Understanding the high-availability helm chart

[mwashley2]

Looking at and generating a helm template with the HA chart I notice there is a stateful set created for each individual node (ie, a statefulset for each coord-0, one for coord-1, etc) vs for example a single statefulset for coordinators with 3 nodes and statefulset for the 2 data nodes. I'm admittedly not an expert in helm/k8s but other technologies I've used such as elastic work in this way.

I ask because I'm currently on a standalone config for testing purposes and preparing to deploy a HA deployment. I've been getting mixed results from AmazonQ specifically and decided I should consult with the authors here before going too much further.

[as51340]

Hi @mwashley2! Great question. The only reason why we chose a single stateful set for each instance was because it was simpler to work with at the beginning. We have primary-secondary type of replication so I had some doubts how to set-up a single StatefulSet.

Do you have some problems in setting-up cluster because of having multiple StatefulSets?

[mwashley2]

Thanks @as51340 All is well now, I really just like to understand everything. I noted a reference in the documentation to each node having a statefulset and rolled with it.

I managed to get the cluster up and running after struggling for a bit seemingly because 0 isn't a valid coordinator id, but once I re-numbered them the basic HA cluster is up and running. I'm currently struggling now with the cluster-setup.yaml which likes to fail adding the second coordinator due to a 2000ms timeout. The coordinator usually does add but is missing bolt_port and then subsequent nodes don't get added, and subsequent job runs detect there are already >1 nodes and exits. I'll open a separate issue for this if I can't get to the bottom of it.

[as51340]

sure, let us know if you are unable to fix it