Strange requests stalling inside nginx

[Maxim Dounin]

Hello!

On Sat, Feb 11, 2017 at 04:06:29PM -0500, MichaelLogutov wrote:

> Hello.
> We have some strange issues when requests seems to stall inside nginx - in
> nginx log we see that request took 1 second (and was terminated by client
> timeout), while excactly the same request (we have special unique headers to
> mark them) from the uwsgi logs took only 100ms to complete.

It is not clear how do you measure that request took 1 second.  
Note well that request time as seen by nginx is unlikely to be 
less that the time seen by your backend, but can easily be much 
longer: for example, nginx has to read the request from client, 
and it sometimes takes significant time.

> We also saw some very strange pauses in nginx debug log:

[...]

I don't see any "strange pauses" in the logs provided.  If you 
think there are any - please elaborate.

Note well that it's almost impossible to analize debug logs for 
pauses if these logs doesn't contain information about processing 
of events.  To make sure all needed information is present it is 
important to configure debug logs at global level, as recommended 
at http://nginx.org/en/docs/debugging_log.html.

-- 
Maxim Dounin
http://nginx.org/