Open jseldess opened 3 years ago
cc @taroface, @mikeCRL, @piyush-singh.
I'd like to add that a single region Kubernetes cluster is likely to be deployed on multiple AZs, setting up localities w/ Regions and AZ in that scenario can distribute range across AZs, give stronger resiliency. Otherwise some ranges could place all replicas in a single AZ and can experience unavailable ranges when multiple nodes fail in an AZ, not to mention AZ failure.
Relates to #10479.
@taroface, did you mean to link to a different issue or PR? Not sure how that one is related...
I had the wrong PR number originally. Or are you referring to #10479?
Ah ok. No, I was just confused by that previously linked PR. All good.
Jesse Seldess commented:
In our manual deployment tutorials, we have a paragraph about the need to add locality for multi-region (end of step 3). However, even for single-region clusters, we should be encouraging customers to set locality as a best practice and in preparation for future scaling out to multiple regions. It would also help us to avoid cases where customers have to do a rolling restart to add localities (see https://github.com/cockroachdb/docs/issues/9970).
Likewise, we should guide users to customize their Kubernetes configs to use locality, even for single-region, in preparation for multi-region scaling.
Jira Issue: DOC-1095