See the linked BF for more details, but an update using the non internal transaction implementation that changes a document's shard key value and is retried on an Interruption error can fail with ConflictingOperationInProgress because a mongos already "started" tracking a transaction with the update's txnNumber. This test retries on interruption errors since it expects them, but in practice a driver won't retry on them, so a test only fix to ignore ConflictingOperationInProgress seems appropriate.