Re: Squid loops on epoll_wait/read

 On Wed, 1 Jun 2011 16:31:45 -0700, Phil Oester wrote:
> I'm having a problem similar to the one described here (and in bug
> 3189):
>
>
> http://squid-web-proxy-cache.1019090.n4.nabble.com/Squid-loops-on-epoll-read-gettimeofday-td1695367.html
>
> where occasionally on a fresh start, squid (3.1.1[12]) will loop
> between epoll_wait and read as follows, using 100% cpu:
>
> epoll_wait(6, {{EPOLLIN, {u32=103, u64=7198046948199235687}}}, 8192,
> 55) = 1
> read(115, 0x7fffa78584f0, 256) = -1 EAGAIN (Resource
> temporarily unavailable)
> epoll_wait(6, {{EPOLLIN, {u32=103, u64=7198046948199235687}}}, 8192,
> 55) = 1
> read(115, 0x7fffa78584f0, 256) = -1 EAGAIN (Resource
> temporarily unavailable)
> epoll_wait(6, {{EPOLLIN, {u32=103, u64=7198046948199235687}}}, 8192,
> 55) = 1
> read(115, 0x7fffa78584f0, 256) = -1 EAGAIN (Resource
> temporarily unavailable)
>
> In the thread noted above, Henrik wondered why epoll_wait was
> returning
> one fd (fd 103), and read was acting upon another (fd 115). He asked
> for cachemgr output showing what each fd was being used for. In my
> case,
> both fd refer to similar items:
>
> 103 Pipe 0 0* 0 async-io
> completetion event: main
> 115 Pipe 0 0* 0 async-io
> completetion event: main
>
> Is it possible this is a simple off by one error within aufs?
>
> A stop/start on squid usually solves this, but occasionally the newly
> started
> instance has the same problem and I have to restart multiple times.
>
> Any ideas?

 The miss-spelling points straight at CommIO::Initialise() in
 src/comm.cc. Which makes that pipe the AIO callback.

 I think off-by-on there would possibly make AIO break completely.

 Could write events on a pipe() trigger an event signal on both FD?

 CommIO::DoneReadFD and CommIO::DoneFD need to be checked to see if they
 both match the expected FD above.

 Amos
Received on Thu Jun 02 2011 - 02:26:45 MDT