Standby cluster bootsrap bug

[imcitius]

Hello
I’m trying to bootstrap standby cluster.
All seems to be ok, but wal-g does not see backups, because there is error in env variable: /run/etc/wal-e.d/env-standby/AWS_REGION contains string loca instead of correct string fr-par, but there is correct value in /run/etc/wal-e.d/env/AWS_REGION.

It seems for me as some templating bug inside Operator, or maybe Spilo.
I digged through Operator and Spilo sources, but didnt get the exact place where there envs is filled.

• Operator version v1.6.3
• Kubernetes
• Production
• Bug report

[apeschel]

The technical reason this happens is due to #1550

The postgres-operator isn't passing AWS_REGION to spilo, and so spilo falls back to attempting to query the AWS API to figure out what region it's in. If this query fails for some reason, there's no error checking, and the AWS_REGION string is set to "loca", which is part of the error string returned by the AWS API.

It seems to me the best fix would be to simply correctly pass AWS_REGION to spilo.

[imcitius]

The only way to pass AWS_REGION to cluster is postgres-pod-configmap which is single per namespace.
But I have few postgres clusters in same namespace, and some should have another AWS_REGION parameter, to backup properly.
Also there it no way to pass STANDBY_* envs, like we can with CLONE_* envs, so I'm stuck here.

[imcitius]

It's quietly disapponting to me that we cannot barely pass any Env vars to cluster pods directly in cluster CRD, and must envolve this very complex logic with variables at operator-CRD level, then variables on pod-configmap level, and in the end it does not work either.

[FxKu]

Passing STANDBY_* env variables through pod_environment_configmap should actually work. But nice idea to pass individual settings per cluster via env section in Postgres CRD.