Public: anyone can read this; the crew can change it.
Full suite red: test/search-production-size.test.ts #12670
Check · Review · Land · Live done — Landed and live in production.
test/search-production-size.test.ts now runs in 5.7 s here, down from 42 s, against a 20 s file budget. The gate and tsc --noEmit printed nothing, and I ran only this file's tests. The time was all in bigForge writing about 600,000 rows.
shipped production is serving it deploy 77d0004
test/search-production-size.test.ts now runs in 5.7 s here, down from 42 s, against a 20 s file budget. The gate and tsc --noEmit printed nothing, and I ran only this file's tests.
The time was all in bigForge writing about 600,000 rows. /search never reads runs, comments or events. The test now keeps production's 9,000 issues and 9,000 pulls and the 1,000 extra repositories, which are the rows search does read. It writes a tenth of the runs, comments, events and review comments (SIZE in the test file).
The test's assertions are unchanged, and it still passes. The /search?q=a app phase measured 19 ms signed out and 28 ms signed in against the 300 ms budget, and the statement ceiling of 40 held. I did not add the file to SLOW in scripts/gate.ts.
To try it: bun test ./test/search-production-size.test.ts.
Test-only, scoped to the red file. SIZE keeps the 9,000 issues, 9,000 pulls and 1,000 repos that /search reads, and trims runs/comments/events/reviewComments to a tenth; the keys match the BigSize fields in test/helpers/big-forge.ts, assertions are untouched, and the comment says why. No UI, security or docs surface touched, so nothing to see in the preview. Two things I would do next: (1) assert in the test that the trimmed tables are not read by /search (or note which tables the statement ceiling covers) so a future search that joins comments is not silently measured against 10% of the data; (2) once the file is stable, check it is under the per-file budget on the forge check, not only locally (5.7 s vs 20 s leaves room).
test/search-production-size.test.tsnow runs in 5.7 s here, down from 42 s, against a 20 s file budget. The gate andtsc --noEmitprinted nothing, and I ran only this file's tests.The time was all in
bigForgewriting about 600,000 rows./searchnever reads runs, comments or events. The test now keeps production's 9,000 issues and 9,000 pulls and the 1,000 extra repositories, which are the rows search does read. It writes a tenth of the runs, comments, events and review comments (SIZEin the test file).The test's assertions are unchanged, and it still passes. The
/search?q=aapp phase measured 19 ms signed out and 28 ms signed in against the 300 ms budget, and the statement ceiling of 40 held. I did not add the file toSLOWinscripts/gate.ts.To try it:
bun test ./test/search-production-size.test.ts.Closes #12646