Re: [Libevent-users] Deadlock driving me nuts

[Sherif Fanous <sherif.fanous@xxxxxxxxx>]

As a matter of fact, the thread in question is usually stuck in event_active_nolock.

The funny thing is that it shows that it's stuck in line 2212 in event.c which is the opening brace of the function.

How would I walk through ctx->events?

Thanks

Sherif

On Fri, Apr 8, 2011 at 9:38 PM, Nick Mathewson <nickm@xxxxxxxxxxxxx> wrote:

On Fri, Apr 8, 2011 at 3:37 AM, Sherif Fanous <sherif.fanous@xxxxxxxxx> wrote:
> Hello
> I'm running into a deadlock using libevent. I've been trying to figure out
> why or where it is happening for the last 3 days but have failed to do so.
> The deadlock occurs with the event base responsible for sending network
> packets. Below is a gdb output showing the deadlocked threads.

Thread 21 (Thread 27350):
#0  evmap_io_active (base=0x872b890, fd=247, events=36) at evmap.c:399
#1  0x080c84ae in epoll_dispatch (base=0x872b890, tv=0xb6e1c334) at epoll.c:436

So, this thread ought to be holding the event_base lock, but I don't
know why it would be freaking out like this.  The place that it seems
to be stuck on is:

       TAILQ_FOREACH(ev, &ctx->events, ev_io_next) {
               if (ev->ev_events & events)
                       event_active_nolock(ev, ev->ev_events & events, 1);
       }

I'm guessing that if you let it run a while, it stays stuck at the
same place? (And not, say, in event_active_nolock)?

So I can only think of two ways that could get stuck.  The first one
is if the ctx->events list had somehow gotten corrupted (say, if it
became circular somehow).  The second is if event_active_nolock had
somehow become stuck... but if that's the case, then I would expect
event_active_nolock to appear on your stack trace.

To see if I guessed right about the first case, could you try walking
through the ctx->events list via ev_io_next and seeing whether it
actually ends, or whether it's corrupted?

--
Nick