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

CUST-WP-0071: T02 done on target-hardware replay (provisional sizing)

Browse source 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
codex 2026-10-21 08:42:16 +02:00
parent 9edd3151ac
commit 648743c640
2 changed files with 23 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

2
WORK-RECORDS.md
View file

@ -464,7 +464,7 @@
| task | CUST-WP-0070-T02 | done | — | workplans/CUST-WP-0070-publication-repo-category.md |
| task | CUST-WP-0070-T03 | done | — | workplans/CUST-WP-0070-publication-repo-category.md |
| task | CUST-WP-0071-T01 | done | — | workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md |
| task | CUST-WP-0071-T02 | progress | — | workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md |
| task | CUST-WP-0071-T02 | done | — | workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md |
| task | CUST-WP-0071-T03 | wait | — | workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md |
| task | CUST-WP-0071-T04 | wait | — | workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md |
| task | CUST-WP-0071-T05 | wait | — | workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md |

24
workplans/CUST-WP-0071-measured-workload-sizing-and-weekly-review.md
View file

@ -94,11 +94,10 @@ updated reef-railiance-k3s owner evidence.
```task
id: CUST-WP-0071-T02
status: progress
status: done
priority: high
assignee: the-custodian
depends_on: [CUST-WP-0071-T01]
blocking_reason: "Configurable synthetic workload and real isolated runtime measurements in preparation under the 2026-10-19 interview decision. Human pilot acceptance remains RAPPS-WP-0014-T03 and does not gate provisional sizing."
state_hub_task_id: "82192370-2fd7-5363-88d2-3c67889d3d68"
```
@ -548,3 +547,24 @@ itself, one owner at a time).
The follow-up adds ≥8-day retention, hub-message delivery with a missed-run
check, and default-deny egress.
No production allocation has changed yet.
## T02 provisional sizing accepted on target-hardware evidence — 2026-10-21
RAPPS-WP-0014-T04 is done (railiance-apps `1af6d82`, harness vergabe-teilnahme
`dca6683`; evidence `railiance-apps/docs/evidence/2026-10-21-vergabe-target-replay/`).
The replay ran the deployed digest `sha256:a26444f5…` with 3 workers at the live
60m/256Mi requests and 1000m/1Gi limits, in an isolated namespace on railiance01.
In-cluster Service latency, phased profile: 3675 ordinary requests with 0 errors
and p95 0.192s; 814 transfers with 0 errors and p95 0.076s; restart to ready in
19.0s. App CPU averaged 775m (p95 998m) with 35.7% of periods throttled; memory
peaked at 483MiB with no OOM. Incremental database CPU was about +320m. The
namespace was torn down and verified gone; the live deployment is unchanged.
Provisional sizing: keep the 60m CPU request and the 1000m/1Gi limits. Raise the
memory request to 512Mi in the next allocation wave. Budget 0.3–0.5 core of
apps-pg demand at synthetic load. Revisit triggers: more users or companies, an
image or worker change, live p95 above 1s, node requests above 95%, sustained free
CPU below 500m, or apps-pg above 600m. Gaps: no SSO, ingress or TLS; a standalone
database instead of CNPG; a synthetic closed loop; a single quiet-hour run. These
results are provisional. They do not prove human workflow acceptance, which stays
with RAPPS-WP-0014-T03 / VERGABE-WP-0019.