[PROF-10589] Add internal metrics for GVL profiling performance #3993
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
…or GVL" This is useful because our main control for "Waiting for GVL" is the waiting_for_gvl_threshold_ns, and thus we can use this stat to monitor how many events are being skipped as being under that threshold. Note that regular (periodic) samples can still observe the "Waiting for GVL" state; but there will not be any special samples created to represent the beginning or end of "Waiting for GVL".
…stead of calling ThreadContext directly This structure is similar to the one we're already using for `sample_from_postponed_job` => `rescued_sample_from_postponed_job` and enables us to have more control over the parameters sent to the ThreadContext collector.
…gvl_running` function Now that the `CpuAndWallTimeWorker` does not call `thread_context_collector_sample_after_gvl_running` directly from a `rb_rescue`, it can use the version of the method with extra arguments directly.
…uring directly This is similar to what we do for `thread_context_collector_sample` and makes it easier for the CpuAndWallTimeWorker to keep stats on how long sampling took without needing to read the clock more than needed.
AlexJF
approved these changes
Oct 11, 2024
BenchmarksBenchmark execution time: 2024-10-11 12:03:44 Comparing candidate commit b2b331b in PR branch Found 0 performance improvements and 0 performance regressions! Performance is the same for 23 metrics, 2 unstable metrics. |
Flaked in CI with...
```
1) Datadog::Profiling::Collectors::CpuAndWallTimeWorker#start when main thread is sleeping but a background thread is working when GVL profiling is enabled when 'Waiting for GVL' periods are below waiting_for_gvl_threshold_ns does not trigger extra samples
Failure/Error:
expect(cpu_and_wall_time_worker.stats.fetch(:gvl_dont_sample))
.to be > 100 # Arbitrary, on my machine I see 250k on a run
expected: > 100
got: 5
```
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #3993 +/- ##
==========================================
- Coverage 97.87% 97.86% -0.02%
==========================================
Files 1314 1314
Lines 78652 78667 +15
Branches 3909 3909
==========================================
+ Hits 76978 76984 +6
- Misses 1674 1683 +9 ☔ View full report in Codecov by Sentry. |
ivoanjo
added
profiling
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 introduces the following new internal metrics/counters for the profiler:
Motivation:
These metrics get automatically reported together with profilers, and allow us to keep an eye on/investigate/debug the performance of GVL profiling.
Additional Notes:
I needed to refactor a bit the way we trigger GVL profiling samples; the change is similar to what we were doing with regular cpu/wall-time samples.
How to test the change?
This change includes test coverage.