quotapool,kvserver: extend quotapool.RateLimit, rate limit queue addition in SystemConfigUpdate#53605
Conversation
|
This change is |
6cfdfc4 to
1898ec9
Compare
Release justification: non-production code changes Release note: None
Release justification: low risk, high benefit changes to existing functionality Release note: None
Release justification: non-production code changes Release note: None
…replicas This has a shockingly profound impact on the runtime of ORM tests. This PR is a fancier implementation of cockroachdb#53603 that permits bursting and makes setting a relatively low rate feel less risky in the face of a small burst of changes. Release justification: low risk, high benefit changes to existing functionality Release note (performance improvement): Limited the frequency of an expensive operation due to schema changes making workload which perform schema changes at a high rate less resource intensive.
1898ec9 to
ab3f53d
Compare
knz
left a comment
There was a problem hiding this comment.
I just reviewed this and I feel rather comfortable by the functional correctness of the thing (i.e. I can't see it blocking unexpectedly)
Can you speak a bit more quantitatively about what motivated this change and the visible benefits?
Reviewed 2 of 2 files at r1, 2 of 2 files at r2.
Reviewable status:complete! 0 of 0 LGTMs obtained (waiting on @knz and @tbg)
knz
left a comment
There was a problem hiding this comment.
Reviewed 1 of 1 files at r4.
Reviewable status:
|
Prior to this change, over 70% of the CPU utilization of the typeorm roachtest was underneath those calls. The change decreases the runtime of the
Before: After |
|
According to a 30s profile from before and after, this change reduces the CPU utilization due to the calls to adding to the queue from 69.7% to 4.9% and reduces the CPU utilization overall dramatically. I'll post runtime differences when they finish. See the below charts: Before: After: |
|
this looks gorgeous! I'm LGTM on this |
|
TFTR! bors r=knz
Before: After: |
|
Build succeeded: |
See individual commits.
This PR is a fancier implementation of #53603 that permits bursting and makes
setting a relatively low rate feel less risky in the face of a small burst of
changes.
Release justification: low risk, high benefit changes to existing functionality
Release note (performance improvement): Limited the frequency of an expensive
operation due to schema changes making workload which perform schema changes
at a high rate less resource intensive.