From 75e62f9447ed6b70fc90bffc067a0fbbeb5440d3 Mon Sep 17 00:00:00 2001
From: hadaq <hadaq>
Date: Tue, 9 Aug 2011 11:33:20 +0000
Subject: [PATCH] Bugfix in eventbuiding: when master stream had lost a trigger
 sequence number (possibly by udp layer), eventbuilder could not recover
 subevent synchronization -> increase of discarded events until regular
 program exit. This was observed at Aug2011 beamtime several times per day
 with cosmic and  test runs. Workaround with a delayed pop of master stream
 queue. JAM (Joern Adamczewski-Musch)

---
 hadaq/evtbuild.c | 52 ++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 46 insertions(+), 6 deletions(-)

diff --git a/hadaq/evtbuild.c b/hadaq/evtbuild.c
index 5fb2bf7..9de629c 100644
--- a/hadaq/evtbuild.c
+++ b/hadaq/evtbuild.c
@@ -1,8 +1,8 @@
-static char *rcsId = "$Header: /misc/hadesprojects/daq/cvsroot/eventbuilder/hadaq/evtbuild.c,v 6.147 2011-08-05 08:31:39 hadaq Exp $";
+static char *rcsId = "$Header: /misc/hadesprojects/daq/cvsroot/eventbuilder/hadaq/evtbuild.c,v 6.148 2011-08-09 11:33:20 hadaq Exp $";
 
 #define _POSIX_C_SOURCE 199309L
 #define SYSLOG_NAMES
-/* #define RFIO */
+#define RFIO
 
 #include <unistd.h>
 #include <getopt.h>
@@ -489,6 +489,8 @@ static int openFile(TheArgs *theArgs, TheStats *theStats)
 	static char outLustrePath[_POSIX_PATH_MAX];
 	static char sec_path[_POSIX_PATH_MAX];
 	static once = 1;
+	/*static unsigned long old_runNr=0;
+	   static unsigned open_count=0; */
 
 	seqNr = 0;
 
@@ -517,6 +519,26 @@ static int openFile(TheArgs *theArgs, TheStats *theStats)
 	iocTime = theArgs->runNr + TIMEOFFSET;
 	strftime(fileName + strlen(fileName), 18, "%y%j%H%M%S", localtime(&iocTime));
 
+
+
+	/* JAM: check here if new run id from epics is our old run id
+	 * this means we closed our file because of reaching file size limit
+	 * before master ioc assigned new run id. In this case, we need to modify file name!
+	 * Still disabled to ensure old naming convention for analysis scripts...
+	 * */
+	/*if(old_runNr!=0 && (theArgs->runNr > old_runNr))
+	   {
+	   open_count=0;  we have a real new run nr, reset file count for this run nr
+	   }
+	   else if(old_runNr!=0 && (theArgs->runNr ==old_runNr))
+	   {
+	   sprintf(msglog, "<I> openFile: begin next file %d for run id: %s", open_count,fileName);
+	   storeLogInfo(theArgs, msglog);
+	   }
+	   sprintf(fileName, "%s_%02d_", fileName, open_count++);  file name is always run id + file count
+	 */
+
+
 	/*   if ebnum == 0 then we assume that there is only 1 EB, 
 	 *   RUNID distribution by IOC is not needed and file name
 	 *   can be in the usual format without EB identificator.
@@ -860,6 +882,8 @@ int main(int argc, char *argv[])
 	uint32_t currTrigTag;
 	uint32_t currId;
 
+	uint32_t failTrigNr = 0;
+
 	signal(SIGINT, sigHandler);
 	signal(SIGTERM, sigHandler);
 	signal(SIGHUP, sigHandler);
@@ -1080,6 +1104,7 @@ int main(int argc, char *argv[])
 	unsigned long popCnt = 0;
 
 	currId = 0;
+	int popMaster = 0;
 
 	while (setjmp(terminateJmp) == 0) {
 		void *evt;
@@ -1157,7 +1182,7 @@ int main(int argc, char *argv[])
 
 		evt = newEvt(EvtDecoding_64bitAligned, EvtId_data);
 		popCnt = 0;				/* Counter of popped up subevents */
-
+		popMaster = 1;			/* flag to pop subevent from master stream i=0 */
 		for (i = 0; i < theArgs->nrOfMsgs && !evtIsBroken; i += step) {
 			uint32_t trigNr;
 			uint32_t trigTag;
@@ -1245,8 +1270,9 @@ int main(int argc, char *argv[])
 
 					evt = Evt_appendSubEvt(evt, subEvt);
 				}
-
-				HadTuQueue_pop(hadTuQueue[i]);
+				if (i > 0) {
+					HadTuQueue_pop(hadTuQueue[i]);	/* delay pop of master stream until broken checked */
+				}
 				step = 1;
 			} else if (trigNr < currTrigNr) {
 				if (theArgs->debugOptsCnt) {
@@ -1257,7 +1283,8 @@ int main(int argc, char *argv[])
 				/* BUGBUG subevt discarded, not in statistic */
 				HadTuQueue_pop(hadTuQueue[i]);
 				step = 0;
-
+				popMaster = 0;	/* do not pop subevent of master channel in this case. needed for next cycle */
+				failTrigNr = trigNr;	/* remember skipped trigger number for debug output */
 				popCnt++;
 				if (popCnt > 10000) {
 					if (theArgs->debugOptsCnt)
@@ -1278,7 +1305,20 @@ int main(int argc, char *argv[])
 				ShmTrans_free(shmTrans[i]);
 			}
 		}
+
+		/* New JAM: treat case that master stream has lost a packet */
+		if (popMaster) {
+			/* Regular case: delayed pop of last master stream entry after check ok */
+			HadTuQueue_pop(hadTuQueue[0]);
+		} else {
+			/* display error and keep last event in hope for later synchronization */
+			sprintf(msglog, "<W> Master message stream has missed trigger sequence nr:%d. Kept subevent nr %d in master queue",
+					failTrigNr, currTrigNr);
+			storeLogInfo(theArgs, msglog);
+		}
+
 		if (!evtIsBroken) {
+
 			if (theArgs->debugOptsCnt) {
 				if (evtBrokenFlag) {
 					Debug_print(theArgs, theStats, theDebug);
-- 
2.43.0