Merge pull request #275 from Obihoernchen/feature/eventlog-sources

Add nodeeventlog source filtering for redfish

confluent_client/bin/nodeeventlog

@@ -53,7 +53,13 @@ argparser.add_option('-t', '--timeframe', type='string',
                           'entries from the last hours or days. '
                           '1h would be one hour, 4d would be four days. '
                           'format <num>h or <num>d'
-                          )  
+                          )
+
+
+argparser.add_option('-s', '--source', type='string',
+                     help='only show entries from the named logs, comma '
+                          'delimited. "-s list" names the logs this node '
+                          'offers instead of showing entries')
 (options, args) = argparser.parse_args()
 try:
     noderange = args[0]
@@ -72,6 +78,19 @@ if len(args) == 2:
         argparser.print_help()
         sys.exit(1)
 
+listmode = bool(options.source) and options.source.lower() == 'list'
+wantsources = None
+if options.source and not listmode:
+    wantsources = set(x.strip().lower() for x in options.source.split(',')
+                      if x.strip())
+if options.source and deletemode:
+    # Clearing a subset is not something the platforms offer, and clearing more
+    # than was asked for is not something to do quietly
+    sys.stderr.write(
+        'Clearing a selection of logs is not supported, so --source cannot be '
+        'combined with clear\n')
+    sys.exit(1)
+
 session = client.Command()
 exitcode = 0
 
@@ -125,6 +144,7 @@ if options.timeframe:
         sys.exit(1)
     timeframe = dt.now() - tdelta
 
+sources_seen = {}
 event_dict = {}
 nodes = []
 for res in session.read('/noderange/{0}/nodes/'.format(args[0])):
@@ -145,6 +165,17 @@ for rsp in func('/noderange/{0}/events/hardware/log'.format(noderange)):
                exitcode |= 1
             if 'events' in thisdata:
                 evtdata = thisdata['events']
+                for evt in evtdata:
+                    if evt.get('log_id', None):
+                        sources_seen.setdefault(node, set()).add(evt['log_id'])
+                if listmode:
+                    continue
+                if wantsources is not None:
+                    # Filter before any tail is taken, or asking for the last
+                    # few entries of one log would answer with fewer
+                    evtdata = [x for x in evtdata if (x.get('log_id', '')
+                                                      or '').lower()
+                               in wantsources]
                 if options.lines:
                     event_dict[node].extend(evtdata)
                 else:
@@ -158,7 +189,34 @@ for rsp in func('/noderange/{0}/events/hardware/log'.format(noderange)):
                         else:
                             print('{0}: {1}'.format(node, format_event(evt))) 
 
-if options.lines:
+if listmode:
+    for node in nodes:
+        found = sorted(sources_seen.get(node, ()))
+        if found:
+            print('{0}: {1}'.format(node, ','.join(found)))
+        else:
+            sys.stderr.write(
+                'No event log sources reported for "{0}"\n'.format(node))
+elif wantsources is not None:
+    for node in nodes:
+        found = set(x.lower() for x in sources_seen.get(node, ()))
+        missing = wantsources - found
+        if not missing:
+            continue
+        if found:
+            sys.stderr.write('{0}: no such log source: {1}, this node has {2}\n'
+                             .format(node, ','.join(sorted(missing)),
+                                     ','.join(sorted(sources_seen[node]))))
+        else:
+            # A log that holds no entries names no source, so an empty answer
+            # here is as often a quiet platform as a wrong name.  Either way
+            # saying nothing would look like the log simply had nothing in it.
+            sys.stderr.write(
+                '{0}: no entries from any log source, so nothing could match '
+                '{1}\n'.format(node, ','.join(sorted(missing))))
+        exitcode |= 1
+
+if options.lines and not listmode:
     for node in nodes:
         evtdata_list = event_dict[node]
         if len(evtdata_list) != 0:
@@ -172,3 +230,5 @@ if options.lines:
                             print('{0}: {1}'.format(node, format_event(evt)))
                 else:
                     print('{0}: {1}'.format(node, format_event(evt)))
+
+sys.exit(exitcode)

confluent_client/doc/man/nodeeventlog.ronn

@@ -3,13 +3,28 @@ nodeeventlog(8) -- Pull eventlog from confluent nodes
 
 ## SYNOPSIS
 
-`nodeeventlog [options] <noderange> [clear]`  
+`nodeeventlog [options] [-s <sources>] <noderange> [clear]`
 
 ## DESCRIPTION
 
 `nodeeventlog` pulls and optionally clears the event log from the requested
 noderange.
 
+A platform may keep its events in more than one log, and some of those are
+noisier than others.  Reading gathers every log the platform describes as an
+event log, wherever it keeps them, and `-s` narrows the output to the ones
+asked for.  `-s list` names the logs a node offers rather than showing entries.
+
+A source is named by the entries that come from it, so a log that holds no
+entries is not among them, and a node whose logs are all empty names none.  An
+ipmi managed node keeps its events in the one log the spec gives it, which is
+reported as `SEL`.
+
+Clearing is deliberately narrower than reading: it only touches the logs the
+platform's own manager publishes, so a log that only a read reaches is never
+destroyed by one.  A selection cannot be cleared, so `-s` and `clear` are
+refused together.
+
 ## OPTIONS
 
 * `-m MAXNODES`, `--maxnodes=MAXNODES`:
@@ -25,10 +40,26 @@ noderange.
    entries from within the last one hour.
                            
   
+* `-s SOURCES`, `--source=SOURCES`:
+   only show entries from the named logs, comma delimited and matched without
+   regard to case. `-s list` names the logs the node offers instead of showing
+   entries. A platform that named a log "list" would be shadowed by that.
+   Asking for a log the node named nothing from is reported on stderr and
+   exits non-zero, rather than looking like an empty log.
+
 * `-h`, `--help`:
   Show help message and exit  
 
 ## EXAMPLES
+* Ask which logs a node keeps:
+  `# nodeeventlog n2 -s list`  
+  `n2: AuditLog,EventLog,Logs,SEL`  
+
+* Pull only the sel and the chassis log, leaving the audit noise out:
+  `# nodeeventlog n2 -s SEL,Logs`  
+  `n2: 07/23/2026 13:03:12 SEL: Processor - Thermal Trip`  
+  `n2: 08/05/2026 04:30:26 Logs: PowerUnit - Power Off`  
+
 * Pull the event log from n2 and n3:
   `# nodeeventlog n2,n3`  
   `n2: 05/03/2017 11:44:25 Event Log Disabled - SEL Fullness - Log clear`  

confluent_server/aiohmi/redfish/oem/generic.py

@@ -570,18 +570,51 @@ class OEMHandler(object):
     @classmethod
     def is_event_log(cls, loginfo):
         """Say whether a log service holds events rather than something else"""
-        identity = '{0} {1}'.format(loginfo.get('Id', ''),
-                                    loginfo.get('Name', '')).lower()
-        for word in cls.noneventlogwords:
-            if word in identity:
-                return False
+        for part in (loginfo.get('Id', ''), loginfo.get('Name', '')):
+            # A build is free to space or punctuate the same name differently,
+            # "BIOS POST Code Log" for a post code log, so compare without any
+            # of that rather than miss it and merge 2719 post codes into the
+            # event log
+            flat = re.sub(r'[^a-z0-9]', '', str(part).lower())
+            for word in cls.noneventlogwords:
+                if word in flat:
+                    return False
         return True
 
+    async def _othereventlogcollections(self, fishclient, everywhere=True):
+        """The log service collections outside the manager.
+
+        A platform is free to keep its event log on the system or the chassis
+        rather than on the manager, and more than one of those may link the very
+        same collection, so the answer is deduplicated.
+        """
+        collections = []
+        candidates = list(self._allsysurls)
+        if everywhere:
+            try:
+                chassiscol = await self._do_web_request('/redfish/v1/Chassis')
+                candidates.extend(
+                    x['@odata.id'] for x in chassiscol.get('Members', []))
+            except Exception:
+                # A platform that will not describe its chassis still has the
+                # rest of its logs to offer
+                pass
+        for url in candidates:
+            try:
+                info = await self._do_web_request(url)
+            except Exception:
+                continue
+            lscollection = info.get('LogServices', {}).get('@odata.id', None)
+            if lscollection and lscollection not in collections:
+                collections.append(lscollection)
+        return collections
+
     async def get_event_log(self, clear=False, fishclient=None, extraurls=[]):
         bmcinfo = await self._do_web_request(await fishclient.get_bmcurl())
+        # A manager need not publish log services at all, and one that does not
+        # is the clearest case of a platform keeping its event log elsewhere, so
+        # carry on to the fallback below rather than answering with nothing
         lsurl = bmcinfo.get('LogServices', {}).get('@odata.id', None)
-        if not lsurl:
-            return
         currtime = bmcinfo.get('DateTime', None)
         correction = timedelta(0)
         utz = tz.tzoffset('', 0)
@@ -612,17 +645,22 @@ class OEMHandler(object):
                     found.append(candidate)
             return found
 
-        lurls = await eventlogurls(lsurl)
-        if not lurls:
+        lurls = await eventlogurls(lsurl) if lsurl else []
+        if not lurls or not clear:
             # Some implementations keep no event log under the manager and put
             # it under the system instead, so fall back to looking there rather
-            # than answering with nothing at all.
-            for sysurl in self._allsysurls:
-                currsysinfo = await self._do_web_request(sysurl)
-                syslsurl = currsysinfo.get('LogServices', {}).get(
-                    '@odata.id', None)
-                if syslsurl:
-                    lurls.extend(await eventlogurls(syslsurl))
+            # than answering with nothing at all.  A read goes further and
+            # gathers every event log the platform publishes, since one that is
+            # never read is one nobody can act on.  Clearing deliberately does
+            # not: a log that only a read reaches must not be destroyed by one.
+            for lscollection in await self._othereventlogcollections(
+                    fishclient, everywhere=not clear):
+                if lscollection == lsurl:
+                    # A platform may link the manager's own collection from the
+                    # system and the chassis as well, and reading it three times
+                    # would answer the same thing three times
+                    continue
+                lurls.extend(await eventlogurls(lscollection))
         lurls.extend([x['@odata.id'] for x in extraurls])
         seenurls = set()
         for lurl in lurls:

confluent_server/confluent/plugins/hardwaremanagement/ipmi.py

@@ -933,6 +933,10 @@ class IpmiHandler:
         await self.output.put(msg.EventCollection(eventout, name=self.node))
 
     def pyghmi_event_to_confluent(self, event):
+        # An ipmi platform keeps its events in one place and the spec has a
+        # name for it, so say so rather than leaving the source blank and
+        # having a caller asking for a named log get nothing back
+        event.setdefault('log_id', 'SEL')
         event['severity'] = _str_health(event.get('severity', 'unknown'))
         if 'event_data' in event:
             event['event'] = '{0} - {1}'.format(