-
Type: Task
-
Resolution: Fixed
-
Priority: Major - P3
-
Affects Version/s: None
-
Component/s: Replication
-
Fully Compatible
-
Sharding 2020-09-21, Sharding 2020-10-05
The existing non-deterministic test theoretically should cover all the possible states, but it's tending to mostly cover the node shutting down after the migration has already committed.
This ticket is to either:
1) Make the test randomly set one of three failpoints before starting the migration thread:
- A failpoint to make the migration hang in the dataSync state. This failpoint doesn't exist yet and could be added here.
- A failpoint to make the migration hang in the blocking state (already exists)
- A failpoint to make the migration abort (already exists)
or
2) Split the test into 4 cases, one for each failpoint (to cover kDataSync, kBlocking, and kAborted) plus one without any failpoints (to cover kCommitted).
- depends on
-
SERVER-50616 TenantMigrationDonor should retry its steps until success, the donor node is stepping down or shutting down, or the donor gets an error that should lead to an abort decision
- Closed