Implement deployment.storage

This is an attribute for a node to indicate preferences for storage.

For now, 'm2' policy will hit m.2 and mirroring kits.

confluent_osdeploy/common/profile/scripts/getinstalldisk

@@ -1,6 +1,9 @@
 #!/usr/bin/python3
+import json
 import subprocess
 import os
+import shutil
+
 
 class SilentException(Exception):
     pass
@@ -71,6 +74,10 @@ class DiskInfo(object):
                 self.subsystype = v.replace('"', '')
             elif k == 'ATTR{ro}' and v == '"1"':
                 raise Exception("Device is read-only")
+            elif k == 'ATTRS{removable}' and v == '"1"':
+                raise Exception("Device is removable")
+            elif k == 'ATTRS{address}':
+                self.busaddr = v.replace('"', '')
         if not self.driver and 'imsm' not in self.mdcontainer and self.subsystype != 'nvm':
             raise Exception("No driver detected")
         if self.driver == 'sr':
@@ -106,6 +113,114 @@ class DiskInfo(object):
         })
 
 
+def get_dmi_type9_info():
+    dmidecode = shutil.which('dmidecode')
+    if not dmidecode:
+        return {}
+    try:
+        dmio = subprocess.check_output(
+            [dmidecode, '--type', '9'], stderr=subprocess.DEVNULL)
+    except Exception:
+        return {}
+    if not isinstance(dmio, str):
+        dmio = dmio.decode('utf8', errors='ignore')
+    slots = {}
+    currentslot = None
+    for line in dmio.split('\n'):
+        stripped = line.strip()
+        if not stripped:
+            if currentslot:
+                busaddr = currentslot.get('bus_address', None)
+                if busaddr:
+                    slots[busaddr] = currentslot
+                currentslot = None
+            continue
+        if stripped.startswith('Handle ') and ', DMI type 9,' in stripped:
+            if currentslot:
+                busaddr = currentslot.get('bus_address', None)
+                if busaddr:
+                    slots[busaddr] = currentslot
+            currentslot = {'handle': stripped.split(',', 1)[0].split(' ', 1)[1]}
+            continue
+        if currentslot is None:
+            continue
+        if ':' not in stripped:
+            continue
+        key, val = stripped.split(':', 1)
+        key = key.strip().lower().replace(' ', '_')
+        currentslot[key] = val.strip()
+    if currentslot:
+        busaddr = currentslot.get('bus_address', None)
+        if busaddr:
+            slots[busaddr] = currentslot
+    return slots
+
+def get_m2_slot_info():
+    slots = get_dmi_type9_info()
+    m2slots = {}
+    for slot in slots.values():
+        if 'm.2 socket' not in slot.get('type', '').lower():
+            continue
+        busaddr = slot.get('bus_address', None)
+        if not busaddr:
+            continue
+        m2slots[busaddr] = slot
+    return m2slots
+
+
+def get_deployment_storage():
+    apiclient = None
+    for clientpath in ('/opt/confluent/bin/apiclient', '/etc/confluent/apiclient'):
+        if os.path.exists(clientpath):
+            apiclient = clientpath
+            break
+    if not apiclient:
+        return None
+    for attriburl in ('/confluent-api/self/myattrib', '/confluent-api/self/myattribs'):
+        try:
+            attribs = subprocess.check_output(
+                ['python3', apiclient, attriburl, '-j'], stderr=subprocess.DEVNULL)
+            if not isinstance(attribs, str):
+                attribs = attribs.decode('utf8')
+            attribs = json.loads(attribs)
+            if not isinstance(attribs, dict):
+                continue
+            storage = attribs.get('deployment.storage', None)
+            if isinstance(storage, str) and storage.strip():
+                return storage.strip()
+        except Exception:
+            continue
+    return None
+
+
+def filter_deployment_storage(disks, storagespec, slotinfo=None):
+    if not storagespec:
+        return disks
+    spec = storagespec.strip().lower()
+    for disk in list(disks):
+        if spec == 'm2':
+            if disk.model.lower() in PRIORITY_MODELS:
+                continue
+            if not getattr(disk, 'busaddr', None):
+                disks.remove(disk)
+                continue
+            if slotinfo is None:
+                slotinfo = get_m2_slot_info()
+            if disk.busaddr not in slotinfo:
+                disks.remove(disk)
+    
+        elif spec.startswith('/dev/'):
+            bspec = os.path.basename(spec)
+            if disk.name.lower() == bspec:
+                return [disk]
+        else:
+            raise Exception("Unknown deployment.storage specification: {0}".format(storagespec))
+    
+    if not disks:
+        raise Exception("No disks match deployment.storage specification: {0}".format(storagespec))
+    return disks
+
+
 def main():
     disks = []
     for disk in sorted(os.listdir('/sys/class/block')):
@@ -116,9 +231,13 @@ def main():
             pass
         except Exception as e:
             print("Skipping {0}: {1}".format(disk, str(e)))
+    slotinfo = get_m2_slot_info()
+    disks = filter_deployment_storage(disks, get_deployment_storage(), slotinfo=slotinfo)
     nd = [x.name for x in sorted(disks, key=lambda x: [x.priority, x.size])]
     if nd:
         open('/tmp/installdisk', 'w').write(nd[0])
+    else:
+        raise Exception("No suitable install disk found")
 
 if __name__ == '__main__':
     main()

confluent_osdeploy/esxi7/profiles/hypervisor/scripts/getinstalldisk

@@ -1,12 +1,88 @@
 #!/usr/bin/python3
 import subprocess
+import json
 import os
 
 class SilentException(Exception):
     pass
 
+def list_slots():
+    slots = {}
+    current = None
+    dump = subprocess.check_output(['smbiosDump'])
+    if not isinstance(dump, str):
+        dump = dump.decode('utf8')
+    for line in dump.split('\n'):
+        if 'System Slot (Type 9):' in line:
+            current = {}
+        elif current is not None:
+            stripped = line.strip()
+            if stripped.startswith('Designation:'):
+                current['designation'] = stripped.split(':', 1)[1].strip().strip('"')
+            if stripped.startswith('Type:'):
+                current['type_label'] = stripped.split(':', 1)[1].strip()
+                current['type_number'] = int(stripped.split(':', 1)[1].strip().split(' ')[0], 16)
+                # 0x14 through 0x17 are M.2 slots, inclusively
+                if current['type_number'] in (0x14, 0x15, 0x16, 0x17):
+                    current['is_m2'] = True
+                else:
+                    current['is_m2'] = False
+            elif stripped.startswith('PCI Address:'):
+                addr = stripped.split(':', 1)[1].strip()
+                current['pci_address'] = addr
+                slots[addr] = current
+                current = None
+    return slots
+
+def list_m2_slots():
+    slots = list_slots()
+    m2slots = {}
+    for addr in slots:
+        slot = slots[addr]
+        if slot.get('is_m2', False):
+            m2slots[addr] = slot
+    return m2slots
+
+def get_deployment_storage():
+    apiclient = None
+    for clientpath in ('/opt/confluent/bin/apiclient', '/etc/confluent/apiclient'):
+        if os.path.exists(clientpath):
+            apiclient = clientpath
+            break
+    if not apiclient:
+        return None
+    for attriburl in ('/confluent-api/self/myattrib', '/confluent-api/self/myattribs'):
+        try:
+            attribs = subprocess.check_output(
+                ['python3', apiclient, attriburl, '-j'], stderr=subprocess.DEVNULL)
+            if not isinstance(attribs, str):
+                attribs = attribs.decode('utf8')
+            attribs = json.loads(attribs)
+            if not isinstance(attribs, dict):
+                continue
+            storage = attribs.get('deployment.storage', None)
+            if isinstance(storage, str) and storage.strip():
+                return storage.strip()
+        except Exception:
+            continue
+    return None
+
+# List of models that are generally used for OS install disk (generally 2-disk mirroring)
+PRIORITY_MODELS = ('m.2 nvme 2-bay raid kit', 
+                   'thinksystem_m.2_vd',
+                   'thinksystem m.2',
+                   'thinksystem_m.2',
+                   'b540p-2hs',  # 7mm or M.2 in rear
+                   'b540d-2hs',  # M.2 in E3.S or 2.5inch
+                   'b550d-2hs',  # Front M.2
+                   'b550p-2hs',  # Rear M.2
+                   'raid b540p-2hs',
+                   'raid b540d-2hs',
+                   'raid b550d-2hs',
+                   'raid b550p-2hs')
+
 class DiskInfo(object):
-    def __init__(self, devname, devinfo):
+    def __init__(self, devname, devinfo, storagespec=None):
         self.name = devname
         self.path = '/dev/' + devname
         self.wwn = None
@@ -23,13 +99,16 @@ class DiskInfo(object):
             raise SilentException("CD-ROM device")
         if self.size < 2048:
             raise SilentException("Too small")
-
-        
+        if storagespec and storagespec == 'm2':
+            if self.model.lower() not in PRIORITY_MODELS and not devinfo.get('m2_slot', None):
+                raise SilentException("Not an M.2 disk")
+               
+            
 
 
     @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'):
+        if self.model.lower() in PRIORITY_MODELS:
             return 0
         if self.driver == 'vmw_ahci':
             return 2
@@ -49,6 +128,7 @@ class DiskInfo(object):
 
 def list_disks():
     current_dev = None
+    m2slots = list_m2_slots()
     disks = {}
     devlist = subprocess.check_output(['localcli', 'storage', 'core', 'device', 'list'])
     if not isinstance(devlist, str):
@@ -96,31 +176,40 @@ def list_disks():
                 adp = ' '.join(line.split()[1:])
                 disks[dev]['adapter'] = adp
                 devbyadp.setdefault(adp, []).append(dev)
-    adapterlist = subprocess.check_output(['localcli', 'storage', 'core', 'adapter', 'list'])
+    adapterlist = subprocess.check_output(['localcli', 'storage', 'core', 'adapter', 'listdetailed'])
     if not isinstance(adapterlist, str):
         adapterlist = adapterlist.decode('utf8')
-    driverbyadp = {}
-    linenum = 0
+    adapters = {}
+    curradpinfo = {}
     for line in adapterlist.split('\n'):
-        linenum += 1
+        line = line.rstrip()
         if not line.strip():
+            curradpinfo = {}
             continue
-        if linenum < 3:
-            continue
-        parts = line.split()
-        if len(parts) < 2:
-            continue
-        adp = parts[0]
-        driver = parts[1]
-        driverbyadp[adp] = driver
+        if not line.startswith(' '):
+            curradpname = line.split(':', 1)[0].strip()
+            curradpinfo['name'] = curradpname
+            adapters[curradpname] = curradpinfo
+        else:
+            if '   Driver:' in line:
+                curradpinfo['driver'] = line.split(maxsplit=1)[1]
+            elif '   Description:' in line:
+                curradpinfo['description'] = line.split(maxsplit=1)[1]
+                if curradpinfo['description'].startswith('('):
+                    curradpinfo['pcieaddress'] = curradpinfo['description'].split('(')[1].split(')')[0]
     for adp in devbyadp:
-        driver = driverbyadp.get(adp, 'Unknown')
+        driver = adapters.get(adp, {}).get('driver', 'Unknown')
+        pcieaddr = adapters.get(adp, {}).get('pcieaddress', None)
         for dev in devbyadp[adp]:
             disks[dev]['adapter_driver'] = driver
+            disks[dev]['adapter_pcieaddress'] = pcieaddr
+            if pcieaddr and pcieaddr in m2slots:
+                disks[dev]['m2_slot'] = m2slots[pcieaddr]
     return disks
         
 def main():
     disks = []
+    storagespec = get_deployment_storage()
     try:
         alldisks = list_disks()
     except Exception as e:
@@ -128,7 +217,7 @@ def main():
         alldisks = {}
     for disk in alldisks:
         try:
-            disks.append(DiskInfo(disk, alldisks[disk]))
+            disks.append(DiskInfo(disk, alldisks[disk], storagespec=storagespec))
         except SilentException:
             pass
         except Exception as e:

confluent_server/confluent/config/attributes.py

@@ -255,6 +255,9 @@ node = {
     'deployment.state_last_updated': {
         'description': ('Timestamp of last state change, as reported by an OS profile, when available'),
     },
+    'deployment.storage': {
+        'description': ('Indicates the storage to use for OS deployment.  This may be a specific device name, or a more generic description such as "m2" to indicate that the OS should be deployed to an M.2 device.  If not specified, the default behavior is to prioritize according to various criteria and use the first disk based on that evaluation.'),
+    },
     'deployment.useinsecureprotocols': {
         'description': ('What phase(s) of boot are permitted to use insecure protocols '
                         '(TFTP and HTTP without TLS.  By default, only HTTPS is used.  However '