[Varnish] #835: Varnish stops receiving incoming connections, but the process is still up

Varnish varnish-bugs at varnish-cache.org
Mon Dec 20 14:56:06 CET 2010


#835: Varnish stops receiving incoming connections, but the process is still up
----------------------+-----------------------------------------------------
 Reporter:  blamer    |       Owner:  kristian                  
     Type:  defect    |      Status:  new                       
 Priority:  normal    |   Milestone:  Varnish 2.1 release       
Component:  varnishd  |     Version:  2.1.4                     
 Severity:  major     |    Keywords:  broken pipe, freeze, crash
----------------------+-----------------------------------------------------
Changes (by kristian):

  * owner:  phk => kristian
  * priority:  high => normal


Comment:

 Can you check if syslog contains any information on Varnish? And/or
 possibly run varnish with -d in the foreground in some way...

 It would also be interesting if you're able to attach gdb to varnish and
 do 'thread apply all bt full' after the fact.

 We're also somewhat curious about general disk performance.

 As for the vcl, you probably want to make sure "connection: close" is set
 for piped connections:

 {{{
 sub vcl_pipe {
 set bereq.http.connection = "close";
 }
 }}}

 (I doubt pipe is the issue here - but it's conceptually possible.)

 The most likely culprit is epoll, so you may want to test a different
 waiter: -p waiter=poll

 Also: How often does this happen? What's the request-pressure when it
 happens?

 Can you check what netstat says at the time?

 Also: ls /proc/`pidof -s varnishd`/fd/ is of some interest (mostly the
 size/number of fd's in use).

-- 
Ticket URL: <http://www.varnish-cache.org/trac/ticket/835#comment:2>
Varnish <http://varnish-cache.org/>
The Varnish HTTP Accelerator




More information about the varnish-bugs mailing list