Resource-Pool Bottleneck fix #29
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* We were running into throughput bottlenecks because it was found that the resource-pool was blocking requests. When The maximum resources per-stripe would get utilized and it would block. * This PR increases the number of stripes from the default 1 to now 4. * The number 4 was chosen here a it was found to be the number with maximum gains without losing stability in local load-testing. A full report on different numbers was presented and can be referred to for a fuller picture.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The PR introduces the fix for hedis fork master-branch performance issue that was found during profiling.
Isolated Issue.
We were running into throughput bottlenecks because during profiling it was found that the resource-pool was blocking requests. When The maximum resources per-stripe would get utilized and it would block.
Proposed fix
The proposed solution here is to introduce more number of stripes when making a pool connection. So that when one pool is saturated, it won't block, increasing the throughput at the cost of a bit more memory.
This PR increases the number of stripes from the default 1 to now 4.
The number 4 was chosen here a it was found to be the number with maximum gains without losing stability in local load-testing. A full report on different numbers used and how they translated to results was presented and can be referred to for a fuller picture.
Results observed after this Fix
The results were that changing the number of stripes in the connection pool settings
on the hedis master branch fork does seem to affect the performance to a good
degree.
RPS observed before -> 188
RPS observed after -> 294
The performance of the API tested here went from
188 RPS->294 RPSwith the inclusion of this fix. More than 100 RPS gain.The time-allocation percentage of the major
takeResourceoffender went down from19.6%-> down to2.8%as found when profiling cost-centres with this fix applied.A full report on the issue, RCA and fix and different combinations was shared internally.
Caveats
There are some minor possible caveats here to be aware of :-
This isn't extensively tested across all repos or scenarios it might possibly be used with, thought it should just work.
If 4 stripes are found unstable for some repos or some deploys with low or tight memory limits, we can lower this to 3 for further increased stability, although with reduced performance, but still a good amount higher than the current status quo.