/********************************************************************* * * * antisender.c * * * * By Andrew L. Stachyra * * and * * Yoshinari Hayato * * * * [currently maintained/updated by Hans Berns] * * * * This program is designed to send data from the SuperKamiokande * * outer detector to a program called event_builder, written by * * Hayato-san, which runs on the main host computer. * * * *********************************************************************/
(reverse order: last update first)
/********************************************************************* * * Version history (reverse order, last comes first): * * 030818 HGB: - new "delay" number isn't working as expected => fixed. * * 030817 HGB: - add column "delay" in sendperf.run_XXXXX logfile to * record&debug collector->sender processing delays, where * "delay" = current local time - last event (gps)time. * * 030319 HGB: - modified the use of errlogfp: now fopen() and fclose() * only when needed, instead of leaving open for the * entire run. * - better limits on repeated "Event count did not increment" * and "Event count jumped" messages, to avoid error * message flooding. * * 030218 HGB: - added stop_evbuild flag to check for evbuild_ready flag, * requested by Shiozawa-san. Hopefully this will eliminate * occasional run end error messages from the event builder. * * 021115 HGB: - changed nblab_file path to /home/online/flags/ * * 021016 HGB: - log ANTI/ODPED/GPS type count in summary file * * 021001 HGB: - added ODPED_DATA_TYPE=52 to process OD pedestal events * * 020929 HGB: - online/net_shm header change requires recompile * * 010223 HGB: - now ignores invalid event builder request (i.e. if * event_requested_to = 0), instead of sending empty block. * Hopefully, that'll finally fix occasional missing OD data * at beginnings of new runs after stop/start. * - additional shift instruction for errsend messages. * * 010108-14: - additional log messages if initial event builder request * is invalid (e.g. leftover from previous run). * * 001025 HGB: - reset net_shm->stab[].from/.to at SETUP phase. * * 001023 HGB: - added checking condition * (event_ctr >= net_shm->stab[server_number].from) * at beginning of a new run. Some runs found to start * with eventbuilder request of final event block of * the previous run. This caused the reformatter to * miss OD events for 1-3 minutes after run start. * * 001018/22: - additional debug logs at run beginning (first 40 blocks * sent to event builder) [response to Hayato's finding of * missing OD data at start of some runs]. * * 000410 HGB: - removed oh_shit script call... * * 000404 HGB: - moved stop_func() after inner while() loop is finished * and added new flag stop_request instead. now the sender * should be able to catch the very final eventbuilder * request, after final nova buffer acknowledged, before * returning to init state. * * 000320/24: - put request() calls in send_next_events() into new * subsection update_net_shm() and rearranged order of * statistics updater to minimize send/request times with * event builder. * - make header/trailer handling in send_data() faster. * - after receiving final nova buffer, adjust final event * to last event requested via net_shm. * * 000317 HGB: - changed MIN_OBSRV_INTERVAL to variable min_obsrv_events * for user selectable heartbeat rate (for future when * SLE rate will be increased again). * * 000313 HGB: - added sendtime.run_XXXX log file for debugging sender's * transfer rate and throughput. * - rearranged summary statistics output * * 000312 HGB: - added FILL_BLANK_HEADERS flag to gpshead_flag to select * whether to generate fake blank headers or no headers * for non-split trigger data. * * 000111 HGB: - added ERR_LOG and PERF_LOG control variables to avoid * using fprintf() to file pointers errlogfp or perflog if * they're neither open nor NULL pointers. Precautionary * change after anticollector problems with file NULL pointer * were discovered. * * 991007 HGB: - adjusted heart-beat rate (MIN_OBSRV_INTERVAL) to 1000 * events and MIN_EVENT_JUMP to 5000 to adjust to recently * increased Super-SLE rate. * * 990812 HGB: - write_summary_file() now for first 5 buffers, too. * * 990810 HGB: - in write_summary_file() replaced net_shm->run_num with * screwed_up_anti_header.run_number so that subroutine will * correctly finish even when there's a network error. * - added printing "*" when in while() setup loops with sleep(). * * 990806 HGB: - reorganized fatal error messages to new subroutine * send_error_messages(), and changed all msgsend() call * before an exit() to errsend(). * * 990730 HGB: - additional error messages if fatal errors are detected * before exit() is called. Better for debugging... * * 990724 HGB: - changed heartbeat frequency to adapt to higher SLE vs. * OD/HE trigger rate; now MIN_OBSRV_INTERVAL = 300. * - place "H" in heartbeat count if only ANTI/GPS header * without TDC data. * * 990504/5: - Network errors found to be mostly fatal. Therefore, * some changes with error message display and summary file. * - changed nblab parametring slightly: squeezed in level 2 * to display NONFATAL_ERR in addition to level 1; levels >2 * moved one level up; so, now levels 0...7 are valid. * * 990503 HGB: - possible cause for occasional segmentatin fault crahes: * send_next_event() seems to be called via signal interrupt * even if stop notification has already been acknowledged. * quick fix: added if(!stop_flag) into send_next_events(). * * 990426 HGB: - added debugging lines before and after stop_func() * is called to trace why the sender still occasionally * crashes with segmentation fault. * * 990423 HGB: - in inner while() loop of main() placed an if(!stop_flag) * to avoid calling read_new_nova_buffer() twice if * final NOVA buffer has already been received. * * 990215/18: - in get_sendlength() take into account GPS headers * from sorter if GPSHEAD is not selected, otherwise * the sender will interpret a GPS header as valid ANTI * data and complain about non-incrementing event numbers. * * 990214 HGB: - added on-the-fly option to change output data from * ANTI-DATA-only to ANTI_DATA-with-GPS_HEADER via flag * file /home/online/flag/gpshead_flag * * 990207 HGB: - a little code cleanup (introducing Summary structure) * and write_summary_file() * * 990205 HGB: - added gps_header datastructure and changed all sections * to deal with additional GPS data at end of each event * as transfered by the new sorter/collector * * 980501 HGB: - changed inner while(1) loop to while(!stop_func) * and removed the if(stop_func==TRUE) section at end * of loop. [I hate "break;" statements!!]. * - removed histogramming of event gap statistics since * it turned out to be useless. * * 980429 HGB: - added one final error message for case of "imperfect" run * (forgot at last version). * * 980212 HGB: - added another variable to histogram.event_gaps for * debugging * * 980205 HGB: - added more debugging features: * - timestamps to performance log file * - histogram file of event gap statistics. * * 980202 HGB: - nova_filled_events had no data -> corrected. * - slight changes in read_new_nova_buffer(). * * 980128 HGB: - changed the NBLAB decision tree and error * messaging slightly. * - introducing log file sendperf.run_XXXX * * 980119 HGB: - start of code cleanup and re-arrangement * * 970710 AS: I changed the heartbeat so that it prints a "." if * substantive data is being passed to the event builder, * and it prints a "_" if empty headers are sent to the * event bulider. Additionally, it prints an "X" if the * low-level sockets routine that sends data ever fails, * although this should be extremely rare. I also shortened * the end-of-run summary message in the errorlog file. * * 970625 AS: I gave the sender the ability to recognize "empty" nova * buffers. These are buffers that do not contain any useful * data. The sorter communicates to the sender that the * buffer is garbage by setting its length equal to * EMPTY_NOVA_BUFFER bytes. * * 970529 AS: I added a "heartbeat" for when the nblab variable is set * really low, so that at a bare minimum one always gets a * dot printed to the screen every so often. * Also I reduced and eliminated some of the error messaging * associated with filling in events for which there is no * data, because it is now the norm for the new split-trigger * configuration that we will be using regularly very soon. * * 970212b AS: I changed one small thing about the errorlog file so that * it makes more intuitive sense. * * 970212 AS: I added in some counters to give us some stats, and * revamped the performance/errorfile, as well as added a few * comments. * * 970211 AS: I added in code to control error messaging. This code * uses a variable called "nblab" to decide which messages to * print. * * 961230 AS: I removed the line which defines DO_ACK at Shiozawa-san's * request. This will increase the rate of data transfer by * eliminating some hand-shaking process. * * 960822 JG: antisender.c.aug22 * ************************************************************************/
