Open ajwerner opened 2 years ago
cc @JeffSwenson we re-discovered this issue during backlog grooming in SQL Schema. is this on your radar?
This is not an issue for Serverless. When a tenant db is configured as multi-region, it configures the tenant settings as global. When CC automation creates a multi-region storage cluster, it uses zone configs to enable global reads on the storage cluster's settings table.
Thanks for explaining! Then I'll leave this in our backlog for now until we see the priority come up in a different way.
Is your feature request related to a problem? Please describe.
Accessing cluster settings is important during cluster startup.
Describe the solution you'd like
The ideal tool to reduce the latency of reading cluster settings would be to set the
system.settings
table to useLOCALITY GLOBAL
.Describe alternatives you've considered Use stale reads and reason about any issues which might arise. The biggest problems seem related to cluster version potentially moving backwards. Perhaps we could work around this by waiting out some wall-clock time period when setting the version. This doesn't seem worth the complexity if we make it possible to set some system tables to
LOCALITY GLOBAL
.Additional context This will depend on https://github.com/cockroachdb/cockroach/issues/63365.
Jira issue: CRDB-18412