Currently, sending remote messages is capped at three retries and local updates areĀ not retried on things such as interrupt due to killOp.
This is to ensure that the migration results (abort & commit) are persisted when they are accidentally interrupted by killOp.