[PROF-10201] Reduce allocation profiling overhead by replacing tracepoint with lower-level API #3805
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
…oint with lower-level API **What does this PR do?** This PR reduces the allocation profiling overhead by replacing the Ruby tracepoint API with the lower-level `rb_add_event_hook2` API. The key insight here is that while benchmarking allocation profiling and looking at what the VM was doing, I discovered that tracepoints are just a thin user-friendlier wrapper around the lower-level API. The lower level API is publicly-available (in "debug.h") but it's listed as "undocumented advanced tracing APIs". **Motivation:** As we're trying to squeeze every bit of performance from the allocation profiling hot-path, it makes sense to make use of the lower-level API. **Additional Notes:** I'm considering experimenting with moving the tracepoint we use for GC profiling to this lower-level API as well, since that's another performance-sensitive code path. **How to test the change?** Functionality-wise, nothing changes, so existing test coverage is enough (and shows this alternative is working correctly). Here's some benchmarking numbers from `benchmarks/profiler_allocation.rb`: ``` ruby 2.7.7p221 (2022-11-24 revision 168ec2b1e5) [x86_64-linux] Warming up -------------------------------------- Allocations (baseline) 1.565M i/100ms Calculating ------------------------------------- Allocations (baseline) 15.263M (± 1.4%) i/s - 153.400M in 10.052624s Warming up -------------------------------------- Allocations (event_hook) 1.240M i/100ms Calculating ------------------------------------- Allocations (event_hook) 12.571M (± 2.1%) i/s - 126.456M in 10.064297s Warming up -------------------------------------- Allocations (tracepoint) 1.183M i/100ms Calculating ------------------------------------- Allocations (tracepoint) 12.225M (± 0.5%) i/s - 123.072M in 10.067487s Comparison: Allocations (baseline): 15262756.4 i/s Allocations (event_hook): 12570772.3 i/s - 1.21x slower Allocations (tracepoint): 12225052.0 i/s - 1.25x slower ``` Here, `event_hook` is with the optimization, whereas `tracepoint` is without it. I am aware these numbers are close to the margin of error. I re-ran my benchmarks a number of times and consistently observed the event_hook version coming out ahead of the tracecpoint version, even if by little.
anmarchenko
approved these changes
Jul 24, 2024
BenchmarksBenchmark execution time: 2024-07-24 10:15:09 Comparing candidate commit 9431929 in PR branch Found 0 performance improvements and 0 performance regressions! Performance is the same for 10 metrics, 2 unstable metrics. |
|
I saw a new flaky profiler spec: https://app.circleci.com/pipelines/github/DataDog/dd-trace-rb/15605/workflows/5b32e4bb-155e-4a94-bf8b-980599eaafae/jobs/568892 . I've looked into it and it's related to #3792 and not this PR so I'll open a separate PR to tackle it. |
Merged
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.
What does this PR do?
This PR reduces the allocation profiling overhead by replacing the Ruby tracepoint API with the lower-level
rb_add_event_hook2API.The key insight here is that while benchmarking allocation profiling and looking at what the VM was doing, I discovered that tracepoints are just a thin user-friendlier wrapper around the lower-level API.
The lower level API is publicly-available (in "debug.h") but it's listed as "undocumented advanced tracing APIs".
Motivation:
As we're trying to squeeze every bit of performance from the allocation profiling hot-path, it makes sense to make use of the lower-level API.
Additional Notes:
I'm considering experimenting with moving the tracepoint we use for GC profiling to this lower-level API as well, since that's another performance-sensitive code path.
How to test the change?
Functionality-wise, nothing changes, so existing test coverage is enough (and shows this alternative is working correctly).
Here's some benchmarking numbers from
benchmarks/profiler_allocation.rb:Here,
event_hookis with the optimization, whereastracepointis without it.I am aware these numbers are close to the margin of error. I re-ran my benchmarks a number of times and consistently observed the event_hook version coming out ahead of the tracecpoint version, even if by little.