-
Notifications
You must be signed in to change notification settings - Fork 3.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
kvserver: increase rebalance action impact req #98889
Conversation
It looks like your PR touches production code but doesn't add or edit any test code. Did you consider adding tests to your PR? 🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf. |
eda0b63
to
c9808eb
Compare
c9808eb
to
8a82539
Compare
Previously, the store rebalancer would attempt to transfer leases or relocate replicas of a range so long as the load of the range (estimated using the local leaseholder replica) was greater than 0.1% of the store's load. This commit updates the value to be higher for both lease rebalancing and range rebalancing, specifically: lease rebalancing: 0.1% -> 0.5% range rebalancing: 0.1% -> 2.0% The documented rationale, as for why these values exist is also the reason for the increase: decrease unnecessary churn. Only the hottest ranges are considered, when the impact of a rebalance action for these ranges is estimated to be less than a small fraction of the store's total load, it is indicative that the distribution of replica load is not skewed. Letting the replicate queue balance replica and lease counts is a better option in such distributions. Making the minimum for range rebalancing higher is also added, as multiple replicas for a range is inherently more costly than moving a lease. Resolves: cockroachdb#98890 Release note: None
8a82539
to
a944e57
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm a little concerned with the potential impact of this change in the sense that it makes some rebalancing less likely to occur, but as you mentioned, the replicate_queue will also do rebalancing. I'm not sure we should backport this as I'm not aware of any field cases that are impacted negativity by this.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @kvoli)
Won't backport. I agree, it's generally a risky change and we don't have a pressing need. |
In light of #104292 I'm going to merge this now and see how we do. |
bors r=andrewbaptist |
Build failed: |
bors r=andrewbaptist |
Build succeeded: |
Previously, the store rebalancer would attempt to transfer leases or relocate replicas of a range so long as the load of the range (estimated using the local leaseholder replica) was greater than 0.1% of the store's load.
This commit updates the value to be higher for both lease rebalancing and range rebalancing, specifically:
lease rebalancing: 0.1% -> 0.5%
range rebalancing: 0.1% -> 2.0%
The documented rationale, as for why these values exist is also the reason for the increase: decrease unnecessary churn. Only the hottest ranges are considered, when the impact of a rebalance action for these ranges is estimated to be less than a small fraction of the store's total load, it is indicative that the distribution of replica load is not skewed. Letting the replicate queue balance replica and lease counts is a better option in such distributions (until we consolidate rebalancing responsibility).
Making the minimum for range rebalancing higher is also added, as multiple replicas for a range is inherently more costly than moving a lease.
Resolves: #98890
Release note: None