vmnic)
 if cfg['ipv4_method'] == 'static':

confluent_osdeploy/esxi7/profiles/hypervisor/scripts/modinstall

@@ -1,9 +1,12 @@
 #!/bin/sh
 mv /etc/confluent/confluent.deploycfg /etc/confluent/confluent.newdeploycfg
 /opt/confluent/bin/apiclient /confluent-public/os/$profile/scripts/makeksnet >> /tmp/makeksnet
+/opt/confluent/bin/apiclient /confluent-public/os/$profile/scripts/getinstalldisk >> /tmp/getinstalldisk
 mv /etc/confluent/confluent.newdeploycfg /etc/confluent/confluent.deploycfg
 chmod +x /tmp/makeksnet
 /tmp/makeksnet > /tmp/ksnet
+localcli system hostname set --host $node
+python3 /tmp/getinstalldisk
 rootpw=$(grep ^rootpassword: /etc/confluent/confluent.deploycfg|sed -e 's/^rootpassword: //')
 echo rootpw --iscrypted $rootpw > /tmp/rootpw
 export BOOT_CMDLINE=ks=/etc/confluent/ks.cfg

confluent_osdeploy/genesis/initramfs/opt/confluent/bin/rungenesis

@@ -56,7 +56,7 @@ cat /tls/*.pem > /etc/pki/tls/certs/ca-bundle.crt
 TRIES=0
 touch /etc/confluent/confluent.info
 TRIES=5
-echo -n "Waitiing for disks..."
+echo -n "Waiting for disks..."
 while [ ! -e /dev/disk/by-label ] && [ $TRIES -gt 0 ]; do
     sleep 1
     TRIES=$((TRIES - 1))

confluent_osdeploy/rhvh4/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/suse15-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/suse15/profiles/hpc/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/suse15/profiles/hpc/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/suse15/profiles/server/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/suse15/profiles/server/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/ubuntu18.04/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/ubuntu18.04/profiles/default/syncfiles

@@ -0,0 +1,29 @@
+# It is advised to avoid /var/lib/confluent/public as a source for syncing.  /var/lib/confluent/public
+# is served without authentication and thus any sensitive content would be a risk. If wanting to host
+# syncfiles on a common share, it is suggested to have /var/lib/confluent be the share and use some other
+# subdirectory other than public.
+#
+# Syncing is performed as the 'confluent' user, so all source files must be accessible by the confluent user.
+#
+# This file lists files to synchronize or merge to the deployed systems from the deployment server
+# To specify taking /some/path/hosts on the deployment server and duplicating it to /etc/hosts:
+# Note particularly the use of '->' to delineate source from target.
+# /some/path/hosts -> /etc/hosts
+
+# If wanting to simply use the same path for source and destinaiton, the -> may be skipped:
+# /etc/hosts
+
+# More function is available, for example to limit the entry to run only on n1 through n8, and to set
+# owner, group, and permissions in octal notation:
+# /example/source -> n1-n8:/etc/target (owner=root,group=root,permissions=600)
+
+# Entries under APPENDONCE: will be added to specified target, only if the target does not already
+# contain the data in the source already in its entirety. This allows append in a fashion that
+# is friendly to being run repeatedly
+
+# Entries under MERGE: will attempt to be intelligently merged. This supports /etc/group and /etc/passwd
+# Any supporting entries in /etc/shadow or /etc/gshadow are added automatically, with password disabled
+# It also will not inject 'system' ids (under 1,000 usually) as those tend to be local and rpm managed.
+MERGE:
+# /etc/passwd
+# /etc/group

confluent_osdeploy/ubuntu20.04-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/ubuntu20.04-diskless/profiles/default/scripts/imageboot.sh

@@ -8,7 +8,9 @@ for addr in $(grep ^MANAGER: /etc/confluent/confluent.info|awk '{print $2}'|sed
     fi
 done
 mkdir -p /mnt/remoteimg /mnt/remote /mnt/overlay
-if grep confluent_imagemethod=untethered /proc/cmdline > /dev/null; then
+TETHERED=1
+if grep -q confluent_imagemethod=untethered /proc/cmdline || grep -q confluent_imagemethod=uncompressed /proc/cmdline; then
+    TETHERED=0
     mount -t tmpfs untethered /mnt/remoteimg
     curl https://$confluent_mgr/confluent-public/os/$confluent_profile/rootimg.sfs -o /mnt/remoteimg/rootimg.sfs
 else
@@ -44,15 +46,46 @@ 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 $memtot > /sys/block/zram0/disksize
-modprobe xfs
 mkdir /sysroot
-mkfs.xfs /dev/zram0 > /dev/null
-mount -o discard /dev/zram0 /mnt/overlay
-if [ ! -f /tmp/mountparts.sh ]; then
+if ! grep -q confluent_imagemethod=uncompressed /proc/cmdline; then
+    modprobe zram
+    memtot=$(grep ^MemTotal: /proc/meminfo|awk '{print $2}')
+    memtot=$((memtot/2))$(grep ^MemTotal: /proc/meminfo | awk '{print $3'})
+    echo $memtot > /sys/block/zram0/disksize
+    modprobe xfs
+    mkfs.xfs /dev/zram0 > /dev/null
+    if [ "$TETHERED" = 1 ]; then
+         mount -o discard /dev/zram0 /mnt/overlay
+    else
+         mount -o discard /dev/zram0 /sysroot
+    fi
+    mount -o discard /dev/zram0 /mnt/overlay
+elif grep -q confluent_imagemethod=uncompressed /proc/cmdline; then
+    mount -t tmpfs disklessroot /sysroot
+fi
+if [ "$TETHERED" = 0 ]; then
+    echo -en "Decrypting and extracting root filesystem: 0%\r"
+        srcsz=$(du -sk /mnt/remote | awk '{print $1}')
+        while [ -f /mnt/remoteimg/rootimg.sfs ]; do
+        dstsz=$(du -sk /sysroot | awk '{print $1}')
+        pct=$((dstsz * 100 / srcsz))
+        if [ $pct -gt 99 ]; then
+            pct=99
+        fi
+        echo -en "Decrypting and extracting root filesystem: $pct%\r"
+        sleep 0.25
+    done &
+    cp -a /mnt/remote/* /sysroot/
+    umount /mnt/remote
+    if [ -e /dev/mapper/cryptimg ]; then
+        dmsetup remove cryptimg
+    fi
+    losetup -d $loopdev
+    rm /mnt/remoteimg/rootimg.sfs
+    umount /mnt/remoteimg
+    wait
+    echo -e "Decrypting and extracting root filesystem: 100%"
+elif [ ! -f /tmp/mountparts.sh ]; then
     mkdir -p /mnt/overlay/upper /mnt/overlay/work
     mount -t overlay -o upperdir=/mnt/overlay/upper,workdir=/mnt/overlay/work,lowerdir=/mnt/remote disklessroot /sysroot
 else
@@ -68,6 +101,7 @@ cp /root/.ssh/* /sysroot/root/.ssh
 chmod 700 /sysroot/root/.ssh
 cp /etc/confluent/* /sysroot/etc/confluent/
 cp /etc/ssh/*key* /sysroot/etc/ssh/
+cp /tls/* /sysroot/etc/ssl/certs
 for pubkey in /etc/ssh/ssh_host*key.pub; do
     certfile=${pubkey/.pub/-cert.pub}
     privfile=${pubkey%.pub}

confluent_osdeploy/ubuntu20.04-diskless/profiles/default/syncfiles

@@ -0,0 +1,29 @@
+# It is advised to avoid /var/lib/confluent/public as a source for syncing.  /var/lib/confluent/public
+# is served without authentication and thus any sensitive content would be a risk. If wanting to host
+# syncfiles on a common share, it is suggested to have /var/lib/confluent be the share and use some other
+# subdirectory other than public.
+#
+# Syncing is performed as the 'confluent' user, so all source files must be accessible by the confluent user.
+#
+# This file lists files to synchronize or merge to the deployed systems from the deployment server
+# To specify taking /some/path/hosts on the deployment server and duplicating it to /etc/hosts:
+# Note particularly the use of '->' to delineate source from target.
+# /some/path/hosts -> /etc/hosts
+
+# If wanting to simply use the same path for source and destinaiton, the -> may be skipped:
+# /etc/hosts
+
+# More function is available, for example to limit the entry to run only on n1 through n8, and to set
+# owner, group, and permissions in octal notation:
+# /example/source -> n1-n8:/etc/target (owner=root,group=root,permissions=600)
+
+# Entries under APPENDONCE: will be added to specified target, only if the target does not already
+# contain the data in the source already in its entirety. This allows append in a fashion that
+# is friendly to being run repeatedly
+
+# Entries under MERGE: will attempt to be intelligently merged. This supports /etc/group and /etc/passwd
+# Any supporting entries in /etc/shadow or /etc/gshadow are added automatically, with password disabled
+# It also will not inject 'system' ids (under 1,000 usually) as those tend to be local and rpm managed.
+MERGE:
+# /etc/passwd
+# /etc/group

confluent_osdeploy/ubuntu20.04/initramfs/scripts/init-premount/confluent

@@ -52,13 +52,40 @@ while ! grep NODENAME /custom-installation/confluent/confluent.info; do
             ipconfig -d $MYIP::$MYGW:$MYNM::$NIC
             echo $NIC > /tmp/autodetectnic
         else
-            configure_networking
-            for dsrv in $deploysrvs; do
-                if openssl s_client -connect $dsrv:443 > /dev/null 2>&1; then
-                    deploysrvs=$dsrv
-                    break
+            rmmod cdc_ether 2> /dev/null
+            while [ ! -f /run/confirmednic ]; do
+                ALLNETDEVS=$(ip a|grep LOWER_UP|grep MULTICAST|awk '{print $2}'|sed -e s/://)
+                rm -rf /run/net* /run/dhcpcd /var/lib/dhcpcd
+                for dev in $(ip a|grep MULTICAST|awk '{print $2}'|sed -e s/://); do
+                   ip a flush $dev
+                   echo 1 > /proc/sys/net/ipv6/conf/$dev/addr_gen_mode
+                   echo 0 > /proc/sys/net/ipv6/conf/$dev/addr_gen_mode
+                done
+                unset DEVICE DEVICE6 IP IP6 dev
+                [ -z "$1" ] || DEVICE=$1
+                shift
+                configure_networking
+                echo $DEVICE > /tmp/autodetectnic
+                for dsrv in $deploysrvs; do
+                    (if openssl s_client -connect $dsrv:443 > /dev/null 2>&1; then
+                        echo $dsrv > /run/confirmednic
+                        break
+                    fi) &
+                    chkpid=$!
+                    ( sleep 10 && kill $chkpid ) &
+                    timeoutpid=$!
+                    wait $chkpid
+                    kill $timeoutpid 2> /dev/null
+                    unset chkpid timeoutpid
+                done
+                if [ ! -f /run/confirmednic ]; then
+                    echo "No connectivity to deployment servers, retrying..."
+                    [ -z "$1" ] && set -- $ALLNETDEVS
                 fi
             done
+            deploysrvs=$(cat /run/confirmednic)
+            rm /run/confirmednic
+
         fi
         MGR=$deploysrvs
         NODENAME=$(grep ^nodename: /tmp/idntmnt/cnflnt.yml | awk '{print $2}')