fix(profiling): release the EventPipe session when the SDK shuts down - #5470
fix(profiling): release the EventPipe session when the SDK shuts down#5470jamescrosswell wants to merge 11 commits into
Conversation
SamplingTransactionProfilerFactory.Dispose() disposed the antecedent Task<SampleProfilerSession> rather than the session it wraps. Fixing that alone changes nothing observable, because two further defects sit between "SDK shuts down" and "EventPipe session released": - SampleProfilerSession.Stop() built _processing as an OnlyOnFaulted continuation, which transitions to Canceled when Process() returns normally - so Wait() threw on every clean shutdown and the disposals after it were never reached. - ProfilingIntegration was not IDisposable, so Hub never registered it for cleanup and the factory's Dispose() was only ever called by tests. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## main #5470 +/- ##
==========================================
- Coverage 74.87% 74.78% -0.09%
==========================================
Files 513 513
Lines 18659 18821 +162
Branches 3636 3678 +42
==========================================
+ Hits 13970 14076 +106
- Misses 3819 3867 +48
- Partials 870 878 +8 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>
Both review bots flagged the same window from different angles, and both were right: - The startup task read _shutdownCts.Token, which throws ObjectDisposedException once Dispose() has disposed the CTS. The token is now captured up front so the task never touches the source. - StartEventPipeSession() can block for up to 30s and isn't cancellable, so Dispose() could run before _session was ever assigned - disposing nothing and leaving the session that arrived later running forever. A lock now hands the session between the two sides so exactly one of them stops it. Adds test seams to SampleProfilerSession so the race can be exercised deterministically rather than by timing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…e' into fix/5418-profiler-session-dispose # Conflicts: # src/Sentry.Profiling/SamplingTransactionProfilerFactory.cs
Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>
Documents the convention in AGENTS.md: this repo favours code that needs no comments, with the code and the PR description as the documentation. Minimal comments only where something genuinely isn't obvious. Applies it to this branch - 34 added comment lines down to 12. What survives is non-intuitive framework behaviour (a ContinueWith whose criteria aren't met is Canceled; reading CancellationTokenSource.Token after Dispose throws) and the justification for an empty catch. Test rationale moved into FluentAssertions "because" strings, where it shows up in failure output instead of sitting in a comment. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…e' into fix/5418-profiler-session-dispose
disposedDuringStartup read as "SDK startup". It means Dispose() ran while the EventPipe session was still being created, so disposedWhileSessionStarting. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ProfilingIntegration.Dispose() disposed the factory but left options.TransactionProfilerFactory pointing at it. Reusing the same SentryOptions for a later Hub meant Register() saw a non-null factory, skipped creating a working one, and profiling silently never started. Reported by Warden on #5470. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
An earlier commit on this branch staged a stale submodule working copy, reverting the pointer that came in from main and leaving the committed Cocoa bindings out of step with it. That fails dirty-check.ps1 on the macOS build. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit d26d7a5. Configure here.
| factory.Dispose(); | ||
| _ownedFactory = null; | ||
| _options = null; | ||
| } |
There was a problem hiding this comment.
Re-init disposes the replacement hub profiler
Medium Severity
SentrySdk.Init constructs the new Hub (and calls Register) before disposing the previous one. Register keeps the existing factory, then the old hub's Dispose stops that factory and clears TransactionProfilerFactory. The replacement hub is left with no profiler, so a second Init silently disables profiling.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit d26d7a5. Configure here.
vaind
left a comment
There was a problem hiding this comment.
I'd look at bugbot's comment, other than that, lgtm


SamplingTransactionProfilerFactory.Dispose()disposed the antecedentTask<SampleProfilerSession>rather than the session it wraps, so theEventPipeSessionwas never released.That line turned out to be the last of three independent defects sitting between "SDK shuts down" and "EventPipe session released" — fixing it alone would not have changed anything observable.
What was broken
SampleProfilerSession.Stop()skipped its own disposals on the happy path._processingwas built withTaskContinuationOptions.OnlyOnFaulted. WheneventSource.Process()returns normally the continuation criteria aren't met, so the continuation transitions toCanceledand_processing.Wait()throwsAggregateException(TaskCanceledException).EventPipeSession.Dispose()andTraceLogEventSource.Dispose()sat after that call and were never reached — thecatchloggedError during sampler profiler session shutdown.and swallowed it. This fired on every clean shutdown, not on some edge case.Nothing ever called
Dispose()on the factory.ProfilingIntegrationwasn'tIDisposable, soHubnever added it to_integrationsToCleanup, and the factory's only other reference isoptions.TransactionProfilerFactory, which nothing disposes. Outside of tests,SamplingTransactionProfilerFactory.Dispose()was unreachable.The reported bug. The
ContinueWithparameter is the antecedent task, soDispose()calledTask.Dispose()— a near no-op on a completed task — instead of disposing the session.What this changes
ProfilingIntegrationnow owns the factory it creates. It implementsIDisposableso the Hub tears it down on shutdown, and disposal also clearsoptions.TransactionProfilerFactory, so aSentryOptionsreused for a later hub gets a working profiler rather than a stale reference. It only touches a factory it created itself — one supplied from elsewhere is left alone.SamplingTransactionProfilerFactory.Dispose()now actually stops the session.StartEventPipeSession()can block for up to 30 seconds and can't be cancelled, so disposal can land before there is any session to stop; a lock hands the session between the startup task andDispose()so exactly one of the two tears it down. The wait for the first event is cancelled on disposal rather than left to block indefinitely. Both waits are bounded at 2 seconds so a wedged session can't hang shutdown.SampleProfilerSession.Stop()moves the disposals into afinally, so the EventPipe connection is released even when stopping the session or draining its events fails.Impact
Low, and narrower than the issue suggests. The session is created once and lives for the SDK's lifetime —
SamplingTransactionProfiler.Stop()only clears_inProgress, it never touches the session — so the only disposal point is SDK shutdown, which is normally process shutdown, where the OS reclaims the handle and theTraceLogregardless. Nothing accumulates while an application runs.It matters where the process outlives the SDK: repeated
Init/Closecycles strand oneEventPipeSessionplus itsTraceLogper cycle, and hosts that close Sentry but keep running never release the pipe.The issue links this to the Windows Service memory growth in #3375. That link doesn't hold up — a service that initialises Sentry once would never reach any of these paths. The steady-state growth reported in #5469 is a separate mechanism and is not addressed here.
Notes
Error during sampler profiler session shutdown.on every clean shutdown with profiling enabled. Worth asking anyone reporting profiler-related resource growth whether they see it.Hub.Dispose()now does real work for profiling users where it previously did none.Sentry.Profiling.TestsisSkip.If(TestEnvironment.IsGitHubActions)because it opens real EventPipe sessions. TheProfilingIntegrationtests added here are plain[Fact]s that do run in CI; the two session-level tests follow the existing skip pattern.AGENTS.mdgains a short section recording the repo's preference for code over comments. Happy to split that into its own PR if it doesn't belong here.Fixes #5418