Fix the Operator rolling update statefulsets unnecessary whien Kube A…

[machine424]

…PI down (#2031)

I think the refactor in https://github.com/zalando/postgres-operator/pull/1848 has introduced this regression, do you think it was intended?

Before the refactor, getPodEnvironmentSecretVariables and getPodEnvironmentConfigMapVariables were called directly in generateStatefulSet and we used to return an error (and halt the sync) instead of just logging the error and continue sync with missing data.

Maybe add a test and adjust documentation somewhere.

[machine424]

Maybe @FxKu as he took care of the refactor.

[FxKu]

@machine424 can you rebase? We merged #2045 which introduced some small changes in the same code parts.

[machine424]

@FxKu, done.

Don't you think we should also release a 1.8.x fix as it is a regression? We have multiple clusters that restart for nothing (we use etcd as dcs to avoid relying on the Kube API) and we'd like to hotfix them.

[FxKu]

:+1:

[idanovinda]

👍

[FxKu]

Thanks again @machine424. It will be included in the next release (should be there soon). There are too many changes already merged to do another 1.8.x release. In the meantime, you could try to patch 1.8.2 yourself with your fix and use a local image.

[machine424]

@FxKu, I see.

If anybody is interested in a 1.8.2 release patched with this fix only, they could try docker pull wiremind/postgres-operator:v1.8.2-640581fb4661dbab852dfc0995b5e9f6140fd8f8