Worker Connections not being released

nathan.eloe

Hello all,
I have been developing an nginx module that allows the sending of zeromq
messages through nginx (inspired by https://github.com/FRiCKLE/ngx_zeromq).
The code base for this can be found here: https://github.com/neloe/ngx_zmq

However, I'm running into an issue where when running with a minimal
location configuration of

location /zmq/{
            zmq_endpoint "tcp://localhost:5555";
            zmq;
        }

I am testing the location with the following curl command:
curl -s localhost:8080/zmq/ -X POST -d "hello"

with 256 worker_connections, 1 worker_process, and keepalive_timeout of 0,
requests to the location will start failing after 254 requests.

Increasing to 1024 worker_connections does not solve the problem, but
allows me to send 1022 requests before the following begins appearing in my
error log:

2014/11/25 14:25:27 [alert] 10732#0: 1024 worker_connections are not enough

Explicitly disabling keepalive/lingering connections does not seem to solve
the problem; it feels like something somewhere is grabbing the worker
connection and never releasing it.

I have tried building the module against nginx 1.5.12, 1.7.4 and 1.7.7.
The older two versions are because they are the versions of openresty I am
trying to use this module in conjunction with.

Any guidance as to what would be causing these connections to never be
released would be very welcome.

Thank you,
Nathan Eloe

agentzh

Hello!

On Tue, Nov 25, 2014 at 12:52 PM, Nathan Eloe wrote:
> I have been developing an nginx module that allows the sending of zeromq
> messages through nginx (inspired by https://github.com/FRiCKLE/ngx_zeromq).

It's worth pointing out that this is the Chinese mailing list for
OpenResty. You're recommended to post to the English mailing list for
OpenResty instead:

https://groups.google.com/forum/#!forum/openresty-en

Regarding your question, it seems that you may

1. not get the r->main->count reference counting right (leading to
downstream connection leaks), and/or
2. not cleanup your upstream connection properly.

Regards,
-agentzh

nathan.eloe

I will move my query to that list, as I have further questions

On Tuesday, November 25, 2014 3:15:44 PM UTC-6, agentzh wrote:

Hello!

On Tue, Nov 25, 2014 at 12:52 PM, Nathan Eloe wrote:
> I have been developing an nginx module that allows the sending of zeromq
> messages through nginx (inspired by https://github.com/FRiCKLE/ngx_zeromq).

It's worth pointing out that this is the Chinese mailing list for
OpenResty. You're recommended to post to the English mailing list for
OpenResty instead:

https://groups.google.com/forum/#!forum/openresty-en

Regarding your question, it seems that you may

1. not get the r->main->count reference counting right (leading to
downstream connection leaks), and/or
2. not cleanup your upstream connection properly.

Regards,
-agentzh