Repository navigation
perf(test): storage suite takes ~90–105s, dominated by a few heavy IO tests #1940
Description
What happened
packages/storage runs 1026 tests in 88.3s locally (104.7s in an earlier run). Wall time is dominated by a handful of tests doing real fs/SQLite/git work. Slowest, from --test-reporter=spec timings on latest main (76e1476):
| Test | Time |
|---|---|
| Git workspace service (describe) | 34.5s |
| rejects missing/changed/extra/symlinked backup payloads (single test) | 32.3s |
| restores a writable Automation authority after a real legacy cutover | 27.1s |
| SQLite Artifact store (describe) | 20.4s |
| cleans interrupted backup and restore staging | 16.3s |
| execution stores / Interaction store / Git worktree executor / operational cutover | 14.5s / 12.9s / 11.5s / 10.9s |
With --concurrency=3, storage is the critical path of the parallel workspace batch.
How to reproduce
cd packages/storage && time node --test "dist/**/*.test.js" (~88–105s).
Environment
- Maka commit:
76e14764a - OS: macOS + Node v26.5.1; likely slower on CI’s 2-core ubuntu runners
- Surface: tests only
Additional context
Heavy tests come from the storage SQLite migration + WAL backup series (#1570/#1694/#1706) and the Git workspace service (#1742). Options: gate the fs/git/sqlite-heavy cases behind the existing MAKA_STORAGE_STRESS env hook (already wired in CI), reduce payload/iteration sizes, or split a fast unit tier from the integration tier.
Activity
UncertaintyDeterminesYou4ndMe commentedon Aug 4, 2026
Re-timed on latest
main(62024d1, cleannpm install+ rootnpm run build:test): the problem this issue measured no longer exists. Recommend closing.rejects missing/changed/extra/symlinked backup payloads— 32.3srestores a writable Automation authority after a real legacy cutover— 27.1scleans interrupted backup and restore staging— 16.3sWhat resolved it: #1994 (
refactor(storage): make SQLite the sole operational authority) deleted the legacy WAL-backup/cutover series that owned the three worst offenders — the migration those tests exercised is complete, so the fs/SQLite/git-heavy payload work went with it. The remaining real-process-crash and process-lock cases are already behindMAKA_STORAGE_STRESS(12 skips locally; CI enables the flag selectively inrun affected standard workspace tests), which is exactly the gating this issue proposed.Slowest remaining items for the record: Git workspace service 11.4s, SQLite Artifact store 7.1s, runtime policy stores 4.8s — none anywhere near critical-path territory at a 12s total.
Environment: macOS 26.6, Node v22.17.0. (One local-only failure, unrelated to timing:
top-level package import does not initialize node:sqliteasserts empty stderr, and Node 22.x prints the SQLite ExperimentalWarning; Node 24 on CI does not.)Astro-Han commentedon Aug 8, 2026
Closing as completed. The current-main remeasurement in the previous comment puts the storage suite at about 12.4 seconds rather than 88–105 seconds; #1994 removed the obsolete migration paths that dominated the original measurement, and the remaining stress cases are gated. A new regression should be filed from a fresh baseline.