It seems in _assertCommandWorked, which we call from assert.commandWorked, we handle writeConcern errors differently from other errors. If assert.commandWorked catches a writeConcern error, it will immediately run the hang analyzer and fail.

As a result, we should not use assert.commandWorked while waiting for a writeConcern inside an assert.soon(), if our intention is to keep retrying the write with the specified writeConcern.

This ticket will specifically fix this bug in user_defined_roles_on_secondaries.js, there may be other places where we try to do this.