Merge pull request #18686 from jakesmith/HPCC-31902-audit-containerinfo

HPCC-31902 Adjust Thor auditing info to contain pod/container meta info Reviewed-by: Gavin Halliday <[email protected]> Merged-by: Gavin Halliday <[email protected]>
hpcc-systems · Jun 14, 2024 · 21ccf10 · 21ccf10
2 parents fcb0440 + d83da9f
commit 21ccf10
Show file tree

Hide file tree

Showing 5 changed files with 51 additions and 35 deletions.
diff --git a/thorlcr/master/thdemonserver.cpp b/thorlcr/master/thdemonserver.cpp
@@ -273,14 +273,6 @@ class DeMonServer : public CSimpleInterface, implements IDeMonServer
         unsigned startTime = msTick();
         graphStarts.append(startTime);
         reportGraph(graph, false, true, startTime, getTimeStampNowValue());
-        const char *graphname = graph->queryJob().queryGraphName();
-        if (memcmp(graphname,"graph",5)==0)
-            graphname+=5;
-        LOG(MCauditInfo,",Progress,Thor,StartSubgraph,%s,%s,%s,%u,%s,%s",
-                queryServerStatus().queryProperties()->queryProp("@thorname"),
-                graph->queryJob().queryWuid(),
-                graphname,
-                (unsigned)graph->queryGraphId(), queryServerStatus().queryProperties()->queryProp("@nodeGroup"), queryServerStatus().queryProperties()->queryProp("@queue"));
     }
     void endGraph(CGraphBase *graph, bool success)
     {

diff --git a/thorlcr/master/thgraphmanager.cpp b/thorlcr/master/thgraphmanager.cpp
@@ -433,10 +433,7 @@ void CJobManager::fatal(IException *e)
     {
         IERRLOG("Unknown exception in CJobManager::fatal");
     }
-    LOG(MCauditInfo,",Progress,Thor,Terminate,%s,%s,%s,exception",
-            queryServerStatus().queryProperties()->queryProp("@thorname"),
-            queryServerStatus().queryProperties()->queryProp("@nodeGroup"),
-            queryServerStatus().queryProperties()->queryProp("@queue"));
+    auditThorSystemEvent("Terminate", {"exception"});
 
     queryLogMsgManager()->flushQueue(10*1000);
 
@@ -890,27 +887,16 @@ bool CJobManager::doit(IConstWorkUnit *workunit, const char *graphName, const So
     JobNameScope activeJobName(wuid);
 
     LOG(MCdebugInfo, "Processing wuid=%s, graph=%s from agent: %s", wuid.str(), graphName, agentep.getEndpointHostText(s).str());
-    LOG(MCauditInfo,",Progress,Thor,Start,%s,%s,%s,%s,%s,%s",
-            queryServerStatus().queryProperties()->queryProp("@thorname"),
-            wuid.str(),
-            graphName,
-            user.str(),
-            queryServerStatus().queryProperties()->queryProp("@nodeGroup"),
-            queryServerStatus().queryProperties()->queryProp("@queue"));
+    auditThorJobEvent("Start", wuid, graphName, user);
+
     Owned<IException> e;
     bool allDone = false;
     try
     {
         allDone = executeGraph(*workunit, graphName, agentep);
     }
     catch (IException *_e) { e.setown(_e); }
-    LOG(MCauditInfo,",Progress,Thor,Stop,%s,%s,%s,%s,%s,%s",
-            queryServerStatus().queryProperties()->queryProp("@thorname"),
-            wuid.str(),
-            graphName,
-            user.str(),
-            queryServerStatus().queryProperties()->queryProp("@nodeGroup"),
-            queryServerStatus().queryProperties()->queryProp("@queue"));
+    auditThorJobEvent("Stop", wuid, graphName, user);
 
     if (e.get()) throw e.getClear();
     return allDone;
@@ -1285,7 +1271,6 @@ void closeThorServerStatus()
     }
 }
 
-
 /*
  * Waits on recv for another wuid/graph to run.
  * Return values:
@@ -1358,6 +1343,42 @@ void publishPodNames(IWorkUnit *workunit, const char *graphName)
     }
 }
 
+static void auditThorSystemEventBuilder(std::string &msg, const char *eventName, std::initializer_list<const char*> args)
+{
+    msg += std::string(",Progress,Thor,") + eventName + "," + getComponentConfigSP()->queryProp("@name");
+    for (auto arg : args)
+        msg += "," + std::string(arg);
+    if (isContainerized())
+        msg += std::string(",") + k8s::queryMyPodName() + "," + k8s::queryMyContainerName();
+    else
+    {
+        const char *nodeGroup = queryServerStatus().queryProperties()->queryProp("@nodeGroup");
+        const char *queueName = queryServerStatus().queryProperties()->queryProp("@queue");
+        msg += std::string(",") + nodeGroup + "," + queueName;
+    }
+}
+
+void auditThorSystemEvent(const char *eventName)
+{
+    std::string msg;
+    auditThorSystemEventBuilder(msg, eventName, {});
+    LOG(MCauditInfo, "%s", msg.c_str());
+}
+
+void auditThorSystemEvent(const char *eventName, std::initializer_list<const char*> args)
+{
+    std::string msg;
+    auditThorSystemEventBuilder(msg, eventName, args);
+    LOG(MCauditInfo, "%s", msg.c_str());
+}
+
+void auditThorJobEvent(const char *eventName, const char *wuid, const char *graphName, const char *user)
+{
+    std::string msg;
+    auditThorSystemEventBuilder(msg, eventName, { wuid, graphName, nullText(user) });
+    LOG(MCauditInfo, "%s", msg.c_str());
+}
+
 void thorMain(ILogMsgHandler *logHandler, const char *wuid, const char *graphName)
 {
     aborting = 0;

diff --git a/thorlcr/master/thgraphmanager.hpp b/thorlcr/master/thgraphmanager.hpp
@@ -34,6 +34,9 @@ int queryExitCode();
 void addConnectedWorkerPod(const char *podName, const char *containerName);
 void publishPodNames(IWorkUnit *workunit, const char *graphName);
 void relayWuidException(IConstWorkUnit *wu, const IException *exception);
+void auditThorSystemEvent(const char *eventName);
+void auditThorSystemEvent(const char *eventName, std::initializer_list<const char*> args);
+void auditThorJobEvent(const char *eventName, const char *wuid, const char *graphName, const char *user);
 
 
 #endif
diff --git a/thorlcr/master/thmastermain.cpp b/thorlcr/master/thmastermain.cpp
@@ -596,10 +596,7 @@ bool ControlHandler(ahType type)
             if (auditStartLogged)
             {
                 auditStartLogged = false;
-                LOG(MCauditInfo,",Progress,Thor,Terminate,%s,%s,%s,ctrlc",
-                    queryServerStatus().queryProperties()->queryProp("@thorname"),
-                    queryServerStatus().queryProperties()->queryProp("@nodeGroup"),
-                    queryServerStatus().queryProperties()->queryProp("@queue"));
+                auditThorSystemEvent("Terminate", {"ctrlc"});
             }
             queryLogMsgManager()->flushQueue(10*1000);
             _exit(TEC_CtrlC);
@@ -968,7 +965,10 @@ int main( int argc, const char *argv[]  )
     getClusterThorQueueName(queueNames, thorName);
 #else
     if (!thorName)
+    {
         thorName = "thor";
+        globals->setProp("@name", thorName);
+    }
     SCMStringBuffer queueNames;
     getThorQueueNames(queueNames, thorName);
 #endif
@@ -992,6 +992,7 @@ int main( int argc, const char *argv[]  )
         masterSlaveMpTag = allocateClusterMPTag();
         kjServiceMpTag = allocateClusterMPTag();
 
+        auditThorSystemEvent("Initializing");
         unsigned numWorkers = 0;
         if (isContainerized())
         {
@@ -1000,7 +1001,6 @@ int main( int argc, const char *argv[]  )
 
             StringBuffer thorEpStr;
             LOG(MCdebugProgress, "ThorMaster version %d.%d, Started on %s", THOR_VERSION_MAJOR,THOR_VERSION_MINOR,thorEp.getEndpointHostText(thorEpStr).str());
-            LOG(MCdebugProgress, "Thor name = %s, queue = %s, nodeGroup = %s",thorname,queueName.str(),nodeGroup.str());
 
             unsigned numWorkersPerPod = 1;
             if (!globals->hasProp("@numWorkers"))
@@ -1126,7 +1126,7 @@ int main( int argc, const char *argv[]  )
             PROGLOG("Persistent Thor group created with group name: %s", uniqueGrpName.str());
         }
 #endif
-        LOG(MCauditInfo, ",Progress,Thor,Startup,%s,%s,%s,%s",nodeGroup.str(),thorname,queueName.str(),logUrl.str());
+        auditThorSystemEvent("Startup");
         auditStartLogged = true;
 
         writeSentinelFile(sentinelFile);
@@ -1140,7 +1140,7 @@ int main( int argc, const char *argv[]  )
 
         // NB: workunit/graphName only set in one-shot mode (if isCloud())
         thorMain(logHandler, workunit, graphName);
-        LOG(MCauditInfo, ",Progress,Thor,Terminate,%s,%s,%s",thorname,nodeGroup.str(),queueName.str());
+        auditThorSystemEvent("Terminate");
         LOG(MCdebugProgress, "ThorMaster terminated OK");
     }
     catch (IException *e) 

diff --git a/thorlcr/thorutil/thormisc.cpp b/thorlcr/thorutil/thormisc.cpp
@@ -1692,4 +1692,4 @@ void saveWuidToFile(const char *wuid)
     if (!wuidFileIO)
         throw makeStringException(0, "Failed to create file 'wuid' to store current workunit for post mortem script");
     wuidFileIO->write(0, strlen(wuid), wuid);
-}
+}