Details
Description
The migration cloner abort sequence does not transmit or validate the migration session id. Because of this it is possible that one shard aborts migration, which has been started by a completely different donor shard.
This bug is exacerbated in 3.3 because failure to execute the _recvChunkStart command is always followed-up by an abort. That way if two migrations with different donors, but the same recipient have the potential of conflicting with each other in which case one of the migrations will remain stuck.
Attachments
Issue Links
- is duplicated by
-
SERVER-25356 _recvChunkAbort should also send the migration session id
-
- Closed
-