implemented the $ActionResumeRetryCount config directive

9 files changed, 104 insertions, 34 deletions

diff --git a/ChangeLog b/ChangeLog
index 9188624e..0d9cf9ec 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,6 @@
 ---------------------------------------------------------------------------
 Version 3.10.4 (rgerhards), 2008-01-??
+- implemented the $ActionResumeRetryCount config directive
 ---------------------------------------------------------------------------
 Version 3.10.3 (rgerhards), 2008-01-28
 - fixed a bug with standard template definitions (not a big deal) - thanks
diff --git a/action.c b/action.c
index 2410c76e..2154fba2 100644
--- a/action.c
+++ b/action.c
@@ -36,10 +36,12 @@
 #include "action.h"
 #include "modules.h"
 #include "sync.h"
+#include "srUtils.h"
 
 
 /* object static data (once for all instances) */
 static int glbliActionResumeInterval = 30;
+int glbliActionResumeRetryCount = 0;		/* how often should suspended actions be retried? */
 
 /* destructs an action descriptor object
  * rgerhards, 2007-08-01
@@ -80,6 +82,7 @@ rsRetVal actionConstruct(action_t **ppThis)
 	}
 
 	pThis->iResumeInterval = glbliActionResumeInterval;
+	pThis->iResumeRetryCount = glbliActionResumeRetryCount;
 	SYNC_OBJ_TOOL_INIT(pThis);
 
 finalize_it:
@@ -191,6 +194,69 @@ rsRetVal actionDbgPrint(action_t *pThis)
 }
 
 
+/* call the DoAction output plugin entry point
+ * rgerhards, 2008-01-28
+ */
+rsRetVal
+actionCallDoAction(action_t *pAction)
+{
+	DEFiRet;
+	int iRetries;
+	int i;
+	int iSleepPeriod;
+
+	assert(pAction != NULL);
+
+	/* here we must loop to process all requested strings */
+	for(i = 0 ; i < pAction->iNumTpls ; ++i) {
+		CHKiRet(tplToString(pAction->ppTpl[i], pAction->f_pMsg, &pAction->ppMsgs[i]));
+	}
+
+	iRetries = 0;
+	do {
+RUNLOG_STR("going into do_action call loop");
+RUNLOG_VAR("%d", iRetries);
+		/* call configured action */
+		iRet = pAction->pMod->mod.om.doAction(pAction->ppMsgs, pAction->f_pMsg->msgFlags, pAction->pModData);
+RUNLOG_VAR("%d", iRet);
+		if(iRet == RS_RET_SUSPENDED) {
+			/* ok, this calls for our retry logic... */
+			++iRetries;
+			iSleepPeriod = pAction->iResumeInterval;
+RUNLOG_VAR("%d", iSleepPeriod);
+			srSleep(iSleepPeriod, 0);
+		} else {
+			break; /* we are done in any case */
+		}
+	} while(pAction->iResumeRetryCount == -1 || iRetries < pAction->iResumeRetryCount); /* do...while! */
+RUNLOG_STR("out of retry loop");
+
+	if(iRet == RS_RET_DISABLE_ACTION) {
+		dbgprintf("Action requested to be disabled, done that.\n");
+		pAction->bEnabled = 0; /* that's it... */
+	}
+
+	if(iRet == RS_RET_SUSPENDED) {
+		dbgprintf("Action requested to be suspended, done that.\n");
+		actionSuspend(pAction);
+	}
+
+	if(iRet == RS_RET_OK)
+		pAction->f_prevcount = 0; /* message processed, so we start a new cycle */
+
+finalize_it:
+	/* cleanup */
+	for(i = 0 ; i < pAction->iNumTpls ; ++i) {
+		if(pAction->ppMsgs[i] != NULL) {
+			free(pAction->ppMsgs[i]);
+			pAction->ppMsgs[i] = NULL;
+		}
+	}
+
+	RETiRet;
+}
+
+
 /*
  * vi:set ai:
  */
diff --git a/action.h b/action.h
index 30eaaf95..0355b834 100644
--- a/action.h
+++ b/action.h
@@ -28,6 +28,13 @@
 #include "syslogd-types.h"
 #include "sync.h"
 
+/* external data - this is to be removed when we change the action
+ * object interface (will happen some time..., at latest when the
+ * config file format is changed). -- rgerhards, 2008-01-28
+ */
+extern int glbliActionResumeRetryCount;
+
+
 /* the following struct defines the action object data structure
  */
 struct action_s {
@@ -37,6 +44,7 @@ struct action_s {
 	short	bSuspended;	/* is the related action temporarily suspended? */
 	time_t	ttResumeRtry;	/* when is it time to retry the resume? */
 	int	iResumeInterval;/* resume interval for this action */
+	int	iResumeRetryCount;/* how often shall we retry a suspended action? (-1 --> eternal) */
 	int	iNbrResRtry;	/* number of retries since last suspend */
 	struct moduleInfo *pMod;/* pointer to output module handling this selector */
 	void	*pModData;	/* pointer to module data - content is module-specific */
@@ -65,6 +73,7 @@ rsRetVal actionTryResume(action_t *pThis);
 rsRetVal actionSuspend(action_t *pThis);
 rsRetVal actionDbgPrint(action_t *pThis);
 rsRetVal actionSetGlobalResumeInterval(int iNewVal);
+rsRetVal actionCallDoAction(action_t *pAction);
 
 #if 1
 #define actionIsSuspended(pThis) ((pThis)->bSuspended == 1)
diff --git a/cfsysline.c b/cfsysline.c
index 7cbd5884..50d0066b 100644
--- a/cfsysline.c
+++ b/cfsysline.c
@@ -875,7 +875,7 @@ void dbgPrintCfSysLineHandlers(void)
 		}
 	}
 	printf("\n");
-
+	ENDfunc
 }
 
 /*
diff --git a/debug.h b/debug.h
index a9891d84..19e91350 100644
--- a/debug.h
+++ b/debug.h
@@ -107,9 +107,12 @@ void dbgPrintAllDebugInfo(void);
 #	define RUNLOG dbgSetExecLocation(dbgCALLStaCK_POP_POINT, __LINE__); dbgprintf("%s:%d: %s: log point\n", __FILE__, __LINE__, __func__)
 #	define RUNLOG_VAR(fmt, x) dbgSetExecLocation(dbgCALLStaCK_POP_POINT, __LINE__);\
 	 		          dbgprintf("%s:%d: %s: var '%s'[%s]: " fmt "\n", __FILE__, __LINE__, __func__, #x, fmt, x)
+#	define RUNLOG_STR(str)    dbgSetExecLocation(dbgCALLStaCK_POP_POINT, __LINE__);\
+				  dbgprintf("%s:%d: %s: %s\n", __FILE__, __LINE__, __func__, str)
 #else
 #	define RUNLOG
 #	define RUNLOG_VAR(fmt, x)
+#	define RUNLOG_STR(str)
 #endif
 
 /* mutex operations */
diff --git a/doc/rsyslog_conf.html b/doc/rsyslog_conf.html
index f69c80fd..929a8a0b 100644
--- a/doc/rsyslog_conf.html
+++ b/doc/rsyslog_conf.html
@@ -34,6 +34,7 @@ development and quite unstable...). So you have been warned ;)</p>
 	<li><a href="rsconf1_actionresumeinterval.html">$ActionResumeInterval</a></li>
 	<li>$AddUnixListenSocket &lt;name-of-socket&gt; adds additional unix socket, 
 	default none -- former -a option</li>
+	<li>$ActionResumeRetryCount &lt;number&gt; [default 0, -1 means eternal]</li>
 	<li><a href="rsconf1_allowedsender.html">$AllowedSender</a></li>
 	<li><a href="rsconf1_controlcharacterescapeprefix.html">$ControlCharacterEscapePrefix</a></li>
 	<li><a href="rsconf1_debugprintcfsyslinehandlerlist.html">$DebugPrintCFSyslineHandlerList</a></li>
diff --git a/srUtils.c b/srUtils.c
index e15456a4..6da1010d 100755
--- a/srUtils.c
+++ b/srUtils.c
@@ -373,6 +373,22 @@ mutexCancelCleanup(void *arg)
 }
 
 
+/* rsSleep() - a fairly portable way to to sleep. It 
+ * will wake up when
+ * a) the wake-time is over
+ * rgerhards, 2008-01-28
+ */
+void
+srSleep(int iSeconds, int iuSeconds)
+{
+	struct timeval tvSelectTimeout;
+
+	BEGINfunc
+	tvSelectTimeout.tv_sec = iSeconds;
+	tvSelectTimeout.tv_usec = iuSeconds; /* micro seconds */
+	select(0, NULL, NULL, NULL, &tvSelectTimeout);
+	ENDfunc
+}
 /*
  * vi:set ai:
  */
diff --git a/srUtils.h b/srUtils.h
index 73109c30..8436cf0b 100755
--- a/srUtils.h
+++ b/srUtils.h
@@ -69,6 +69,7 @@ int getNumberDigits(long lNum);
 rsRetVal timeoutComp(struct timespec *pt, long iTimeout);
 long timeoutVal(struct timespec *pt);
 void mutexCancelCleanup(void *arg);
+void srSleep(int iSeconds, int iuSeconds);
 
 /* mutex operations */
 /* some macros to cancel-safe lock a mutex (it will automatically be released
diff --git a/syslogd.c b/syslogd.c
index c46b5507..d60abade 100644
--- a/syslogd.c
+++ b/syslogd.c
@@ -535,6 +535,7 @@ static rsRetVal resetConfigVariables(uchar __attribute__((unused)) *pp, void __a
 	iMainMsgQWrkMinMsgs = 100;
 	bMainMsgQSaveOnShutdown = 1;
 	MainMsgQueType = QUEUETYPE_FIXED_ARRAY;
+	glbliActionResumeRetryCount = 0;
 
 	return RS_RET_OK;
 }
@@ -2351,7 +2352,6 @@ fprintlog(action_t *pAction)
 	msg_t *pMsgSave;	/* to save current message pointer, necessary to restore
 				   it in case it needs to be updated (e.g. repeated msgs) */
 	DEFiRet;
-	int i;
 
 	pMsgSave = NULL;	/* indicate message poiner not saved */
 	/* first check if this is a regular message or the repeation of
@@ -2371,10 +2371,7 @@ fprintlog(action_t *pAction)
 		if((pMsg = MsgDup(pAction->f_pMsg)) == NULL) {
 			/* it failed - nothing we can do against it... */
 			dbgprintf("Message duplication failed, dropping repeat message.\n");
-			return RS_RET_ERR;
-			/* This return is OK. The finalizer frees strings, which are not
-			 * yet allocated. So we can not use the finalizer.
-			 */
+			ABORT_FINALIZE(RS_RET_ERR);
 		}
 
 		/* We now need to update the other message properties.
@@ -2397,36 +2394,9 @@ fprintlog(action_t *pAction)
 	/* When we reach this point, we have a valid, non-disabled action.
 	 * So let's execute it. -- rgerhards, 2007-07-24
 	 */
-	/* here we must loop to process all requested strings */
-
-	for(i = 0 ; i < pAction->iNumTpls ; ++i) {
-		CHKiRet(tplToString(pAction->ppTpl[i], pAction->f_pMsg, &pAction->ppMsgs[i]));
-	}
-	/* call configured action */
-	iRet = pAction->pMod->mod.om.doAction(pAction->ppMsgs, pAction->f_pMsg->msgFlags, pAction->pModData);
-
-	if(iRet == RS_RET_DISABLE_ACTION) {
-		dbgprintf("Action requested to be disabled, done that.\n");
-		pAction->bEnabled = 0; /* that's it... */
-	}
-
-	if(iRet == RS_RET_SUSPENDED) {
-		dbgprintf("Action requested to be suspended, done that.\n");
-		actionSuspend(pAction);
-	}
-
-	if(iRet == RS_RET_OK)
-		pAction->f_prevcount = 0; /* message processed, so we start a new cycle */
+	iRet = actionCallDoAction(pAction);
 
 finalize_it:
-	/* cleanup */
-	for(i = 0 ; i < pAction->iNumTpls ; ++i) {
-		if(pAction->ppMsgs[i] != NULL) {
-			free(pAction->ppMsgs[i]);
-			pAction->ppMsgs[i] = NULL;
-		}
-	}
-
 	if(pMsgSave != NULL) {
 		/* we had saved the original message pointer. That was
 		 * done because we needed to create a temporary one
@@ -3158,6 +3128,8 @@ static void dbgPrintInitInfo(void)
 	dbgprintf("Main queue watermarks: high: %d, low: %d, discard: %d, discard-severity: %d\n",
 		   iMainMsgQHighWtrMark, iMainMsgQLowWtrMark, iMainMsgQDiscardMark, iMainMsgQDiscardSeverity);
 	/* TODO: add
+	iActionRetryCount = 0;
+	iActionRetryInterval = 30000;
 	static int iMainMsgQtoWrkShutdown = 60000;
 	static int iMainMsgQtoWrkMinMsgs = 100;	
 	static int iMainMsgQbSaveOnShutdown = 1;
@@ -4570,6 +4542,7 @@ static rsRetVal loadBuildInModules(void)
 	 * This, I think, is the right thing to do. -- rgerhards, 2007-07-31
 	 */
 	CHKiRet(regCfSysLineHdlr((uchar *)"workdirectory", 0, eCmdHdlrGetWord, NULL, &pszWorkDir, NULL));
+	CHKiRet(regCfSysLineHdlr((uchar *)"actionresumeretrycount", 0, eCmdHdlrInt, NULL, &glbliActionResumeRetryCount, NULL));
 	CHKiRet(regCfSysLineHdlr((uchar *)"mainmsgqueuefilename", 0, eCmdHdlrGetWord, NULL, &pszMainMsgQFName, NULL));
 	CHKiRet(regCfSysLineHdlr((uchar *)"mainmsgqueuesize", 0, eCmdHdlrInt, NULL, &iMainMsgQueueSize, NULL));
 	CHKiRet(regCfSysLineHdlr((uchar *)"mainmsgqueuehighwatermark", 0, eCmdHdlrInt, NULL, &iMainMsgQHighWtrMark, NULL));