Patch release. No migrations.
- Fixed a 2.29.0 replay stall. A
select()whose condition member had a timeout could stop withhistory_shape_mismatchonce that member's timer fired. Affected runs resume after upgrading. Each run takes the member outcome first recorded in its history. - Condition waits with a timeout now follow durable input order, as named signal waits already did and as the docs describe. An update accepted before the wait's own timeout fired satisfies it, even if the workflow task runs after the deadline (for example during a worker outage). Updates accepted after the timeout fired cannot satisfy it. The first update that makes the predicate true wins, and later updates don't reverse it. This applies to standalone waits and to conditions inside
select()andall(). - Replay compatibility: runs whose history already records the wait as satisfied or timed out replay that outcome unchanged. 2.29.0 records a timeout before acting on it, so decided runs keep their behavior.
See #1193 and durable-workflow/.github#230.