nginx access logs, timestamps "jump back"

Maxim Dounin mdounin at
Wed Mar 16 17:25:19 MSK 2011


On Tue, Mar 15, 2011 at 10:26:16PM -0400, assistlydavid wrote:

> Hi Maxim,
> Thanks for the response.
> How are the workers allocated work?  I'm still curious as to why 
> I'd see a large block of ELB-only traffic flushed to the log at 
> the same time, rather than a mix.  Any thoughts on that?

Workers accept() connections and process requests on accepted 

Depending on settings (notably accept_mutex) and number of already 
accepted connections conditions on which workers call accept() 
vary.  With accept_mutex turned on (default) it's likely that on 
underloaded servers all/most of new connections will be accepted by 
the same worker.

> Presumably, the chance of multiple workers flushing the same 
> type of log messages (ELB-only) at the same time is very low, 
> especially when we're seeing constant traffic from end-users 
> more frequently than ELB health checks.  Am I perhaps seeing all 
> ELB-only traffic flushed from the buffer of a single worker?  

Most likely health checks use single keepalive connection, and 
that's why all health checks go to the same worker.

And new client connections are currently accepted by another 

> Are workers chosen based on server/location directives?


Maxim Dounin

More information about the nginx mailing list