SPDY/1.3.4: strange connection buildup

[Valentin V. Bartenev]

On Monday 13 August 2012 14:40:07 Tom van der Woerdt wrote:
> Op 8/13/12 12:26 PM, Valentin V. Bartenev schreef:
> > On Monday 13 August 2012 14:11:41 Tom van der Woerdt wrote:
> >> Op 8/13/12 12:07 PM, Valentin V. Bartenev schreef:
> >>> On Monday 13 August 2012 13:12:18 Tom van der Woerdt wrote:
> >>>> Hi,
> >>>> 
> >>>> I've attached two Munin screenshots that show a strange issue I've
> >>>> been having with nginx. I'm running the 1.3.4 version with the SPDY
> >>>> patch (latest, 51) and it really looks like nginx sometimes forgets
> >>>> to close connections.
> >>> 
> >>> [...]
> >>> 
> >>> In case of SPDY these numbers are not real connections. It also counts
> >>> SPDY-streams. And it's pretty common when a browser opens about 10-20
> >>> streams per page.
> >>> 
> >>> But it is not ruled out, that there's may be a problem with the
> >>> counting.
> >> 
> >> If streams are counted instead of connections then the number is even
> >> more wrong.
> > 
> > Streams + connections.
> > 
> >> A stream gets closed immediately after the response has been
> >> sent. My server has no "slow" streams so there should be maybe 1 or 2
> >> streams visible on the graph, not 250.
> > 
> > Could you check the error log for segfault alerts? If worker process dies
> > then counters will stay in a wrong state.
> 
> [root at server nginx]# cat error.log* | grep alert
> 2012/08/12 18:52:18 [alert] 25293#0: worker process 32049 exited on
> signal 11
> 2012/08/10 23:02:13 [alert] 25293#0: worker process 16308 exited on
> signal 11
> 2012/08/09 17:55:50 [alert] 25293#0: worker process 25294 exited on
> signal 11
> 
[...]

Well, I was right, it's segfaulting.

Please, try a patch with the latest changes: 
http://nginx.org/patches/spdy/patch.spdy-52.txt

If this doesn't help, I will ask you to provide a core dump: 
http://wiki.nginx.org/Debugging#Core_dump

 wbr, Valentin V. Bartenev