M12.3: publish and recover durable exact populations
Foundation / checks (ecs) (push) Successful in 2m41s
Foundation / checks (tsan) (push) Failing after 18m12s
Foundation / server-image (push) Canceled after 0s
Foundation / image (push) Canceled after 0s
Foundation / checks (network) (push) Canceled after 20m41s
Foundation / checks (postgres) (push) Canceled after 20m43s
Foundation / checks (checks) (push) Canceled after 20m49s
Foundation / checks (deploy) (push) Successful in 5s
Foundation / checks (client) (push) Failing after 17m4s
Foundation / postgres-tools (push) Canceled after 0s

This commit is contained in:
dingo
2026-10-17 17:10:17 -04:00
parent d5d10a0cc7
commit 627411983f
35 changed files with 2637 additions and 102 deletions
+46
View File
@@ -1,5 +1,51 @@
# Validation report
## 2026-10-17 - M12.3b durable exact population publication
ADR-0142 affects portable lifecycle batches/checkpoint manifests, optional
PostgreSQL adapters/schema 44, the owning cellular host and opt-in worker batching.
No biological version or golden changes; no deployment or World0 operation.
Executed:
- MSVC 19.44 Release configure/build and **76/76 CTests** (final run 9.53 s).
- GCC 14.2 actual QUIC/PostgreSQL worker and integration-test build. Final private
`postgres-check.sh --population-only` **passed, exit 0, 5 s** after checked-access
refinements. Real reserved founders fund the wildlife fixture; tests exercise
exact membership oscillation, duplicate/conflicting requests, mutation/profile
race guards, missing/corrupt payloads, stale authority, host recreation and
materialization before player operations.
- Real PostgreSQL lease-row contention makes the adapter time out. One queued
request later commits and another backend is terminated before commit; both
resolve/retry the retained bytes. Forked publisher SIGKILLs occur before the
transaction, after commit and after a pipe ACK, with fresh-connection recovery.
This does not claim a population-specific network ACK or exec'd worker kill.
- Clang 19 actual server build; explicit clang-tidy 19 on changed optional store,
SQL adapter, cellular recovery, host, population integration tests and worker.
Final refinements use separately checked stable optional values; those follow-up
tidy runs pass. Analysis-preset portable checks and repository format check pass.
- Portable analysis CTests pass **7/7** (29.70 s); final batch-builder/checkpoint
checks pass **2/2** (3.73 s). ASan/UBSan population checkpoint/replay checks pass
(final 12.50 s), with no reports. Tests use explicit checks under NDEBUG.
The broad regression first exposed an over-broad mutation guard and the schema-44
upgrade expectation; both were corrected. The historical failed-migration drill
now supplies all required checksums, so it reaches its intended bad-v8-checksum
failure rather than an unrelated missing psql variable. The C++ suite, backup and
DR passed in the next run, but a concurrent local relink interrupted the upgrade
drill with `Text file busy`; this is not counted as a passing full run. Final
qualification uses a frozen copy of the executables to avoid that interference.
Final frozen-binary `postgres-check.sh --full` **passed, exit 0, 496 s**: complete
C++ suite (421 s), independent reservation connections, backup/restore, host-loss
DR, upgrade, migration history, permissions and checksum drift. The final optional
checked-access refinements additionally passed the targeted private population
suite above; all populations remain opt-in and no live region was changed.
Remaining scope: [M12_DURABLE_POPULATIONS](M12_DURABLE_POPULATIONS.md) and
[M12_REMAINING_GATES](M12_REMAINING_GATES.md). No compression, deployment-host
capacity, durable interval admission or multi-day ecological acceptance is inferred.
## 2026-10-17 - M12.4a unattended demand and confirmed-head pacing
ADR-0141 changes the existing cellular worker's demand/pacing and metrics, with