Main INDEX
Monthly INDEX
PREV
NEXT
User name R. Michaels
Log entry time 00:57:50 on June25,2000
Entry number 44845
keyword=re: delayed end-run entries
Here is some reasoning for why some end-run halog updates
are delayed by several hours. I credit M. Rvachev for figuring
out much of this, though nobody was able to solve it.
Problem seems to be with 'caget' used in scripts to get EPICS
variables. If you look at the process list on adaqs2, you typically
see at least one 'caget' command which is hanging. If you kill it
with 'kill -9', it will end (and data appear normal), but the next
'caget' command in the script's list will hang. Apparently, as long
as one 'caget' command is hanging, all others function. No particular
variable or network destination is affected; all are the same.
Now, every end-run we clean up processes by killing caget's, among
other things. So, after enough runs the entire list of variables is cycled
through, and the hung up end-run script finishes (and then the next one
gets hung up for hours).
We thought caget might get hung up because it has something to
print to stderr or stdout which it cannot inside the scripts. But piping
these to /dev/null did not help. Other things were tried too. If
someone has a good idea how to fix this, let me know (and not just
a band-aid fix).
Finally, I point out that all start-run, end-run, halog-com, and halog-end
files are written in obvious format to /home/adev/epics/runfiles and will
be saved at end of experiment. It is another (and trivial) database.