Comment 28 for bug 802850

Thanks, either I or someone else will look at this in the next day or so.

Sent from my C64

On Oct 7, 2011, at 8:36 AM, Marc Easen <email address hidden> wrote:

> During high load I was seeing the same issue but this is what was happening:
>
> strace of worker:
>
> sendto(45, "\0REQ\0\0\0'\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> sendto(45, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> poll([{fd=45, events=POLLIN}], 1, 5000) = 1 ([{fd=45, revents=POLLIN}])
> getsockopt(45, SOL_SOCKET, SO_ERROR, [1459455931862482944], [4]) = 0
> sendto(45, "\0REQ\0\0\0'\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> recvfrom(45, "\0RES\0\0\0\n\0\0\0\0\0RES\0\0\0\n\0\0\0\0", 8192, 0, NULL, NULL) = 24
> sendto(45, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
>
>> From what I can see is the following
>
> 1) The worker is requesting work
> 2) The reponse of GEARMAN_LOST_CONNECTION is returned from the recv() function
> 3) The worker then sends a PRE_SLEEP command
> 4) The worker then request for work
> 4) Gearmand then responds with the response for two requests for jobs, which causes and odd spike in the server (I have located where this is happening)
> 5) The worker then send a PRE_SLEEP command
>
> I have fixed this issue by removing the break for a
> GEARMAN_LOST_CONNECTION status, along with saving the state of the
> worker, which allows it continue where it left off.
>
> (Please ignore the git references, I downloaded 0.24 and imported it
> into a git repo so I could track the changes)
>
>
> diff --git a/libgearman/worker.cc b/libgearman/worker.cc
> index 21488e0..745ca20 100644
> --- a/libgearman/worker.cc
> +++ b/libgearman/worker.cc
> @@ -672,17 +672,11 @@ gearman_job_st *gearman_worker_grab_job(gearman_worker_st *worker,
>
> if (gearman_failed(*ret_ptr))
> {
> - if (*ret_ptr == GEARMAN_IO_WAIT)
> - {
> - worker->state= GEARMAN_WORKER_STATE_GRAB_JOB_RECV;
> - }
> - else
> + worker->state= GEARMAN_WORKER_STATE_GRAB_JOB_RECV;
> + if (*ret_ptr != GEARMAN_IO_WAIT)
> {
> gearman_job_free(worker->job);
> worker->job= NULL;
> -
> - if (*ret_ptr == GEARMAN_LOST_CONNECTION)
> - break;
> }
>
> return NULL;
>
> --
> You received this bug notification because you are subscribed to
> Gearman.
> https://bugs.launchpad.net/bugs/802850
>
> Title:
> Gearman 100% cpu usage, workers in a loop (PHP, 0.22, 0.23)
>
> Status in Gearman Server and Client Libraries:
> New
>
> Bug description:
> After some time, minutes to hours, with a slight load on the gearman
> server (~1 job/min), workers get lost in a loop (per strace) and
> gearmand eats up 100% cpu.
>
> Strace of a worker:
> getsockopt(7, SOL_SOCKET, SO_ERROR, [117528996916232192], [4]) = 0
> sendto(7, "\0REQ\0\0\0\36\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> recvfrom(7, "\0RES\0\0\0\n\0\0\0\0", 8192, 0, NULL, NULL) = 12
> sendto(7, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> poll([{fd=7, events=POLLIN}], 1, 5000) = 1 ([{fd=7, revents=POLLIN}])
> getsockopt(7, SOL_SOCKET, SO_ERROR, [117528996916232192], [4]) = 0
> sendto(7, "\0REQ\0\0\0\36\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> recvfrom(7, "\0RES\0\0\0\n\0\0\0\0", 8192, 0, NULL, NULL) = 12
> sendto(7, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> poll([{fd=7, events=POLLIN}], 1, 5000) = 1 ([{fd=7, revents=POLLIN}])
> getsockopt(7, SOL_SOCKET, SO_ERROR, [117528996916232192], [4]) = 0
> sendto(7, "\0REQ\0\0\0\36\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> recvfrom(7, "\0RES\0\0\0\n\0\0\0\0", 8192, 0, NULL, NULL) = 12
> sendto(7, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> poll([{fd=7, events=POLLIN}], 1, 5000) = 1 ([{fd=7, revents=POLLIN}])
> getsockopt(7, SOL_SOCKET, SO_ERROR, [117528996916232192], [4]) = 0
> sendto(7, "\0REQ\0\0\0\36\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> recvfrom(7, "\0RES\0\0\0\n\0\0\0\0", 8192, 0, NULL, NULL) = 12
> sendto(7, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> poll([{fd=7, events=POLLIN}], 1, 5000) = 1 ([{fd=7, revents=POLLIN}])
> getsockopt(7, SOL_SOCKET, SO_ERROR, [117528996916232192], [4]) = 0
> sendto(7, "\0REQ\0\0\0\36\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> recvfrom(7, "\0RES\0\0\0\n\0\0\0\0", 8192, 0, NULL, NULL) = 12
> sendto(7, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> poll([{fd=7, events=POLLIN}], 1, 5000) = 1 ([{fd=7, revents=POLLIN}])
> getsockopt(7, SOL_SOCKET, SO_ERROR, [117528996916232192], [4]) = 0
> sendto(7, "\0REQ\0\0\0\36\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> recvfrom(7, "\0RES\0\0\0\n\0\0\0\0", 8192, 0, NULL, NULL) = 12
> sendto(7, "\0REQ\0\0\0\4\0\0\0\0", 12, MSG_NOSIGNAL, NULL, 0) = 12
> poll([{fd=7, events=POLLIN}], 1, 5000) = 1 ([{fd=7, revents=POLLIN}])
> ....
>
>
> Strace of gearmand:
>
>
> # strace -p 2820
> Process 2820 attached - interrupt to quit
> clock_gettime(CLOCK_MONOTONIC, {3794803, 727265822}) = 0
> epoll_wait(3,
>
> (... and nothing more.... )
>
>
> All workers are PHP based.
>
>
> # php --ri gearman
>
> gearman
>
> gearman support => enabled
> extension version => 0.8.0
> libgearman version => 0.22
> Default TCP Host => 127.0.0.1
> Default TCP Port => 4730
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/gearmand/+bug/802850/+subscriptions