fix(ci): allow-list Publish trigger events in benchmark.yml instead of excluding 'push'

[carlos-alm]

Summary

• Both jobs in benchmark.yml gated on github.event.workflow_run.event != 'push'. That's a negative exclusion, not a positive allow-list of the event types that actually warrant a benchmark run.
• fix(ci): move dev builds from every push to a daily schedule #2113 changes Publish's dev-build trigger from push to schedule — != 'push' doesn't exclude 'schedule', so once that merges, scheduled dev-build Publish completions would newly pass this filter, and:
  • record-benchmarks would try to download benchmark-files/benchmark-results-json artifacts that don't exist for schedule-triggered Publish runs (only release/workflow_dispatch go through pre-publish-benchmark), failing the job
  • embedding-benchmark would kick off a redundant 2.5+ hour run against the latest stable npm version once a day, for nothing
• Fix: replace the exclusion with an explicit allow-list (event == 'release' || event == 'workflow_dispatch') so it can't silently drift out of sync with Publish's trigger set again

Why

Direct correctness dependency of #2113 (Publish dev-build scheduling). Also corrects my own earlier issue #2109, which incorrectly assumed this workflow ran the full benchmark suite on every Publish completion — it already didn't, this fix just keeps that true after #2113.

Closes #2109
Depends on #2113 (should merge together or #2113 first)

Test plan

• YAML parses, actionlint shows no new findings (pre-existing shellcheck notes only)
• After both PRs merge: verify the first scheduled dev-build Publish completion does not trigger record-benchmarks or embedding-benchmark
• Verify a real release still triggers both as before

[greptile-apps]

Greptile Summary

This PR hardens the if: conditions on both jobs in benchmark.yml by replacing a negative event exclusion (!= 'push') with an explicit allow-list (== 'release' || == 'workflow_dispatch'). This is a correctness fix needed before PR #2113 lands, which changes Publish's dev-build trigger from push to schedule.

• record-benchmarks: now only runs when Publish was triggered by release or workflow_dispatch — the only paths that produce the benchmark-files/benchmark-results-json artifacts this job consumes.
• embedding-benchmark: same allow-list applied; prevents a spurious 2.5+ hour benchmark run firing daily once schedule becomes a valid Publish trigger.
• Direct workflow_dispatch of benchmark.yml itself is preserved for the embedding-benchmark job.

Confidence Score: 5/5

Safe to merge; the two-line condition change is logically sound and directly prevents unwanted benchmark runs once #2113 lands.

The change converts a fragile negative exclusion into a tight positive allow-list. Both jobs are updated consistently, the asymmetry between them is intentional and correctly preserved. No artifacts, secrets, or permissions are touched.

No files require special attention; the single changed file has a minimal, well-commented diff.

Important Files Changed

Filename	Overview
.github/workflows/benchmark.yml	Both record-benchmarks and embedding-benchmark job conditions updated from != 'push' exclusion to `== 'release'

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    PW["Publish workflow completes"]
    BW["benchmark.yml triggered via workflow_run"]

    PW --> |"event = 'release'"| BW
    PW --> |"event = 'workflow_dispatch'"| BW
    PW --> |"event = 'push' (legacy)"| X1["FILTERED — was allowed before fix"]
    PW --> |"event = 'schedule' (PR #2113)"| X2["FILTERED — would have been allowed before fix"]

    BW --> FILTER{"if condition (after fix)"}
    FILTER --> |"event == 'release' OR 'workflow_dispatch'"| JOBS["Run record-benchmarks + embedding-benchmark"]
    FILTER --> |"any other event"| SKIP["Skip — no run"]

    DW["Direct workflow_dispatch on benchmark.yml"] --> |"Only embedding-benchmark"| EMBED["embedding-benchmark runs (record-benchmarks skipped)"]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    PW["Publish workflow completes"]
    BW["benchmark.yml triggered via workflow_run"]

    PW --> |"event = 'release'"| BW
    PW --> |"event = 'workflow_dispatch'"| BW
    PW --> |"event = 'push' (legacy)"| X1["FILTERED — was allowed before fix"]
    PW --> |"event = 'schedule' (PR #2113)"| X2["FILTERED — would have been allowed before fix"]

    BW --> FILTER{"if condition (after fix)"}
    FILTER --> |"event == 'release' OR 'workflow_dispatch'"| JOBS["Run record-benchmarks + embedding-benchmark"]
    FILTER --> |"any other event"| SKIP["Skip — no run"]

    DW["Direct workflow_dispatch on benchmark.yml"] --> |"Only embedding-benchmark"| EMBED["embedding-benchmark runs (record-benchmarks skipped)"]

Loading

_{Reviews (3): Last reviewed commit: "Merge remote-tracking branch 'origin/mai..." | Re-trigger Greptile}

[carlos-alm]

Addressed Claude's review feedback: none required — Greptile's review (5/5 confidence, "Safe to merge") did not flag any findings or gaps against the workflow diff itself.

What was done since that review:

• CI had 2 failing checks (CI Testing Pipeline, Pre-publish benchmark gate), both caused by tests/benchmarks/regression-guard.test.ts flagging Full build/1-file rebuild timing regressions against the stale 3.15.0 baseline — this branch predated the fix for that (PR fix(bench): exempt 3.15.0 baseline drift for Full build / 1-file rebuild #2107, already on main).
• Merged origin/main into this branch (no conflicts) to pick up fix(bench): exempt 3.15.0 baseline drift for Full build / 1-file rebuild #2107's baseline-drift exemption. Verified the merge did not alter this PR's actual change (the benchmark.yml allow-list diff is identical before/after the merge).
• Ran npm run lint and the regression-guard test suite locally to confirm green before pushing.

[carlos-alm]

@greptileai