Alexander Kukushkin

https://www.postgresql.org/message-id/flat/CAPs%2BM8LCex6d%3DDeneofdsoJVijaG59m9V0ggbb3pOH7hZO4%2BcQ%40mail.gmail.com -- looks very similar

It is not a problem of postgre-operator or spilo. They only operate with the configuration you provide. If something is broken - you have to figure out what and fix...

If Patroni is killed with -9 something is terribly wrong. This is a clear indicator of OOM and my guess would be that you set `memory.limits` to low, probably something...

The Patroni issue you are referring to is actually a different one. There they are killing the postmaster process, not patroni. Maybe in your case OOM kills not only Patroni...

> This is strictly not a valid assumption - the only postgres processes in the spilo container will be those managed by patroni. There is a false assumption in your...

Yes, thanks to @hughcapet! now there is a [feature branch](https://github.com/zalando/spilo/tree/feature/postgresql-15) with Postgres 15, but you'll have to build the image on your own.

Oh, it seems that you have a lot of failovers... It is possible to reduce the number of history lines stored in the annotation by using **max_timelines_history** [parameter](https://patroni.readthedocs.io/en/latest/SETTINGS.html).

> Also, though less important, Why master service is not using selector to choose the master pod ? Because Patroni manages the endpoint and therefore can guaranty that there is...

> What a about using the pod names instead of IP Addresses ? Guess how many times people got into trouble due to misbehaving DNS? No thanks, I played this...

> and would allow the pooler to detect an IP change fixing https://github.com/zalando/postgres-operator/issues/1928 Not really. There always be a race condition and clients (or pgbouncer) will connect to the wrong...