mirror of
https://github.com/MBombeck/HealthLog.git
synced 2026-11-08 12:32:43 +00:00
Marc-André Bombeck
a451a114a2
The profile section in docs/contributing/benchmarks.md says how to run scripts/bench/boot-profile.ts, what the profile of the v1.44 runtime found (the worker graph as one 753 ms block, a live heap of about 165 MB of which the query compiler holds 50 and the catalogs 33) and what changed. scripts/bench/results/v1.45.0-*.md hold the before and after numbers per area, from back-to-back runs on the same machine. The size baselines take the lowered numbers.
1.5 KiB
1.5 KiB
v1.45.0: write path
Same runs as v1.45.0-read-path.md.
| request | before p50 ms | after p50 ms | statements before / after |
|---|---|---|---|
| batch ingest, 500 fresh rows per request | 166 | 186 | 53 / 50 |
Fresh rows were already set-based (v1.44); the benchmark's ingest is all fresh rows, so it shows no gain, as expected. The changes are for re-posts:
- Unchanged
stats:re-posts write nothing and start nothing. One read over the batch'sstats:ids replaces a reconciler transaction (about ten round trips) per row, and an all-unchanged batch no longer refolds, enqueues rollups or PR detection, or marks caches stale. The per-entry status staysupdatedand the summary'supdatedcount is unchanged.tests/integration/routes/measurements-batch-unchanged-stats.test.ts: four unchanged re-posts take at most 12 statements in total (before: about 40) with zero rollup, invalidation and PR calls; a changed value, a changed device or a tombstoned row still goes through the reconciler. - Apple Health re-import skips the UPDATE of a row stored with exactly the
values it would write; one read per chunk instead of one UPDATE per row.
tests/integration/apple-health-reimport-unchanged.test.ts: a re-import of 20 stored rows reports the same per-type counts and sends no UPDATE. - Narrower invalidation. The batch route tells the cache how far back it
wrote (
touchedFrom), so closed blocks of older readings stay.