summaryrefslogtreecommitdiffstats
path: root/runtime/queue.c
diff options
context:
space:
mode:
Diffstat (limited to 'runtime/queue.c')
-rw-r--r--runtime/queue.c77
1 files changed, 71 insertions, 6 deletions
diff --git a/runtime/queue.c b/runtime/queue.c
index 280ebd94..bdd57f99 100644
--- a/runtime/queue.c
+++ b/runtime/queue.c
@@ -12,7 +12,7 @@
* function names - this makes it really hard to read and does not provide much
* benefit, at least I (now) think so...
*
- * Copyright 2008, 2009 Rainer Gerhards and Adiscon GmbH.
+ * Copyright 2008-2011 Rainer Gerhards and Adiscon GmbH.
*
* This file is part of the rsyslog runtime library.
*
@@ -83,6 +83,11 @@ static rsRetVal ConsumerDA(qqueue_t *pThis, wti_t *pWti);
static rsRetVal batchProcessed(qqueue_t *pThis, wti_t *pWti);
static rsRetVal qqueueMultiEnqObjNonDirect(qqueue_t *pThis, multi_submit_t *pMultiSub);
static rsRetVal qqueueMultiEnqObjDirect(qqueue_t *pThis, multi_submit_t *pMultiSub);
+static rsRetVal qAddDirect(qqueue_t *pThis, void* pUsr);
+static rsRetVal qDestructDirect(qqueue_t __attribute__((unused)) *pThis);
+static rsRetVal qConstructDirect(qqueue_t __attribute__((unused)) *pThis);
+static rsRetVal qDelDirect(qqueue_t __attribute__((unused)) *pThis);
+static rsRetVal qDestructDisk(qqueue_t *pThis);
/* some constants for queuePersist () */
#define QUEUE_CHECKPOINT 1
@@ -592,6 +597,47 @@ static rsRetVal qDelLinkedList(qqueue_t *pThis)
/* -------------------- disk -------------------- */
+/* The following function is used to "save" ourself from being killed by
+ * a fatally failed disk queue. A fatal failure is, for example, if no
+ * data can be read or written. In that case, the disk support is disabled,
+ * with all on-disk structures kept as-is as much as possible. Instead, the
+ * queue is switched to direct mode, so that at least
+ * some processing can happen. Of course, this may still have lots of
+ * undesired side-effects, but is probably better than aborting the
+ * syslogd. Note that this function *must* succeed in one way or another, as
+ * we can not recover from failure here. But it may emit different return
+ * states, which can trigger different processing in the higher layers.
+ * rgerhards, 2011-05-03
+ */
+static inline rsRetVal
+queueSwitchToEmergencyMode(qqueue_t *pThis, rsRetVal initiatingError)
+{
+ pThis->iQueueSize = 0;
+ pThis->nLogDeq = 0;
+ qDestructDisk(pThis); /* free disk structures */
+
+ pThis->qType = QUEUETYPE_DIRECT;
+ pThis->qConstruct = qConstructDirect;
+ pThis->qDestruct = qDestructDirect;
+ pThis->qAdd = qAddDirect;
+ pThis->qDel = qDelDirect;
+ pThis->MultiEnq = qqueueMultiEnqObjDirect;
+ if(pThis->pqParent != NULL) {
+ DBGOPRINT((obj_t*) pThis, "DA queue is in emergency mode, disabling DA in parent\n");
+ pThis->pqParent->bIsDA = 0;
+ pThis->pqParent->pqDA = NULL;
+ /* This may have undesired side effects, not sure if I really evaluated
+ * all. So you know where to look at if you come to this point during
+ * troubleshooting ;) -- rgerhards, 2011-05-03
+ */
+ }
+
+ errmsg.LogError(0, initiatingError, "fatal error on disk queue '%s', emergency switch to direct mode",
+ obj.GetName((obj_t*) pThis));
+ return RS_RET_ERR_QUEUE_EMERGENCY;
+}
+
+
static rsRetVal
qqueueLoadPersStrmInfoFixup(strm_t *pStrm, qqueue_t __attribute__((unused)) *pThis)
{
@@ -794,10 +840,7 @@ finalize_it:
static rsRetVal qDeqDisk(qqueue_t *pThis, void **ppUsr)
{
DEFiRet;
-
- CHKiRet(obj.Deserialize(ppUsr, (uchar*) "msg", pThis->tVars.disk.pReadDeq, NULL, NULL));
-
-finalize_it:
+ iRet = obj.Deserialize(ppUsr, (uchar*) "msg", pThis->tVars.disk.pReadDeq, NULL, NULL);
RETiRet;
}
@@ -1312,6 +1355,7 @@ static int qqueueChkDiscardMsg(qqueue_t *pThis, int iQueueSize, void *pUsr)
if(iRetLocal == RS_RET_OK && iSeverity >= pThis->iDiscardSeverity) {
DBGOPRINT((obj_t*) pThis, "queue nearly full (%d entries), discarded severity %d message\n",
iQueueSize, iSeverity);
+ STATSCOUNTER_INC(pThis->ctrNFDscrd, pThis->mutCtrNFDscrd);
objDestruct(pUsr);
ABORT_FINALIZE(RS_RET_QUEUE_FULL);
} else {
@@ -1693,7 +1737,18 @@ ConsumerReg(qqueue_t *pThis, wti_t *pWti)
ISOBJ_TYPE_assert(pThis, qqueue);
ISOBJ_TYPE_assert(pWti, wti);
- CHKiRet(DequeueForConsumer(pThis, pWti));
+ iRet = DequeueForConsumer(pThis, pWti);
+ if(iRet == RS_RET_FILE_NOT_FOUND) {
+ /* This is a fatal condition and means the queue is almost unusable */
+ d_pthread_mutex_unlock(pThis->mut);
+ DBGOPRINT((obj_t*) pThis, "got 'file not found' error %d, queue defunct\n", iRet);
+ iRet = queueSwitchToEmergencyMode(pThis, iRet);
+ // TODO: think about what to return as iRet -- keep RS_RET_FILE_NOT_FOUND?
+ d_pthread_mutex_lock(pThis->mut);
+ }
+ if (iRet != RS_RET_OK) {
+ FINALIZE;
+ }
/* we now have a non-idle batch of work, so we can release the queue mutex and process it */
d_pthread_mutex_unlock(pThis->mut);
@@ -1943,6 +1998,13 @@ qqueueStart(qqueue_t *pThis) /* this is the ConstructionFinalizer */
CHKiRet(statsobj.AddCounter(pThis->statsobj, UCHAR_CONSTANT("full"),
ctrType_IntCtr, &pThis->ctrFull));
+ STATSCOUNTER_INIT(pThis->ctrFDscrd, pThis->mutCtrFDscrd);
+ CHKiRet(statsobj.AddCounter(pThis->statsobj, UCHAR_CONSTANT("discarded.full"),
+ ctrType_IntCtr, &pThis->ctrFDscrd));
+ STATSCOUNTER_INIT(pThis->ctrNFDscrd, pThis->mutCtrNFDscrd);
+ CHKiRet(statsobj.AddCounter(pThis->statsobj, UCHAR_CONSTANT("discarded.nf"),
+ ctrType_IntCtr, &pThis->ctrNFDscrd));
+
pThis->ctrMaxqsize = 0; /* no mutex needed, thus no init call */
CHKiRet(statsobj.AddCounter(pThis->statsobj, UCHAR_CONSTANT("maxqsize"),
ctrType_Int, &pThis->ctrMaxqsize));
@@ -2328,6 +2390,7 @@ doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr)
STATSCOUNTER_INC(pThis->ctrFull, pThis->mutCtrFull);
if(pThis->toEnq == 0 || pThis->bEnqOnly) {
DBGOPRINT((obj_t*) pThis, "enqueueMsg: queue FULL - configured for immediate discarding.\n");
+ STATSCOUNTER_INC(pThis->ctrFDscrd, pThis->mutCtrFDscrd);
objDestruct(pUsr);
ABORT_FINALIZE(RS_RET_QUEUE_FULL);
} else {
@@ -2339,6 +2402,7 @@ doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr)
timeoutComp(&t, pThis->toEnq);
if(pthread_cond_timedwait(&pThis->notFull, pThis->mut, &t) != 0) {
DBGOPRINT((obj_t*) pThis, "enqueueMsg: cond timeout, dropping message!\n");
+ STATSCOUNTER_INC(pThis->ctrFDscrd, pThis->mutCtrFDscrd);
objDestruct(pUsr);
ABORT_FINALIZE(RS_RET_QUEUE_FULL);
}
@@ -2480,6 +2544,7 @@ DEFpropSetMeth(qqueue, iLowWtrMrk, int)
DEFpropSetMeth(qqueue, iDiscardMrk, int)
DEFpropSetMeth(qqueue, iFullDlyMrk, int)
DEFpropSetMeth(qqueue, iDiscardSeverity, int)
+DEFpropSetMeth(qqueue, iLightDlyMrk, int)
DEFpropSetMeth(qqueue, bIsDA, int)
DEFpropSetMeth(qqueue, iMinMsgsPerWrkr, int)
DEFpropSetMeth(qqueue, bSaveOnShutdown, int)
generated by cgit (git 2.34.1) at 2024-05-03 18:41:07 +0000