As a part of the preSetup we run the precmdShardKey function which runs a chunk split. The split can trigger a metadata refresh that has a majority write concern. That write concern will fail since previously in the testing code we have taken down one of the secondaries.

To resolve this we need to make the test resilient to write concern failures in the preSetup. Either we ignore the errors and skip the test, or figure out some other workaround.