Comment by imiric
6 months ago
I was never a fan of images from Bitnami. They always used complicated entrypoint and setup scripts, and introduced weird quirks to the software. More than once have I experienced issues or ran into configuration limitations with Bitnami images that didn't exist in official ones.
So good riddance, as far as I'm concerned. I recommend anyone to avoid using them, and switch to official images or to build them yourself if they're not provided. That's the more secure approach, anyway.
I concur. There was supposedly a migration path from their postgresql image & chart to the postgresql-ha image & chart.
Aside of having to re-mount the data disk and move things around manually; the -ha chart has numerous other issues where it always requires the master to be node-0. And with pods being rescheduled within a statefulset, good look having the master be on node-0. If there was an outage and the master is anywhere else, node-0 will just 'wait' for a master to come online, time out and shoot itself in the head thinking it is in a network partition and that retrying may help.
The algorithm implemented by postgresql-ha turned out to be plain broken. Only able to survive pods neatly shutting down.
Sometimes, over engineered approaches are necessary to make older software work with environment variables and configmaps, because said software is still designed for traditional VM deployments.
Agreed, Bitnami images often feel over-engineered.