Skip to main content

2026-10-09-top-level-transaction-retry

title: Every top-level ctx.transaction retries a deadlock affects: dartway_core_server: "0.21.0-dev.17"​

Who is affected​

A project that opens ctx.transaction outside a transactional command — in a route, a transactional: false command, a leased (transactional: false) job or a startup step — or that wraps one in a retry loop of its own. A top-level ctx.transaction now retries a deadlock (40P01) or a serialization failure up to three attempts, with the memo cleared between them; a transactional queued job and a recurring job re-run their handler in place the same way. Nothing breaks if a project does neither step below.

What to change​

  1. Read every such ctx.transaction body for work outside the database — an outbound HTTP call, a message sent, a file written. The body may now run up to three times: move that work before or after the transaction.
  2. Delete hand-rolled retry loops around ctx.transaction that catch DwSerializationFailure or SQLSTATE 40P01. Left in place, such a loop nests its own attempts around the framework's, up to three times three, and does not clear the memo between them.

How to check​

grep -rn "40P01\|DwSerializationFailure" <project>_server/lib finds no retry loop, and no ctx.transaction body in a route, a non-transactional command, a leased job or a startup step calls another service.