Repository navigation
[Task] Preserve atomic BPMN edits in Dapper workflow definitions #8293
Description
Activity
The isolated Dapper implementation now uses one SERIALIZABLE read/predicate/write transaction, preserving existing SaveAsync writer participation without a new CAS stamp. Nine conformance scenarios pass independently on SQLite, PostgreSQL 17 and SQL Server 2022: current metadata, failed predicate, missing row, atomic draft creation, rollback, competing legacy writer, two-worker contention and tenant preservation/rejection. SQL Server with an artificially short one-second command timeout correctly propagated timeout rather than misreporting it as a conflict; its normal 15-second bound allowed deadlock detection and the suite passed. PostgreSQL proof exposed and fixed existing primary-key omission in Upsert and integer-vs-Boolean version predicates. Final patch/guard tests and review are in progress; no source import or package publication is claimed.
Preparation PR: #8297 . Final suite totals are 28/28 SQLite (including existing query/bookmark tests), 15/15 PostgreSQL and 15/15 SQL Server. A clean patch replay matched the reviewed source byte-for-byte and passed 28 SQLite tests. Exact pins, hashes, database images, limits and reproduction instructions are in the PR. This task stays open until incorporation and validation in the actual consolidated source.
Preparation PR #8297 merged as 9fee71f after all exact-head checks, Greptile5/5 and resolved review threads. Final recorded tests:29/29 SQLite (also clean patch replay),16/16 PostgreSQL,16/16 SQL Server; all declared net8/net9/net10 targets build. This is the reviewed patch artifact, not actual imported source; #8293 stays open until consolidated integration and final-layout validation. The merge subject suppresses automatic publishing; no package feed publication was authorized.
- added a commit that references this issue
on Sep 27, 2026 #8506 merged into draft import #8409 as
ce1111e4421b0e2af9852d55d9e2ebde937fcab5(merge commit). It passed exact-head checks, includingmapped-source-artifact-proofandimport-rehearsal-tests, and all review threads were resolved (Greptile 4/5, one non-blocking P2).Why it was needed: the Dapper
TryUpdateLatestAsyncthat had reached the import (from Extensions #216) was unreviewed and diverged from the reviewed #8297 patch:- no transaction, and tenant scope ignored;
- whole stale rows written back;
- missing the PostgreSQL primary-key and Boolean-filter fixes.
The reviewed tests fail against it: 4 on SQLite, all 17 on PostgreSQL.
What changed: the imported code now matches the reviewed patch. One deliberate difference: a superseded version returns Conflict, matching the other stores.
Tests:
- SQLite 35/35 on net10 and net9.
- PostgreSQL 17.11 and SQL Server 2022 CU27: 17/17 each, over five runs.
- Net8 is build-only; a test dependency doesn't support it.
Receipts: the new sixth receipt (
source-tip-refresh-2026-09-27-r6.json) pins the change. r1 and r5 defer to it only for the two superseded paths.Follow-up: running the PostgreSQL and SQL Server tests in CI is tracked separately.
This issue stays open until the history-preserving import reaches
main. No package publication occurred.
Program: #8194
Parent: #8286
Related: #8287
Deliverable
Implement atomic latest-definition compare-and-swap in the Dapper persistence provider so BPMN document PUT uses the same concurrency contract as the in-memory and EF Core stores.
Acceptance
Scope boundary
This task implements and verifies the Dapper provider path only. It does not claim MongoDB or every distributed persistence provider is concurrency-safe, and it does not publish packages.