tailsamplingprocessor: Optimize tag mutator memory allocations #27889
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
Since each `tailSamplingSpanProcessor`'s instance is not concurrently called by the ticker worker (it's a 1-to-1 relationship) we can safely reuse a slice for the tag mutators used in `makeDecision`. Additionally the tag mutators themselves were causing a lot of allocations and since they are static, we created constants for them preventing allocations on each execution of `makeDecision`. This improved the `makeDecision` benchmark by ~31%. ``` benchstat old.txt new.txt name old time/op new time/op delta Sampling-10 51.8µs ± 1% 35.7µs ± 1% -30.94% (p=0.008 n=5+5) ```
jpkrohling
approved these changes
Oct 20, 2023
|
This looks cool. I jumped into this PR from the YouTube video as well. Nice job. |
|
A test is failing on a place that is related to the changes: Anyone willing to pick this up? I can probably take a look eventually, but not right now. |
…ield mutatorsBuf and panic in CICD. Added this field to all struct in ut.
I appreciate the optimizations made by the original author. The issue of the unit test only requires a very small change, but since I am unable to submit it on the original branch, I have opened a separate PR #28597 , which could be merged after #27889 . But yes @brancz feel free to add those fix lines to the original branch and I will delete the new pr ^_^ |
|
I think I was able to add your commits to this PR, @jiekun. Thank you! |
Codecov ReportAll modified and coverable lines are covered by tests ✅
... and 46 files with indirect coverage changes 📢 Thoughts on this report? Let us know!. |
jpkrohling
approved these changes
Oct 26, 2023
sigilioso
pushed a commit
to carlossscastro/opentelemetry-collector-contrib
that referenced
this pull request
Oct 27, 2023
…telemetry#27889) **Description:** Since each `tailSamplingSpanProcessor`'s instance is not concurrently called by the ticker worker (it's a 1-to-1 relationship) we can safely reuse a slice for the tag mutators used in `makeDecision`. Additionally the tag mutators themselves were causing a lot of allocations and since they are static, we created constants for them preventing allocations on each execution of `makeDecision`. This improved the `makeDecision` benchmark by ~31%. ``` benchstat old.txt new.txt name old time/op new time/op delta Sampling-10 51.8µs ± 1% 35.7µs ± 1% -30.94% (p=0.008 n=5+5) ``` **Testing:** Unit tests unchanged; added a benchmark **Documentation:** Perf improvement so no documentation changes needed. This was all based on production profiling data at Polar Signals running the collector. Here is a snapshot of the original profiling data we started with: https://pprof.me/52a7fab/ Judging by the production profiling data, a 31% improvement on the `makeDecision` codepath, should translate roughly into a 6% baseline CPU improvement our production deployment of the opentelemetry collector. The profiling data after improving: https://pprof.me/58c0e84/ This improvement was done as part of the Let's Profile Livestream where we optimize popular open-source projects live: https://www.youtube.com/watch?v=vkMQRjiNTHM --------- Co-authored-by: Jiekun <zhujiekun@52tt.com>
jmsnll
pushed a commit
to jmsnll/opentelemetry-collector-contrib
that referenced
this pull request
Nov 12, 2023
…telemetry#27889) **Description:** Since each `tailSamplingSpanProcessor`'s instance is not concurrently called by the ticker worker (it's a 1-to-1 relationship) we can safely reuse a slice for the tag mutators used in `makeDecision`. Additionally the tag mutators themselves were causing a lot of allocations and since they are static, we created constants for them preventing allocations on each execution of `makeDecision`. This improved the `makeDecision` benchmark by ~31%. ``` benchstat old.txt new.txt name old time/op new time/op delta Sampling-10 51.8µs ± 1% 35.7µs ± 1% -30.94% (p=0.008 n=5+5) ``` **Testing:** Unit tests unchanged; added a benchmark **Documentation:** Perf improvement so no documentation changes needed. This was all based on production profiling data at Polar Signals running the collector. Here is a snapshot of the original profiling data we started with: https://pprof.me/52a7fab/ Judging by the production profiling data, a 31% improvement on the `makeDecision` codepath, should translate roughly into a 6% baseline CPU improvement our production deployment of the opentelemetry collector. The profiling data after improving: https://pprof.me/58c0e84/ This improvement was done as part of the Let's Profile Livestream where we optimize popular open-source projects live: https://www.youtube.com/watch?v=vkMQRjiNTHM --------- Co-authored-by: Jiekun <zhujiekun@52tt.com>
RoryCrispin
pushed a commit
to ClickHouse/opentelemetry-collector-contrib
that referenced
this pull request
Nov 24, 2023
…telemetry#27889) **Description:** Since each `tailSamplingSpanProcessor`'s instance is not concurrently called by the ticker worker (it's a 1-to-1 relationship) we can safely reuse a slice for the tag mutators used in `makeDecision`. Additionally the tag mutators themselves were causing a lot of allocations and since they are static, we created constants for them preventing allocations on each execution of `makeDecision`. This improved the `makeDecision` benchmark by ~31%. ``` benchstat old.txt new.txt name old time/op new time/op delta Sampling-10 51.8µs ± 1% 35.7µs ± 1% -30.94% (p=0.008 n=5+5) ``` **Testing:** Unit tests unchanged; added a benchmark **Documentation:** Perf improvement so no documentation changes needed. This was all based on production profiling data at Polar Signals running the collector. Here is a snapshot of the original profiling data we started with: https://pprof.me/52a7fab/ Judging by the production profiling data, a 31% improvement on the `makeDecision` codepath, should translate roughly into a 6% baseline CPU improvement our production deployment of the opentelemetry collector. The profiling data after improving: https://pprof.me/58c0e84/ This improvement was done as part of the Let's Profile Livestream where we optimize popular open-source projects live: https://www.youtube.com/watch?v=vkMQRjiNTHM --------- Co-authored-by: Jiekun <zhujiekun@52tt.com>
6 tasks
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.
Description:
Since each
tailSamplingSpanProcessor's instance is not concurrentlycalled by the ticker worker (it's a 1-to-1 relationship) we can safely
reuse a slice for the tag mutators used in
makeDecision. Additionallythe tag mutators themselves were causing a lot of allocations and since
they are static, we created constants for them preventing allocations on
each execution of
makeDecision.This improved the
makeDecisionbenchmark by ~31%.Testing: Unit tests unchanged; added a benchmark
Documentation: Perf improvement so no documentation changes needed.
This was all based on production profiling data at Polar Signals running the collector. Here is a snapshot of the original profiling data we started with: https://pprof.me/52a7fab/
Judging by the production profiling data, a 31% improvement on the
makeDecisioncodepath, should translate roughly into a 6% baseline CPU improvement our production deployment of the opentelemetry collector.The profiling data after improving: https://pprof.me/58c0e84/
This improvement was done as part of the Let's Profile Livestream where we optimize popular open-source projects live: https://www.youtube.com/watch?v=vkMQRjiNTHM