Hi Oleg, we've finally managed to freeze the log file after the spinning was first detected and this is the output of the spinning session:
https://gist.github.com/Turbo87/db7365bac349c7e56690 Tobias 2014-09-16 13:36 GMT+02:00 Oleg Kalnichevski <[email protected]>: > On Tue, 2014-09-16 at 13:08 +0200, Tobias Bieniek wrote: >> > If you do not want me to make wild guesses, please make available a >> > _complete_ context / header log of the session. >> >> I'll see what I can do. The problem is that with our production >> traffic on the server the log files are rotating very fast and it is >> hard to catch a log of one entire request session, but without the >> production traffic the problem doesn't seem to be reproducible. >> >> Since yesterday more of the I/O workers have started spinning again >> and are cluttering the log files now. >> >> I know it is wild guessing, but could the problem be related to >> streaming file downloads over SSL that are cancelled on the client >> side? It appears as if the SSL session might be staying in the CLOSING >> state forever with some data in the buffer but unable to flush it out. >> >> Tobias >> > > Tobias > > This is quite likely to be related to abnormal SSL connection > termination causing the I/O session to spin while trying to complete SSL > session handshake. But I need to see the session log to be 100% sure. > > Oleg > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
