Prev: iovecs don't print right
Next: Solaris 5.10 cpu hike --- _flockget --- possible deadlock ?
From: Eric Sosman on 4 Jul 2008 15:17
Raxit(a)mykavita.com wrote:
> [...]
> ff140090 lwp_park (0, 0, 0)
> ff12409c _flockget (ff16c3b8, c9610, 44274, ff117594, ff16c308,
> c9610) + 94
> ff118110 fprintf (c9640, a7ae8, f7afa558, 501b0, ff118460, ff168284)
> + 50
> 00059644 logstr
>
> many different threads (approx 100) executing/stcuk in above.

It looks like the hundred threads are all trying to write
a log message at the same time, and are patiently waiting for
their turn to use fprintf on the same FILE* stream. (Verify:
Check whether the first fprintf argument is the same in all
the stacks.) If that's the case, it may be that the CPU spike
is associated with whatever produces all that log activity.
Have you looked at the log messages?

--
Eric Sosman
esosman(a)ieee-dot-org.invalid
 | 
Pages: 1
Prev: iovecs don't print right
Next: Solaris 5.10 cpu hike --- _flockget --- possible deadlock ?