benchmark-results-storage

Benchmark results of the rolldown repo, stored as append-only JSON Lines — one line per measurement.

Files

File	Contents
benchmark-node-output.jsonl	Canonical, append-only. One line per (build × bench case × metric) from the Benchmarks Node workflow on rolldown main.
benchmark-legacy.jsonl	Archived series from before 2024-04-16 (earlier runner setup, different case set). Never appended to.
benchmark-node-output.json	Legacy nested format (the github-action-benchmark data file). Dual-written during the migration window, frozen afterwards.
scripts/migrate.mjs	The one-time converter that produced the .jsonl files. Self-checking (counts, endpoints, spot samples); safe to rerun.

Line schema

The same Entry schema rolldown/metric uses for its metric.json:

{"case":"threejs10x (default)","metric":"production build time","timestamp":1713278381040,"commit":"0c3f2baa25c141cb91392074e38171e74ef96b5a","unit":"ms","records":{"rolldown":297.49},"repoUrl":"https://github.com/rolldown/rolldown"}

• case — benchmark case name
• metric — production build time (unit ms) or peak memory (unit byte, Rust-side peak from rolldown's tracking allocator)
• timestamp — unix ms of the benchmark run
• commit — the rolldown commit the run measured
• records — value per bundler, keyed by name (currently only rolldown)

Lines are append-only and each workflow run appends all of its lines in one commit, so a commit is one benchmark run.

• Writer: .github/workflows/benchmark-node.yml in rolldown/rolldown
• Reader: https://rolldown.github.io/metric/ (rolldown/metric)