txn_participant_adds_additional_partipants and txn_participant_adds_additional_partipants_with_aborts rely on failpoints and transaction metrics from serverStatus to test the appropriate behavior. serverStatus provides metrics around all transactions that have run and it is not possible to differentiate between different sessions. Therefore the current implementation provides a very loose assertion around the behavior we would like to test.

We should refactor the tests / code to be able to have stronger assertions. One option would be to convert the integration tests into unit tests. Or we could refactor the code to provide more metrics / information about distinct transactions so that we do not have to rely on serverStatus.