Repository navigation
feat: performance regression gate on release tags (benchmark CI proposal, draft) - #1182
Draft
GOODBOY008 wants to merge 2 commits into
Draft
GOODBOY008 wants to merge 2 commits into
GOODBOY008 wants to merge 2 commits into
Conversation
GOODBOY008
force-pushed
the
feat/benchmark-regression-gate
branch
7 times, most recently
from
October 10, 2026 10:08
0636776 to
575fc9c
Compare
Adds a fesod-benchmark module with JMH benchmarks for Fesod, kept to pure JMH classes per review feedback: no CI integration, no report or analysis code. Closes apache#572. Suites (org.apache.fesod.sheet.benchmark): - FesodBenchmark: read/write x XLSX/CSV x 1K/10K/100K rows (average time, 3 forks, fixed 1g G1 heap, stable execution contract) - FesodVsPoiBenchmark: Fesod vs Apache POI on identical data (write, DOM read, batched streaming read; XLSX + XLS) - deterministic data generation (fixed seed + fixed date anchor) and unit tests for the generator and file utilities Benchmarks run manually via the shaded benchmarks.jar; see fesod-benchmark/benchmark.md.
Wire FesodBenchmark into a performance regression gate (split out of - BaselineRunner pins the execution contract (3 forks, 3x1s warmup, 5x2s measurement, fixed JVM args, gc profiler) and writes JMH JSON - BaselineComparator compares against the committed baseline with tiered, noise-aware verdicts (alloc/op hard-fail; time regressions gated on JMH error-bar overlap) and renders the Markdown report - benchmark.yml runs the gate on release tags and manual dispatch only, bootstraps the baseline on first run via an automated PR, and posts regression reports on the matching GitHub Release (best-effort notification ladder) No baseline is committed up front - the first run on an apache runner bootstraps it through an automated PR, so the committed baseline has apache-runner provenance.
GOODBOY008
force-pushed
the
feat/benchmark-regression-gate
branch
from
October 12, 2026 01:22
575fc9c to
5917023
Compare
This branch has not been deployed
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.
What this proposes
Wire the
FesodBenchmarksuite (from #575) into a performance regression gate that runs on release tags and manual dispatch only — never per pull request.Why a gate at all
Design highlights
workflow_dispatch(never per-PR — shared-runner time noise is ±15–20%)gc.alloc.rate.norm(allocation per op) — measured noise ±0.1%, effectively deterministic → regression beyond threshold failsubuntu-24.04+ JDK 17 (Temurin) runners; first run bootstraps it via an automated PR; passing tags advance it via automated PRsactions/*only; bot pushes only tobenchmark/baseline-*branches; baseline JSON enters the repo only through reviewed PRsThresholds default to warn 10% / fail 20%, overridable per dispatch or via repo variables (
BENCHMARK_WARN_PCT/BENCHMARK_FAIL_PCT).What's in the diff
baseline/BaselineRunner— pinned-contract JMH runner (3 forks, 3×1s warmup, 5×2s measurement,-Xms1g -Xmx1g -XX:+UseG1GC, gc profiler always on)baseline/BaselineComparator— tiered comparator, Markdown report, GitHub Actions annotations.github/workflows/benchmark.yml— run / compare / notify-regression / update-baseline jobsbaseline/README.md+ a gate section inbenchmark.mdNo baseline is committed up front: the first run on an apache runner bootstraps it through an automated PR, so the committed baseline has apache-runner provenance.
Context
@delei's review on #575 asked to keep the benchmark module to the JMH classes and not run benchmarks through CI currently — #575 now does exactly that. This draft carries the gate half so the idea is reviewable separately: happy to close it if the community isn't ready for it yet, or to adjust the design per feedback (e.g. different triggers, thresholds, or notification behavior).