Major I've a problem with ODS 1.3.2 (or rather the 1.3-branch, as of rev. 5653,
but I've seen it for all 1.3-versions) running on a RHE 5.7 system.
The ods-signerd now and then (every second week or so) becomes stuck in a
CPU loop. Most of the threads use CPU. It looks as if the problem is
related to a mutex lock (futex, to be more specific)..
(The pthread_cond_timedwait call in ods_thread_wait in signer/src/shared/locks.c).
While stuck in the loop, it also keeps a lock on the kasp database.
I've some other processes (backups of the key database for example) that
wait for the same database lock. If it is a race condition when accessing
the kasp database or if it is something internal to ods-signerd is unclear.
It may even be a RHE Linux bug. It looks as if futex locks have had some
problems in earlier Linux versions (and programming using them is
tricky in any version).
I attach some gdb- and other output from the process and its threads. I
planned to leave ODS in the "loop" state to allow further info collection
but after some strace commands, the hang was resolved. However, the next
sign got stuck in a waitpid-call and the only way to resolve that was
to stop (ods-control stop and kill of some processes). Maybe some
internal state got confused by the long time in a CPU-bound loop.
Known problem? Linux problem? Some resource problem? Other?
/ Göran Bengtson
Chalmers Univ. of Technology