Using TCP keepalive on Azure

[amotl]

Dear @MarkusH, @chaudum, @lukasbals and @Taliik,

regarding the infamous Azure TCP idle connection drop problem (crate/crate#10779), I would like to give you a heads-up in the context of Kubernetes operators (Azure/AKS#1877).

According to dev-obs/actus#272, AKS clusters running Kubernetes v1.19+ are currently receiving a bugfix

The AKS control plane will always send RST for idle connections after 4min.

So, this will probably mitigate any problems in this regard for connections through kubernetes-asyncio.

With kind regards,
Andreas.

P.S.: However, I don't know whether any connections through aiopg might still be affected.

[amotl]

However, I don't know whether any connections through aiopg might still be affected.

On the psycopg2 side, this has been discussed within psycopg/psycopg2#113 and psycopg/psycopg2#561. On aiopg, aio-libs/aiopg#491 might be related.

In order to mitigate that, efforts on psycopg2 coming from psycopg/psycopg2#565 and psycopg/psycopg2#974 apparently have been wrapped into psycopg2-pool, see also liberapay/postgres.py#80.

So, while I want to apologize that I didn't look into how socket options can be set directly like crate/crate-python#374 when using aiopg, I am inclined to believe that the pool_recycle option of aiopg.create_pool() might help us here:

pool_recycle (float) – number of seconds after which connection is recycled, helps to deal with stale connections in pool, default value is -1, means recycling logic is disabled.

[amotl]

GH-548 addresses this. Thank you, @SStorm.