Based on this comment, it appears we expect deleting the txn coordinator doc at the end of commitTxn should be done with w:1 writeConcern. It seems we use DBDirectClient to run the delete operation, which does not specify writeConcern. As a result, we may be utilizing the implicit default writeConcern for the shard, which could be majority instead of w:1. This might be the case if we do not manually specify writeConcern on the opCtx before issuing the DBDirectClient, as DBDirectClient reuses the opCtx to perform the write.

We should start this ticket by trying to create a repro and see if the delete succeeds without waiting for majority.