This hopefully addresses #1427. I still don't fully understand why this is the only test that occasionally fails in this manner. In theory a lot of our tests that sleep for exactly 1 epoch duration and expect the epoch to have advanced are making a big assumption about thread scheduling. Most should probably relax their sleep time like this change.
Coverage increased (+0.007%) to 76.53% when pulling b7b36382ebaf4aca4197c145ebcef60ed944436c on mbutrovich:serializable_txn_test into 898219f9fe6e1d2b6901ea3797caac5c93850213 on cmu-db:master.
This hopefully addresses #1427. I still don't fully understand why this is the only test that occasionally fails in this manner. In theory a lot of our tests that sleep for exactly 1 epoch duration and expect the epoch to have advanced are making a big assumption about thread scheduling. Most should probably relax their sleep time like this change.