|
|
||
|---|---|---|
| .. | ||
| benchmark-kafka-shard-1 | ||
| benchmark-kafka-shard-2 | ||
| benchmark-webhook-shard-1 | ||
| benchmark-webhook-shard-2 | ||
| README.md | ||
sizing-matrix aggregator fixture
Real, captured Playwright test-results.json output from the Test: E2E Infrastructure
benchmark run — not synthetic RunReportBuilder mocks. Used to exercise the
sizing-matrix aggregator end-to-end against the exact shape the benchmark lanes emit.
Provenance: captured from run 27965055540 (master, 2026-06-22). Trimmed to keep
each lane's suite structure (suite.file, spec.title) and the inline
run-report.json attachments; per-step stdout, traces, and metric:* attachments
were stripped to keep the fixture small (~250 KB vs ~9 MB raw).
Why captured, not generated: the aggregator's failure modes live in the gap between
the real artifact shape and the idealized RunReport type — reports arrive as inline
base64 attachments (not loose files), and scenario.spec is a human title (not a spec
path). A fabricated fixture re-encodes the same wrong assumptions the code makes, so the
test passes while CI fails. Only a real artifact catches it.
Consumed by: the aggregator end-to-end test (DEVP-531) — asserts the pipeline
produces a non-empty matrix (cells.length > 0) with concurrency per cell.
Across these four lanes it should resolve the six measured cells: L-S0/S1/S2, D-S1, X-S0/S1.
The lane files are named test-results.json on purpose — the aggregator matches on that
exact basename, so the fixture must use it to exercise the real extraction path. That
name is in the repo .gitignore, so these files are tracked via git add -f.
To refresh: download the benchmark-*-shard-* artifacts from a recent infra run and
re-trim (keep run-report.json attachments + suite file/title; drop the rest), then
git add -f the lane test-results.json files.