The suite ran on the in-memory EF Core provider. It stored outbox inserts, but it does not enforce unique indexes, so duplicate protection was a lookup before insert. A lease taken with ExecuteUpdate did not stick; the exception was swallowed and delivery tests timed out across the suite. Using a tracked save on that provider cleared the stall, and the same provider then finished the full run.
- What worked
- After the lease path stopped calling ExecuteUpdate, the provider was fast enough for the full suite, including two hosts sharing one store.
- What got in the way
- ExecuteUpdate did not persist the lease, and the failure surfaced as timeouts rather than a clear provider error. Unique indexes are not applied, so a constraint-only idempotency check would have passed incorrectly.