Increase gunicorn timeout and workers

[henhuy]

No description provided.

[nesnoj]

Did not solve hanging of app. Solution brought by fix in mapengine 111f422

[henhuy]

Increasing worker amount still could be a good solution - but if it works properly now, maybe we should keep it as is?!

[nesnoj]

Increasing worker amount still could be a good solution - but if it works properly now, maybe we should keep it as is?!

Yes, you're right.

The docs say
"Gunicorn relies on the operating system to provide all of the load balancing when handling requests. Generally we recommend (2 x $num_cores) + 1 as the number of workers to start off with."
which would result in 49 which is way too large according to other info I found.
This thread explains best, however there's no single truth and the optimal config depends on the number of requests, their duration and synchronicity, the docs recommend "tune using TTIN and TTOU signals while the application is under load". I'd like to avoid this finetuning.
So maybe --workers=4 is a good starting point.

And it might make sense to increase threads too.

Moreover, the docs recommend async workers for certain use cases, e.g. "Applications making long blocking calls (Ie, external web services)" to avoid processes being blocked. Except for WMS layers we won't have ext. resources and they are not supposed to be "long blocking calls". ESys optimization processes run in a separate container and checked by the app frequently, so they do not block the web processes, is this correct?
What do you think about synchronicity?

[henhuy]

So maybe --workers=4 is a good starting point.

Yep - maybe we start with this

Moreover, the docs recommend async workers for certain use cases, e.g. "Applications making long blocking calls (Ie, external web services)" to avoid processes being blocked. Except for WMS layers we won't have ext. resources and they are not supposed to be "long blocking calls". ESys optimization processes run in a separate container and checked by the app frequently, so they do not block the web processes, is this correct?
What do you think about synchronicity?

Yes, I think you are right - except for ES simulation, we don't need async stuff and ES is already using celery, so we are fine there IMHO.

[nesnoj]

ok, thx!