Run the SUSP-5 checkpoint gate as advisory on hosted CI runners#20
Merged
Conversation
The measured checkpoint overhead sits at the 5% budget boundary: identical test-only changes scored a 4.63% geomean in run 29678923617 and 5.11% in run 29680420318, so the strict SUSP-5 verdict on shared GitHub runners is decided by machine noise, not by the code. Follow the existing NFR-1-startup/SNAP-6 pattern: the gate still measures and reports every pair plus limitMet, but enforcement becomes opt-out via --advisory-checkpoint-overhead, wired through the kwasm.benchmark.enforceCheckpointOverhead Gradle property, and the Performance workflow passes false. Local and dedicated-hardware runs keep the strict default. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
After #14 removed the noisy cross-runner history comparison,
JVM performance gatesstill fails intermittently — now on SUSP-5 (checkpoint overhead ≤ 5% geomean). The product sits right at that boundary, so on shared runners the verdict is a coin flip decided by machine noise:The worst pair in the failing run is the ~1 µs/op
hostToGuestmicrobenchmark at 7.99% — exactly the scale where hosted-runner jitter dominates.Fix
Same pattern the gate tool already uses for NFR-1-startup and SNAP-6: keep measuring and reporting (all pairs,
geomeanRatio, newlimitMetfield), but make enforcement switchable.performance_gate.py verify --advisory-checkpoint-overhead→ SUSP-5 reportsenforced: falseand no longer fails the run; strict remains the default.kwasm.benchmark.enforceCheckpointOverhead(defaulttrue) wires it through:benchmarks:<target>PerformanceGate.falsefor the jvm and native jobs.performanceGateToolTeststill runs in CI.Local runs and any future dedicated hardware keep strict enforcement by default.
🤖 Generated with Claude Code