Eventbuilder: Added evaluation of trb3 trigger type from first data
stream as "eventID" for event header (new
function SubEvt_trigType_trb3).This is read from bit 4..7 of decoding
word of subevent header. If this trb3 trigger type is non zero, it will
define eventID. If not, fallback to trb3 behaviour: eventID is evaluated
from CTS data word 2 with shift/mask procedure of SubEvt_trigType()
function.
Bugfix in openFile as suggested by Rafal Lalik:
Use additional fileNameFull string to compose full filename instead of
sprintf with identical source and dest string.
JAM: disabled that slave eventbuilder can close their run when exceeding 1.8 G limit.
This could lead to files with the "duplicate run id", which are not properly handled by oracle+ analysis. Instead, continue writing extra long files and put a warning in log after epics master has ended the run.
JAM: avoid duplilcate filenames in case of old runid (epics). to be tested.
bugfix: do not discard res events for all new files, only on process startup.
JAM: Added daq_evtbuild option --resbeginevent for resfile to skip first n events. Workaround to avoid QA monitor crashes in case of initial frontend rubbish.
hadaq [Fri, 25 Nov 2011 09:00:33 +0000 (09:00 +0000)]
JAM: improved fast cleanup of eventbuilder queues in case of too many trigger mismatch. Now complete double buffers are discarded. This works (always?), and is fast.
hadaq [Thu, 24 Nov 2011 18:09:02 +0000 (18:09 +0000)]
JAM: new approach to clean up eventbuilder queues when too many subevents were dropped. Actually, my previous try was very wrong, since queues were not cleared correctly. Still under development.
hadaq [Tue, 16 Aug 2011 10:54:41 +0000 (10:54 +0000)]
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
hadaq [Fri, 12 Aug 2011 20:49:45 +0000 (20:49 +0000)]
Changed log output for "popped subevents" case. Now frequency of warning is downscaled. Introduced again termination in this case after reaching another limit. Use defines for tuning the limits.
Some small code corrections.
JAM
hadaq [Wed, 10 Aug 2011 11:55:41 +0000 (11:55 +0000)]
Removed again wrong bugfix for subevent sync (nice try...).
Disabled bailing out when too many subevents are popped in one queue for a warning message.
Provided default value of 30% to ensure that a reasonable critical event discarding condition is used if not specified in command line.
hadaq [Wed, 10 Aug 2011 11:45:07 +0000 (11:45 +0000)]
Removed again wrong bugfix for subevent sync (nice try...).
Disabled bailing out when too many subevents are popped in one queue for a warning message.
Provided default value of 30% to ensure that a reasonable critical event discarding condition is used if not specified in command line.
hadaq [Tue, 9 Aug 2011 14:56:08 +0000 (14:56 +0000)]
Second try: Bugfix in eventbuilding: 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.
Try: Added free of shared mem and queue, since eb was stuck with previous version...
JAM
hadaq [Tue, 9 Aug 2011 11:33:20 +0000 (11:33 +0000)]
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 [Fri, 5 Aug 2011 08:31:39 +0000 (08:31 +0000)]
Changed message timeout for getRunId to 600s instead 2s.
Bugfix in rfio: when "no file" (-d null) was specified at startup, rfio was opened and written anyway. JAM (Joern Adamczewski-Musch)
jspc29 / daqdata.git / log