[crdb] Remove useless indexes#1517
Conversation
|
@the-glu, thank you for this. Can you please provide evidence of no degradation ? |
Performances tests I ran (locus ones) show no changes, but they are really synthetic. This also really depends on the planner / statistics / data in the table, so it's mostly based on query analysis that measurable data. |
|
Please update the NEXT_RELEASE_NOTES |
|
This looks reasonable to me, but as @barroco notes, users with existing deployments will need/want to do something because of this change so we need something in NEXT_RELEASE_NOTES. Beyond this PR, we should make sure it's clear how to perform a DB migration and what is and isn't required. Specifically, one user group thought they needed to synchronize the actions of all pool participants to a narrow time window to update both the schema and software version, while I think we have a design intent to avoid that need (instead allowing a leisurely schema update on the previous software version, then leisurely transitioning of each DSS instance to the new software version, or perhaps in the reverse order). |
|
Release notes updated |
This PR contrib to general performance and to #813 by removing useless indexes in CRDB.
Cells index or primary keys should be used looking at queries, making those indexes useless.
ends_at / updated_at are not removed to use indexes when using evict tool.
owner on
rid.subscriptionsis used inMaxSubscriptionCountInCellsByOwnerNotice that not the case for yugabyte and nothing changed there.