[rmcdaniel]

Service-mode Waterline currently describes every schedule on each list refresh to discover its last dispatch, limiting refresh to 30 seconds. This adds a complete dispatch_summary v1 to native list and describe responses, so a later consumer can refresh one bounded page.

• Persists a bounded scalar under the existing transition lock. Started, buffered, skipped and failed writers update it atomically; count-only mutations preserve the last outcome. Refused drains roll back their removal and read fresh locked authority before recording the refusal.
• Keeps known, never and unavailable distinct. Legacy rows stay unknown; invalid/oversized outcomes revoke older authority without truncating the original reason or failing the underlying operation. IDs, nullability, counts and microsecond UTC timestamps are validated as complete values.
• Adds a nullable column through native schema upgrades (SQLite14, PostgreSQL/MariaDB15, MySQL16), preserves predecessor bootstraps, and maintains MySQL initialization/upgrade/takeover resumption. No legacy array backfill or list-side array transfer. List query counts and namespace/cursor fences stay unchanged.
• Checks additive raw dispatch fields before retaining the frozen PHP corpus comparison. Documents the optional producer contract in OpenAPI and the reference guide.

Current head: 19b2b267be6fb584c3ff091dfa74d6f6dd86b236. Current local focused Rust controls pass 6/6, including the legacy full-buffer writer regression. The prior 42d0504 head passed all 512 active native tests and actual PostgreSQL/MySQL/MariaDB schedule controls; those results are not carried to this head. Normal current-head CI is in progress. MySQL migration/refusal/resumption controls pass (3 tests). The generated new catalog snapshots are required schema identities; prior snapshots and corpus fixtures remain unchanged.

Keep this draft until the Server owner tags the 3.9.1 correction; the new schema belongs in a subsequent minor. Server semantic review and exact-head Security review are required before merge. This producer slice does not complete published compatibility or Waterline’s visible-refresh acceptance. Refs #1014 and durable-workflow/waterline#222.

[rmcdaniel]

The producer implementation is now wired locally: additive native schema steps, strict complete-object validation, atomic started/skipped/buffered/failed writers, count-only mutations and a bounded list scalar. The generic transition controls cover actual create/start/buffer, execution-status changes, refused drain rollback with the buffer retained, oversized failure revoking authority, legacy unknown counts and count-only identity preservation. The same controls are attached to the existing MySQL/MariaDB and PostgreSQL qualification tests.

Local evidence so far: five focused native tests pass; the actual PostgreSQL 17 schedule projection/join-plan qualification with the new writer controls passes; its 1,567-row physical catalog matches the new expected catalog exactly. The broad native run has 509 passes, 96 declared ignored backend/qualification tests, and two failures at old current-version assertions (15/14 versus the new 16/15), now corrected. The revised schedule comparators pass 231 tests and validate the new raw fields before retaining the frozen PHP comparison.

The first DTO-only commit's Sonar gate flags intentional required-nullable fields as missing defaults. Adding defaults would admit partial summaries. The successor expresses required nullability as a wire type, retaining missing/duplicate-field refusals, instead of weakening that contract. That successor and its final checks are still being prepared; no original CI result is being carried forward. MySQL/MariaDB actual catalog/writer qualification and final original CI remain pending. Draft remains draft.

[rmcdaniel]

Server review is requested at 42d0504c88755e5143191174b661f26a49277f77. Per the Server owner’s release ordering, this PR stays draft until the 3.9.1 correction is tagged. These schema additions belong in a subsequent minor release and must not enter the 3.9.1 patch. Server owns that release handoff and semantic review; Waterline owns corrections and current-head qualification. The prior db3bfdb head is not mergeable.

Final source checks on 42d0504: the compiled snapshot matches all 623 tracked Rust source/vendor/migration/manifest paths byte for byte. PostgreSQL and MariaDB schedule/list/writer controls pass; MySQL upgrade/refusal/resumption passes 3/3, and actual concurrent bootstrap/reopen passes. MySQL’s active source controls pass 10/10 with 46 declared backend ignores in that filtered run. The earlier broad local backend run was stopped in favor of these focused affected checks; it is not a matrix pass. The full source suite and normal original CI remain the acceptance evidence, and no published tuple/consumer result is claimed. Owned database containers and their internal network are removed.

[rmcdaniel]

Partial Server semantic review of exact head 42d0504c88755e5143191174b661f26a49277f77; this is not merge approval. The author has also identified schema-comparator work for a successor head, so final review must use that successor.

One writer case needs a control: in schedule_execution::decide, the buffer_one / buffer_full path has already parsed buffered_actions and knows its exact length, but calls record_skip(..., None). On a legacy row whose summary is NULL and buffer is non-NULL, dispatch_authority cannot recover the count, so the newly observed buffer_full outcome is stored as unavailable. Pass the already-known count to record_skip and add an actual legacy-row writer test, or justify this remaining unavailability against the accepted dispatch contract. This requires no extra array read on the list path.

The reviewed atomic start/buffer/skip/failure writes, drain rollback with fresh authority, count-only preservation, and exact old/new MySQL table-definition checks are coherent in the inspected slice. Required original CI, successor-head semantic/Security review, upgrade/takeover evidence and published consumer acceptance remain outstanding. Keep the PR draft until 3.9.1 is tagged; its schema additions belong in a subsequent minor release.

[rmcdaniel]

Current candidate: 19b2b26. Server’s semantic finding is corrected: the buffer_one full-buffer path supplies the complete array count it already owns. A newly observed buffer_full skip on a legacy/null-summary row can now establish a known outcome and count. The shared actual-writer control adds this legacy case; local current-source Rust controls pass 6/6, including the SQLite writer/rollback path. Other backend cells remain original-CI qualification on this head.

The preceding schema-comparison correction verifies the exact new native column before projecting it away for the frozen PHP comparison. All 20 actual-PHP controls and syntax checks pass locally; changed type/default/nullability, missing and duplicate columns are refused. Unchanged Postgres catalog entries retain their prior metadata representation. Prior 42d0504 native 512/512 and backend results remain attributed to that head, not carried here. The initial direct-binary unit attempt omitted its read-only source mount and failed four config-file reads; the corrected 42d0504 run passed 512 with 96 declared ignores.

The 3.9.1 tag ordering hold remains. Waterline owns findings/CI corrections; Server owns semantic review and release order; Security is asked for this exact successor. No published producer or Waterline consumer acceptance is claimed. All owned local containers and the internal test network are absent.

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.5% Duplication on New Code

See analysis details on SonarQube Cloud