coulomb/the-custodian
5
0
Fork 0
Code Issues 1 Pull requests Projects Releases Packages Wiki Activity Actions

Integrate configurable synthetic baseline and quota comparison
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
Details
CI Smoke / container-smoke (push) Successful in 11s
Details
Reject generated-only commits / generated-commit-guard (push) Successful in 11s
Details

Browse source
This commit is contained in:
codex 2026-10-19 15:00:49 +02:00
parent 04797021fd
commit 96be6d028e
1 changed files with 21 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

21
workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md
View file

@ -438,3 +438,24 @@ collector's grouping labels. T03 must reconcile these with owner sources and
representative demand before recommending a node budget. The successful
STATE-WP-0091 preflight applies to its own modeled release sequence only.
This capture changes no allocation and does not satisfy T02/T03/T04 acceptance.
## Configurable synthetic baseline delivered — 2026-10-19
Vergabe commit `11f7c8a` supplies the maintained HTTP/PostgreSQL driver,
`config/synthetic-pilot.json`, separate sizing-margin configuration and retained
raw/executed-source evidence under `docs/evidence/2026-10-19-synthetic-sizing/`.
The isolated run measured 30 seconds idle plus 180.889 seconds of ordinary work:
two users/workers, 100 initial tenders, 1MiB transfers and 0.2-second cycle think
time. All 3028 operations completed with zero errors; ordinary p95 was 227ms,
with transfer p95 95ms reported separately. Six focused owner tests pass.
This aggressive, unbounded local-host run measured aggregate application CPU
p95 1298m, observed peak 1917m and peak summed RSS 194MiB. Its configured safety
margins yield an experimental envelope of 1950m/2900m CPU and 256Mi/320Mi memory;
these are not approved deployment values. The CPU envelope exceeds both the
current 1000m application limit and the 395m reservation remainder. Do not raise
limits or claim the 60m reservation adequate from this evidence. The next
comparison uses an isolated app container with the existing 1000m/1Gi limits.
Source review also requested bounded cleanup on driver exceptions and explicit
measurement gaps on sampler failure. T02 stays progress while that comparison
and review proceed; human workflow acceptance remains separate.