[Rivet-svn] r2847 - in trunk: . bin

[blackhole at projects.hepforge.org]

Author: buckley
Date: Sat Dec 11 21:03:58 2010
New Revision: 2847

Log:
Adding an 3600 second timeout for initialising an event file. If it takes longer
than (or anywhere close to) this long, chances are that the event source is
inactive for some reason (perhaps accidentally unspecified and stdin is not
active, or the event generator has died at the other end of the pipe. The reason
for not making it something shorter is that e.g. Herwig++ or Sherpa can have
long initialisation times to set up the MPI handler or to run the matrix element
integration. An timeout after an hour is still better than a batch job which
runs for two days before you realise that you forgot to generate any events!

Modified:
   trunk/ChangeLog
   trunk/bin/rivet

Modified: trunk/ChangeLog
==============================================================================
--- trunk/ChangeLog	Sat Dec 11 19:26:18 2010	(r2846)
+++ trunk/ChangeLog	Sat Dec 11 21:03:58 2010	(r2847)
@@ -1,3 +1,16 @@
+2010-12-11  Andy Buckley  <andy at insectnation.org>
+
+	* Adding an 3600 second timeout for initialising an event file. If
+	it takes longer than (or anywhere close to) this long, chances are
+	that the event source is inactive for some reason (perhaps
+	accidentally unspecified and stdin is not active, or the event
+	generator has died at the other end of the pipe. The reason for
+	not making it something shorter is that e.g. Herwig++ or Sherpa
+	can have long initialisation times to set up the MPI handler or to
+	run the matrix element integration. An timeout after an hour is
+	still better than a batch job which runs for two days before you
+	realise that you forgot to generate any events!
+
 2010-12-10  Andy Buckley  <andy at insectnation.org>
 
 	* Fixing unbooked-histo segfault in UA1_1990_S2044935 at 63 GeV.

Modified: trunk/bin/rivet
==============================================================================
--- trunk/bin/rivet	Sat Dec 11 19:26:18 2010	(r2846)
+++ trunk/bin/rivet	Sat Dec 11 21:03:58 2010	(r2847)
@@ -349,7 +349,7 @@
     if n % 10000 == 0:
         nevtloglevel = logging.CRITICAL
     timecurrent = time.time()
-    timeelapsed = timecurrent - starttime;
+    timeelapsed = timecurrent - starttime
     if maxevtnum is None:
         logging.log(nevtloglevel, "Event %d (%d s elapsed)" % (n, timeelapsed))
     else:
@@ -368,9 +368,10 @@
     a_up = a.upper()
     ## Print warning message and exit if not a valid analysis name
     if not a_up in all_analyses:
-        print "'%s' is not a valid analysis. Available analyses are:" % a_up
+        logging.warning("'%s' is not a valid analysis. Available analyses are:" % a_up)
         for aa in all_analyses:
-            print "    %s" % aa
+            logging.warning("    %s" % aa)
+            logging.warning("Exiting...")
         sys.exit(1)
     logging.debug("Adding analysis '%s'" % a_up)
     ah.addAnalysis(a_up)
@@ -387,34 +388,63 @@
 import platform
 logging.info("Rivet running on machine %s (%s)" % (platform.node(), platform.machine()))
 
+
+## Timeout handlers for the event initialisation and loop
+EVENT_TIMEOUT = 600
+def evtinithandler(signum, frame):
+    logging.warn("It has taken more than %d secs to get the first event! Is the input event stream working?" % EVENT_TIMEOUT)
+    raise Exception("Event initialisation timeout")
+# def evtloophandler(signum, frame):
+#     global evtnum, HEPMCFILES
+#     logging.warn("Event #%i timeout: it has taken more than %d secs to process this event! Is the input event stream working?" % (evtnum, EVENT_TIMEOUT))
+#     logging.warn("Abandoning processing of events from %s; is this input event stream actually working?" % hepmcfile)
+#     raise Exception("Event generation timeout")
+
+
 ## Init run based on one event
-evtfile = HEPMCFILES[0]
-if not run.init(evtfile):
-    logging.error("Failed to initialise on event file %s" % evtfile)
-    sys.exit(2)
+hepmcfile = HEPMCFILES[0]
+signal.signal(signal.SIGALRM, evtinithandler)
+signal.alarm(EVENT_TIMEOUT)
+try:
+    init_ok = run.init(hepmcfile)
+    if not init_ok:
+        logging.error("Failed to initialise on event file %s... exiting" % hepmcfile)
+        sys.exit(2)
+except:
+    logging.error("Timeout in initialisation from event file %s... exiting" % hepmcfile)
+    sys.exit(3)
+
 
 ## Event loop
-starttime = time.time()
 evtnum = 0
-for fileidx in range(len(HEPMCFILES)):
-    logging.info("Reading events from '%s'" % HEPMCFILES[fileidx])
+starttime = time.time()
+for fileidx, hepmcfile in enumerate(HEPMCFILES):
+    ## Open next HepMC file (does not apply to first file: it was already used for the run init)
+    if fileidx > 0:
+        run.openFile(hepmcfile)
+        if not run.readEvent():
+            logging.warning("Could not read events from '%s'" % hepmcfile)
+            continue
+    logging.info("Reading events from '%s'" % hepmcfile)
     while opts.MAXEVTNUM is None or evtnum < opts.MAXEVTNUM:
         evtnum += 1
         logNEvt(evtnum, starttime, opts.MAXEVTNUM)
-        if not run.processEvent():
+        ## Process this event
+        processed_ok = run.processEvent()
+        if not processed_ok:
             logging.warn("Event processing failed for evt #%i!" % evtnum)
             break
+        ## Exit the loop if signalled
         if RECVD_KILL_SIGNAL is not None:
             break
-        if not run.readEvent():
+        ## Read next event
+        read_ok = run.readEvent()
+        if not read_ok:
             break
+        ## Write a histo file snapshot if appropriate
         if opts.HISTO_WRITE_INTERVAL is not None:
             if evtnum % opts.HISTO_WRITE_INTERVAL == 0:
                 ah.writeData(opts.HISTOFILE)
-    if fileidx < len(HEPMCFILES)-1:
-        run.openFile(HEPMCFILES[fileidx+1])
-        if not run.readEvent():
-            continue
 logging.info("Finished event loop")
 run.finalize()