maxwellby...@gmail.com wrote:
> Hi,
> 
> I'm running Starman 0.4014 behind nginx 1.10.0 on Ubuntu 16.04.1
> with Linux 4.4.0 (with nginx and Starman on separate servers).
> 
> I get immediate 502 errors from Nginx if all five Starman worker
> processes are busy, even if I have recently (re-)started the
> Starman master process.
> 
> As the --backlog option of Starman defaults to 1024,
> and /proc/sys/net/core/somaxconn is 128, shouldn't these
> requests be queued by Starman on the backend,
> rather than being rejected?

Maybe all 128 slots in the queue are immediately used up by
clients?  It happens on busy sites...

The ss(8) tool in Linux can show how much of the backlog you
use in the Recv-Q (2nd) column.

        ss -tl

(-t for TCP, -l for LISTEN)

You can leave out the -l to see the client sockets hitting
:6548, too.

> Configuration is through Plack::Runner:
> 
> 
> my %args = (
>   '--port' => 6548,    # 3000 would be the default
>   '--server' => "Starman",

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"psgi-plack" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to psgi-plack+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to