1 files changed, 94 insertions, 152 deletions

diff --git a/runtime/wti.c b/runtime/wti.c
index abdf4add..0ba1fe54 100644
--- a/runtime/wti.c
+++ b/runtime/wti.c
@@ -39,9 +39,10 @@
 #include <pthread.h>
 #include <errno.h>
 
-#ifdef OS_SOLARIS
-#	include <sched.h>
-#endif
+/// TODO: check on solaris if this is any longer needed - I don't think so - rgerhards, 2009-09-20
+//#ifdef OS_SOLARIS
+//#	include <sched.h>
+//#endif
 
 #include "rsyslog.h"
 #include "stringbuf.h"
@@ -99,11 +100,10 @@ wtiGetState(wti_t *pThis, int bLockMutex)
 
 
 /* send a command to a specific thread
- * bActiveOnly specifies if the command should be sent only when the worker is
- * in an active state. -- rgerhards, 2008-01-20
+ * rgerhards, 2008-01-20
  */
 rsRetVal
-wtiSetState(wti_t *pThis, qWrkCmd_t tCmd, int bActiveOnly, int bLockMutex)
+wtiSetState(wti_t *pThis, qWrkCmd_t tCmd, int bLockMutex)
 {
 	DEFiRet;
 	qWrkCmd_t tCurrCmd;
@@ -116,8 +116,7 @@ wtiSetState(wti_t *pThis, qWrkCmd_t tCmd, int bActiveOnly, int bLockMutex)
 
 	tCurrCmd = pThis->tCurrCmd;
 	/* all worker states must be followed sequentially, only termination can be set in any state */
-	if(   (bActiveOnly && (tCurrCmd < eWRKTHRD_RUN_CREATED))
-	   || (tCurrCmd > tCmd && !(tCmd == eWRKTHRD_TERMINATING || tCmd == eWRKTHRD_STOPPED))) {
+	if(tCurrCmd > tCmd && !(tCmd == eWRKTHRD_STOPPED)) {
 		DBGPRINTF("%s: command %d can not be accepted in current %d processing state - ignored\n",
 			  wtiGetDbgHdr(pThis), tCmd, tCurrCmd);
 	} else {
@@ -125,32 +124,16 @@ wtiSetState(wti_t *pThis, qWrkCmd_t tCmd, int bActiveOnly, int bLockMutex)
 		/* we could replace this with a simple if, but we leave the switch in in case we need
 		 * to add something at a later stage. -- rgerhards, 2008-09-30
 		 */
-		switch(tCmd) {
-			case eWRKTHRD_TERMINATING:
-				/* TODO: re-enable meaningful debug msg! (via function callback?)
-				dbgprintf("%s: thread terminating with %d entries left in queue, %d workers running.\n",
-					  wtiGetDbgHdr(pThis->pQueue), pThis->pQueue->iQueueSize,
-					  pThis->pQueue->iCurNumWrkThrd);
-				*/
-				pthread_cond_signal(&pThis->condExitDone);
-				dbgprintf("%s: worker terminating\n", wtiGetDbgHdr(pThis));
-				break;
-			/* these cases just to satisfy the compiler, we do (yet) not act an them: */
-			case eWRKTHRD_RUNNING:
-			case eWRKTHRD_STOPPED:
-			case eWRKTHRD_RUN_CREATED:
-			case eWRKTHRD_RUN_INIT:
-			case eWRKTHRD_SHUTDOWN:
-			case eWRKTHRD_SHUTDOWN_IMMEDIATE:
-				/* DO NOTHING */
-				break;
+		if(tCmd == eWRKTHRD_STOPPED) {
+			dbgprintf("%s: worker almost stopped, assuming it has\n", wtiGetDbgHdr(pThis));
+			pThis->thrdID = 0; /* invalidate the thread ID so that we do not accidently find reused ones */
 		}
 		/* apply the new state */
+dbgprintf("worker terminator will write stateval %d\n", tCmd);
 		unsigned val = ATOMIC_CAS_VAL(pThis->tCurrCmd, tCurrCmd, tCmd);
 		if(val != tCurrCmd) {
 			DBGPRINTF("wtiSetState PROBLEM, tCurrCmd %d overwritten with %d, wanted to set %d\n", tCurrCmd, val, tCmd);
 		}
-
 	}
 
 	END_MTX_PROTECTED_OPERATIONS(&pThis->mut);
@@ -172,13 +155,11 @@ wtiCancelThrd(wti_t *pThis)
 
 	d_pthread_mutex_lock(&pThis->mut);
 
-	wtiProcessThrdChanges(pThis, MUTEX_ALREADY_LOCKED); /* process state change, so that we have current state vars */
-
-	if(pThis->tCurrCmd >= eWRKTHRD_TERMINATING) {
+	if(pThis->tCurrCmd != eWRKTHRD_STOPPED) {
 		dbgoprint((obj_t*) pThis, "canceling worker thread, curr stat %d\n", pThis->tCurrCmd);
 		pthread_cancel(pThis->thrdID);
-		wtiSetState(pThis, eWRKTHRD_TERMINATING, 0, MUTEX_ALREADY_LOCKED);
-		ATOMIC_STORE_1_TO_INT(pThis->pWtp->bThrdStateChanged); /* indicate change, so harverster will be called */
+		/* TODO: check: the following check should automatically be done by cancel cleanup handler! 2009-07-08 rgerhards */
+		wtiSetState(pThis, eWRKTHRD_STOPPED, MUTEX_ALREADY_LOCKED);
 	}
 
 	d_pthread_mutex_unlock(&pThis->mut);
@@ -190,26 +171,15 @@ wtiCancelThrd(wti_t *pThis)
 /* Destructor */
 BEGINobjDestruct(wti) /* be sure to specify the object type also in END and CODESTART macros! */
 CODESTARTobjDestruct(wti)
-	/* if we reach this point, we must make sure the associated worker has terminated. It is
-	 * the callers duty to make sure the worker already knows it shall terminate.
-	 * TODO: is it *really* the caller's duty? ...mmmhhhh.... smells bad... rgerhards, 2008-01-25
-	 */
-	wtiProcessThrdChanges(pThis, LOCK_MUTEX); /* process state change one last time */
-
-	d_pthread_mutex_lock(&pThis->mut);
-	if(wtiGetState(pThis, MUTEX_ALREADY_LOCKED) != eWRKTHRD_STOPPED) {
-		dbgprintf("%s: WARNING: worker %p shall be destructed but is still running (might be OK) - joining it\n",
+	if(Debug && wtiGetState(pThis, MUTEX_ALREADY_LOCKED) != eWRKTHRD_STOPPED) {
+		dbgprintf("%s: WARNING: worker %p shall be destructed but is still running (might be OK) - ignoring\n",
 			  wtiGetDbgHdr(pThis), pThis);
-		/* let's hope the caller actually instructed it to shutdown... */
-		pthread_cond_wait(&pThis->condExitDone, &pThis->mut);
-		wtiJoinThrd(pThis);
 	}
-	d_pthread_mutex_unlock(&pThis->mut);
 
 	/* actual destruction */
-	pthread_cond_destroy(&pThis->condExitDone);
 	pthread_mutex_destroy(&pThis->mut);
 
+	free(pThis->batch.pElem);
 	free(pThis->pszDbgHdr);
 ENDobjDestruct(wti)
 
@@ -217,7 +187,6 @@ ENDobjDestruct(wti)
 /* Standard-Constructor for the wti object
  */
 BEGINobjConstruct(wti) /* be sure to specify the object type also in END macro! */
-	pthread_cond_init(&pThis->condExitDone, NULL);
 	pthread_mutex_init(&pThis->mut, NULL);
 ENDobjConstruct(wti)
 
@@ -229,75 +198,20 @@ rsRetVal
 wtiConstructFinalize(wti_t *pThis)
 {
 	DEFiRet;
+	int iDeqBatchSize;
 
 	ISOBJ_TYPE_assert(pThis, wti);
 
 	dbgprintf("%s: finalizing construction of worker instance data\n", wtiGetDbgHdr(pThis));
 
 	/* initialize our thread instance descriptor */
-	pThis->pUsrp = NULL;
 	pThis->tCurrCmd = eWRKTHRD_STOPPED;
 
-	RETiRet;
-}
-
-
-/* join a specific worker thread
- * we do not lock the mutex, because join will sync anyways...
- */
-rsRetVal
-wtiJoinThrd(wti_t *pThis)
-{
-	DEFiRet;
-
-	ISOBJ_TYPE_assert(pThis, wti);
-	dbgprintf("waiting for worker %s termination, current state %d\n", wtiGetDbgHdr(pThis), pThis->tCurrCmd);
-	if (pThis->thrdID == 0) {
-		dbgprintf("worker %s was already stopped\n", wtiGetDbgHdr(pThis));
-	} else {
-		pthread_join(pThis->thrdID, NULL);
-		wtiSetState(pThis, eWRKTHRD_STOPPED, 0, MUTEX_ALREADY_LOCKED); /* back to virgin... */
-		pThis->thrdID = 0; /* invalidate the thread ID so that we do not accidently find reused ones */
-	dbgprintf("worker %s has stopped\n", wtiGetDbgHdr(pThis));
-	}   
-
-	RETiRet;
-}
-
-/* check if we had a worker thread changes and, if so, act
- * on it. At a minimum, terminated threads are harvested (joined).
- */
-rsRetVal
-wtiProcessThrdChanges(wti_t *pThis, int bLockMutex)
-{
-	DEFiRet;
-	DEFVARS_mutexProtection;
-
-	ISOBJ_TYPE_assert(pThis, wti);
-
-	BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex);
-	switch(pThis->tCurrCmd) {
-		case eWRKTHRD_TERMINATING:
-			/* we need to at least temporarily release the mutex, because otherwise
-			 * we may deadlock with the thread we intend to join (it aquires the mutex
-			 * during termination processing). -- rgerhards, 2008-02-26
-			 */
-			END_MTX_PROTECTED_OPERATIONS(&pThis->mut);
-			iRet = wtiJoinThrd(pThis);
-			BEGIN_MTX_PROTECTED_OPERATIONS(&pThis->mut, bLockMutex);
-			break;
-		/* these cases just to satisfy the compiler, we do not act an them: */
-		case eWRKTHRD_STOPPED:
-		case eWRKTHRD_RUN_CREATED:
-		case eWRKTHRD_RUN_INIT:
-		case eWRKTHRD_RUNNING:
-		case eWRKTHRD_SHUTDOWN:
-		case eWRKTHRD_SHUTDOWN_IMMEDIATE:
-			/* DO NOTHING */
-			break;
-	}
-	END_MTX_PROTECTED_OPERATIONS(&pThis->mut);
+	/* we now alloc the array for user pointers. We obtain the max from the queue itself. */
+	CHKiRet(pThis->pWtp->pfGetDeqBatchSize(pThis->pWtp->pUsr, &iDeqBatchSize));
+	CHKmalloc(pThis->batch.pElem = calloc((size_t)iDeqBatchSize, sizeof(batch_obj_t)));
 
+finalize_it:
 	RETiRet;
 }
 
@@ -321,13 +235,12 @@ wtiWorkerCancelCleanup(void *arg)
 	DBGPRINTF("%s: cancelation cleanup handler called.\n", wtiGetDbgHdr(pThis));
 	
 	/* call user supplied handler (that one e.g. requeues the element) */
-	pWtp->pfOnWorkerCancel(pThis->pWtp->pUsr, pThis->pUsrp);
+	pWtp->pfOnWorkerCancel(pThis->pWtp->pUsr, pThis->batch.pElem[0].pUsrp);
 
 	pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave);
 	d_pthread_mutex_lock(&pWtp->mut);
-	wtiSetState(pThis, eWRKTHRD_TERMINATING, 0, MUTEX_ALREADY_LOCKED);
+	wtiSetState(pThis, eWRKTHRD_STOPPED, MUTEX_ALREADY_LOCKED);
 	/* TODO: sync access? I currently think it is NOT needed -- rgerhards, 2008-01-28 */
-	ATOMIC_STORE_1_TO_INT(pWtp->bThrdStateChanged); /* indicate change, so harverster will be called */
 
 	d_pthread_mutex_unlock(&pWtp->mut);
 	pthread_setcancelstate(iCancelStateSave, NULL);
@@ -335,75 +248,105 @@ wtiWorkerCancelCleanup(void *arg)
 }
 
 
+/* wait for queue to become non-empty or timeout
+ * helper to wtiWorker
+ * IMPORTANT: mutex must be locked when this code is called!
+ * rgerhards, 2009-05-20
+ */
+static inline void
+doIdleProcessing(wti_t *pThis, wtp_t *pWtp, int *pbInactivityTOOccured)
+{
+	struct timespec t;
+
+	BEGINfunc
+	DBGPRINTF("%s: worker IDLE, waiting for work.\n", wtiGetDbgHdr(pThis));
+	pWtp->pfOnIdle(pWtp->pUsr, MUTEX_ALREADY_LOCKED);
+
+	if(pWtp->toWrkShutdown == -1) {
+		/* never shut down any started worker */
+		d_pthread_cond_wait(pWtp->pcondBusy, pWtp->pmutUsr);
+	} else {
+		timeoutComp(&t, pWtp->toWrkShutdown);/* get absolute timeout */
+		if(d_pthread_cond_timedwait(pWtp->pcondBusy, pWtp->pmutUsr, &t) != 0) {
+			DBGPRINTF("%s: inactivity timeout, worker terminating...\n", wtiGetDbgHdr(pThis));
+			*pbInactivityTOOccured = 1; /* indicate we had a timeout */
+		}
+	}
+	ENDfunc
+}
+
+
 /* generic worker thread framework
  */
 #pragma GCC diagnostic ignored "-Wempty-body"
 rsRetVal
 wtiWorker(wti_t *pThis)
 {
-	DEFiRet;
-	DEFVARS_mutexProtection;
-	struct timespec t;
+	DEFVARS_mutexProtection_uncond;
 	wtp_t *pWtp;		/* our worker thread pool */
 	int bInactivityTOOccured = 0;
+	rsRetVal localRet;
+	rsRetVal terminateRet;
+	bool bMutexIsLocked;
+	DEFiRet;
 
 	ISOBJ_TYPE_assert(pThis, wti);
 	pWtp = pThis->pWtp; /* shortcut */
 	ISOBJ_TYPE_assert(pWtp, wtp);
 
 	dbgSetThrdName(pThis->pszDbgHdr);
-	pThis->pUsrp = NULL;
 	pthread_cleanup_push(wtiWorkerCancelCleanup, pThis);
 
-	BEGIN_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr, LOCK_MUTEX);
+	BEGIN_MTX_PROTECTED_OPERATIONS_UNCOND(pWtp->pmutUsr);
 	pWtp->pfOnWorkerStartup(pWtp->pUsr);
-	END_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr);
+	END_MTX_PROTECTED_OPERATIONS_UNCOND(pWtp->pmutUsr);
 
 	/* now we have our identity, on to real processing */
 	while(1) { /* loop will be broken below - need to do mutex locks */
-		/* process any pending thread requests */
-		wtpProcessThrdChanges(pWtp);
-
-		/* if we have a rate-limiter set for this worker pool, let's call it. Please
-		 * keep in mind that the rate-limiter may hold us for an extended period
-		 * of time. -- rgerhards, 2008-04-02
-		 */
-		if(pWtp->pfRateLimiter != NULL) {
+		if(pWtp->pfRateLimiter != NULL) { /* call rate-limiter, if defined */
 			pWtp->pfRateLimiter(pWtp->pUsr);
 		}
 		
 		wtpSetInactivityGuard(pThis->pWtp, 0, LOCK_MUTEX); /* must be set before usr mutex is locked! */
-		BEGIN_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr, LOCK_MUTEX);
-
-		if(  (bInactivityTOOccured && pWtp->pfIsIdle(pWtp->pUsr, MUTEX_ALREADY_LOCKED))
-		   || wtpChkStopWrkr(pWtp, LOCK_MUTEX, MUTEX_ALREADY_LOCKED)) {
-			END_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr);
-			break; /* end worker thread run */
+		BEGIN_MTX_PROTECTED_OPERATIONS_UNCOND(pWtp->pmutUsr);
+		bMutexIsLocked = TRUE;
+
+		/* first check if we are in shutdown process (but evaluate a bit later) */
+		terminateRet = wtpChkStopWrkr(pWtp, LOCK_MUTEX, MUTEX_ALREADY_LOCKED);
+		if(terminateRet == RS_RET_TERMINATE_NOW) {
+			/* we now need to free the old batch */
+			localRet = pWtp->pfObjProcessed(pWtp->pUsr, pThis);
+			dbgoprint((obj_t*) pThis, "terminating worker because of TERMINATE_NOW mode, del iRet %d\n",
+				 localRet);
+			break;
 		}
-		bInactivityTOOccured = 0; /* reset for next run */
 
-		/* if we reach this point, we are still protected by the mutex */
-
-		if(pWtp->pfIsIdle(pWtp->pUsr, MUTEX_ALREADY_LOCKED)) {
-			DBGPRINTF("%s: worker IDLE, waiting for work.\n", wtiGetDbgHdr(pThis));
-			pWtp->pfOnIdle(pWtp->pUsr, MUTEX_ALREADY_LOCKED);
-
-			if(pWtp->toWrkShutdown == -1) {
-				/* never shut down any started worker */
-				d_pthread_cond_wait(pWtp->pcondBusy, pWtp->pmutUsr);
-			} else {
-				timeoutComp(&t, pWtp->toWrkShutdown);/* get absolute timeout */
-				if(d_pthread_cond_timedwait(pWtp->pcondBusy, pWtp->pmutUsr, &t) != 0) {
-					DBGPRINTF("%s: inactivity timeout, worker terminating...\n", wtiGetDbgHdr(pThis));
-					bInactivityTOOccured = 1; /* indicate we had a timeout */
-				}
+		/* try to execute and process whatever we have */
+		/* This function must and does RELEASE the MUTEX! */
+		localRet = pWtp->pfDoWork(pWtp->pUsr, pThis, iCancelStateSave);
+		bMutexIsLocked = FALSE;
+
+		if(localRet == RS_RET_IDLE) {
+			if(terminateRet == RS_RET_TERMINATE_WHEN_IDLE) {
+				break;	/* end of loop */
+			}
+
+			if(bInactivityTOOccured) {
+				/* we had an inactivity timeout in the last run and are still idle, so it is time to exit... */
+				break; /* end worker thread run */
 			}
-			END_MTX_PROTECTED_OPERATIONS(pWtp->pmutUsr);
+			BEGIN_MTX_PROTECTED_OPERATIONS_UNCOND(pWtp->pmutUsr);
+			doIdleProcessing(pThis, pWtp, &bInactivityTOOccured);
+			END_MTX_PROTECTED_OPERATIONS_UNCOND(pWtp->pmutUsr);
 			continue; /* request next iteration */
 		}
 
-		/* if we reach this point, we have a non-empty queue (and are still protected by mutex) */
-		pWtp->pfDoWork(pWtp->pUsr, pThis, iCancelStateSave);
+		bInactivityTOOccured = 0; /* reset for next run */
+	}
+
+	/* if we exit the loop, the mutex may be locked and, if so, must be unlocked */
+	if(bMutexIsLocked) {
+		END_MTX_PROTECTED_OPERATIONS_UNCOND(pWtp->pmutUsr);
 	}
 
 	/* indicate termination */
@@ -413,8 +356,7 @@ wtiWorker(wti_t *pThis)
 
 	pWtp->pfOnWorkerShutdown(pWtp->pUsr);
 
-	wtiSetState(pThis, eWRKTHRD_TERMINATING, 0, MUTEX_ALREADY_LOCKED);
-	ATOMIC_STORE_1_TO_INT(pWtp->bThrdStateChanged); /* indicate change, so harverster will be called */
+	wtiSetState(pThis, eWRKTHRD_STOPPED, MUTEX_ALREADY_LOCKED);
 	d_pthread_mutex_unlock(&pThis->mut);
 	pthread_setcancelstate(iCancelStateSave, NULL);