[darcs-users] test_network failing with deadlock

Dmitry Kurochkin dmitry.kurochkin at gmail.com
Fri Aug 15 08:16:00 UTC 2008


Hi Karel.

On Fri, Aug 15, 2008 at 11:39 AM, Karel Gardas
<kgardas at objectsecurity.com> wrote:
>
> Hello,
>
> few days ago Zooko asked for curl versions on my Solaris and OpenBSD
> buildbots. I've replied that not only those are affected by the issue in
> test_network test. Also linuxes are affected. Today I see that heffalump
> Linux-? Debian-lenny(ish) ? also do have problem with running
> test_network for a long time.
>
> Anyway, at least on my Solaris where I'm able to use pstack utility to
> get a stack of all threads of running process, the state of such darcs
> running test_network indefinitely looks:
>
> 24154:  /buildbot/workspace/karel
> solaris/build/tests-network.dir/../darcs get
> -----------------  lwp# 1 / thread# 1  --------------------
>  fee5f559 lwp_park (0, 0, 0)
>  fee59656 cond_wait_queue (85a426c, 85a427c, 0, 0) + 41
>  fee59b3a _cond_wait (85a426c, 85a427c) + 69
>  fee59b78 cond_wait (85a426c, 85a427c) + 24
>  fee59bb2 pthread_cond_wait (85a426c, 85a427c, 85a427c, 0, fe772000,
> fe912a00) + 1e
>  084fc3e0 waitCondition (0, 0, fe772000, 0, 0, 10300) + 10
>  08598860 MainCapability ()
> -----------------  lwp# 2 / thread# 2  --------------------
>  fee635a5 pollsys  (fe6fbe80, 1, fe6fbf18, 0)
>  fee1a16e pselect  (4, fe77e018, fe77e0a0, feef8608, fe6fbf18, 0) + 19e
>  fee1a47e select   (4, fe77e018, fe77e0a0, 0, fe77e128) + 7e
>  0849abce ???????? (fe7862a6, 4, fe77e0a0, fe77e018, fe7862b8, 849b1f8)
>  fe786316 ???????? (75fffe83, 45c722, 849aa9c, e9f4c783, 21b80, c7043c7)
>  14775c7b ???????? ()
> -----------------  lwp# 3 / thread# 3  --------------------
>  fee5f559 lwp_park (0, 0, 0)
>  fee59656 cond_wait_queue (85a5df4, 85a5e04, 0, 0) + 41
>  fee59b3a _cond_wait (85a5df4, 85a5e04) + 69
>  fee59b78 cond_wait (85a5df4, 85a5e04) + 24
>  fee59bb2 pthread_cond_wait (85a5df4, 85a5e04, 0, 8598953, 85bcf30,
> fe8c0a00) + 1e
>  084fc3e0 waitCondition (0, 0, 0, 0, 0, 10300) + 10
>  08598860 MainCapability ()
> -----------------  lwp# 5 / thread# 5  --------------------
>  fee5f559 lwp_park (0, 0, 0)
>  fee59656 cond_wait_queue (85bcf4c, 85bcf5c, 0, 0) + 41
>  fee59b3a _cond_wait (85bcf4c, 85bcf5c) + 69
>  fee59b78 cond_wait (85bcf4c, 85bcf5c) + 24
>  fee59bb2 pthread_cond_wait (85bcf4c, 85bcf5c, 0, 0, 0, fe8c1a00) + 1e
>  084fc3e0 waitCondition (0, 0, 0, 0, 0, 10300) + 10
>  08598860 MainCapability ()
>
>
> when run over the time, stack trace differer only in one line:
>
> --- /tmp/s1.txt Fri Aug 15 08:47:13 2008
> +++ /tmp/s4.txt Fri Aug 15 09:38:02 2008
> @@ -11,7 +11,7 @@
>  fee635a5 pollsys  (fe6fbe80, 1, fe6fbf18, 0)
>  fee1a16e pselect  (4, fe77e018, fe77e0a0, feef8608, fe6fbf18, 0) + 19e
>  fee1a47e select   (4, fe77e018, fe77e0a0, 0, fe77e128) + 7e
> - 0849abce ???????? (fe7862a6, 4, fe77e0a0, fe77e018, fe7862b8, 849b1f8)
> + 0849abce ???????? (fe78640e, 4, fe77e0a0, fe77e018, fe786420, 849b1f8)
>  fe786316 ???????? (75fffe83, 45c722, 849aa9c, e9f4c783, 21b80, c7043c7)
>  14775c7b ???????? ()
>  -----------------  lwp# 3 / thread# 3  --------------------
>
>
> but this is rather not so common change, since for example for last hour
> the stack trace looks the same.
>
> Any idea what's going wrong here? My still bet is something bad going
> either in curl or in darcs/curl interaction...

>From the stacks everything looks normal to me. I have just run "make
test_network" on debian unstable and it completed successfully. I
think what can help us understand the problem is:
1. determine which command hangs (just run ps).
2. reproduce it outside of test_network with --debug and --debug-http
options. That would hopefully give us a hint on what is going wrong.

Regards,
  Dmitry

>
> Thanks,
> Karel
> --
> Karel Gardas                  kgardas at objectsecurity.com
> ObjectSecurity Ltd.           http://www.objectsecurity.com
> _______________________________________________
> darcs-users mailing list
> darcs-users at darcs.net
> http://lists.osuosl.org/mailman/listinfo/darcs-users
>


More information about the darcs-users mailing list