From: hadaq <hadaq>
Date: Tue, 16 Aug 2011 10:54:41 +0000 (+0000)
Subject: Try: flush all queues when encountering too many discarded subevents at synchronizati... 
X-Git-Url: https://jspc29.x-matter.uni-frankfurt.de/git/?a=commitdiff_plain;h=ba3b9ab17a8edf46c29c8551bd10a4176baa193f;p=daqdata.git

Try: flush all queues when encountering too many discarded subevents at synchronization with master channel. May help system to recover when driven with high data rate?
JAM
---

diff --git a/hadaq/evtbuild.c b/hadaq/evtbuild.c
index d8efa84..c4fb600 100644
--- a/hadaq/evtbuild.c
+++ b/hadaq/evtbuild.c
@@ -1,4 +1,4 @@
-static char *rcsId = "$Header: /misc/hadesprojects/daq/cvsroot/eventbuilder/hadaq/evtbuild.c,v 6.153 2011-08-14 14:18:39 hadaq Exp $";
+static char *rcsId = "$Header: /misc/hadesprojects/daq/cvsroot/eventbuilder/hadaq/evtbuild.c,v 6.154 2011-08-16 10:54:41 hadaq Exp $";
 
 #define _POSIX_C_SOURCE 199309L
 #define SYSLOG_NAMES
@@ -53,8 +53,8 @@ static RFILE *fRemote = NULL;
 #define DAQVERSION 2			/* needed by Hydra unpackers */
 
 #define RUNID_INFO_DT  600		/* time in s until not available runid info is logged  */
-#define LOST_SUBEVT_LIMIT 50000	/* number of popped subevents per trigger sequence until warning */
-#define LOST_SUBEVT_TERM  20	/* terminate after this number of lost sub warnings per trigger sequence */
+#define LOST_SUBEVT_LIMIT 10000	/* number of popped subevents per trigger sequence until warning */
+#define LOST_SUBEVT_DROP  2		/* drop all queue contents after this number of lost sub warnings per trigger sequence */
 
 #define NEVTIDS 64UL			/* must be 2^n */
 #define NEVTIDS_IN_FILE 0UL		/* must be 2^n */
@@ -296,7 +296,7 @@ int isMismatchCritical(TheArgs *theArgs, TheStats *my)
 	int retVal = 0;
 
 	if (theArgs->maxTrigMismatch == -1) {
-		theArgs->maxTrigMismatch = 30;	/* JAM: set reasonable default if not specified */
+		theArgs->maxTrigMismatch = 90;	/* JAM: set reasonable default if not specified */
 		sprintf(msglog, "<I> Critical trigger mismatch ratio takes default of %d %% ", theArgs->maxTrigMismatch);
 		storeLogInfo(theArgs, msglog);
 		/*return retVal; */
@@ -874,7 +874,7 @@ static int rfio_closeConnection(TheArgs *theArgs)
 /* BUGBUG bailOut not proper yet */
 int main(int argc, char *argv[])
 {
-	int i;
+	int i, q;
 	TheArgs theArgsS, *theArgs = &theArgsS;
 	TheStats theStatsS, *theStats = &theStatsS;
 	TheDebug theDebugS, *theDebug = &theDebugS;
@@ -1302,9 +1302,21 @@ int main(int argc, char *argv[])
 						storeLogInfo(theArgs, msglog);
 						popTermCnt++;
 					}
-					if (popTermCnt > LOST_SUBEVT_TERM) {
-						storeLogInfo(theArgs, "<E> too many lost subevents. Exiting now!");
-						goto bailOut1;
+					if (popTermCnt > LOST_SUBEVT_DROP) {
+						/*  storeLogInfo(theArgs, "<E> too many lost subevents. Exiting now!");
+						   goto bailOut1; */
+
+						/* throw away all queues and start again: */
+						storeLogInfo(theArgs, "<E> too many lost subevents. Dropping all queues!");
+						for (q = 0; q < theArgs->nrOfMsgs; q++) {
+							HadTuQueue_clear(hadTuQueue[q]);
+							desHadTuQueue(hadTuQueue[q]);
+							free(hadTuQueue[q]);
+							hadTuQueue[q] = NULL;
+							ShmTrans_free(shmTrans[q]);
+						}
+						i = 0;	/* start again with evaluating other double buffer */
+						step = 1;
 					}
 				}
 			} else {