Investigate 200k Money published-refresh tail latency #1580

Open
opened 2026-10-11 18:28:05 +00:00 by kayg · 0 comments
Owner

Evidence

The bench/money-register-688.mjs release profile ran through perf-run on perf-test-69-63 under the shared benchmark lock. The recorded in-lock load average was 2.70 2.10 2.21. It measured 200,000 rows with MONEY688_PUBLISHED_REFRESH_ONLY=1:

  • Published refresh serial, n=5: p50 1,397.3 ms; p95 9,044.4 ms.
  • Published refresh burst, n=4: p50 2,428.8 ms; p95 5,078.3 ms.
  • The write-count assertions passed: 11 row updates, zero row inserts/deletes, zero Category writes, and 40 consumed hints.

The baseline's 200k cleared-mark p50/p95 is 870.4/992.8 ms, with only one completed sample. This is a signal to reproduce and investigate, not a clean before/after comparison: the current profile includes write-count instrumentation and its JSON says qualification=local because MONEY688_QUALIFICATION=perf-vm was omitted, although execution was on the perf VM through perf-run.

Follow-up

Repeat the release profile with corrected qualification metadata and compare the same operation and measurement method. Check the high serial p95 before changing the implementation. This is a performance follow-up and did not block merge40 qualification.

## Evidence The `bench/money-register-688.mjs` release profile ran through `perf-run` on perf-test-69-63 under the shared benchmark lock. The recorded in-lock load average was `2.70 2.10 2.21`. It measured 200,000 rows with `MONEY688_PUBLISHED_REFRESH_ONLY=1`: - Published refresh serial, n=5: p50 1,397.3 ms; p95 9,044.4 ms. - Published refresh burst, n=4: p50 2,428.8 ms; p95 5,078.3 ms. - The write-count assertions passed: 11 row updates, zero row inserts/deletes, zero Category writes, and 40 consumed hints. The baseline's 200k cleared-mark p50/p95 is 870.4/992.8 ms, with only one completed sample. This is a signal to reproduce and investigate, not a clean before/after comparison: the current profile includes write-count instrumentation and its JSON says `qualification=local` because `MONEY688_QUALIFICATION=perf-vm` was omitted, although execution was on the perf VM through `perf-run`. ## Follow-up Repeat the release profile with corrected qualification metadata and compare the same operation and measurement method. Check the high serial p95 before changing the implementation. This is a performance follow-up and did not block merge40 qualification.
Commenting is not possible because the repository is archived.
No labels
No milestone
No project
No assignees
1 participant
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
kayg/calternal-v0#1580
No description provided.