The title of this ticket is my best guess at what's happening in the following situation . (This came from this slack conversation). See the attached test; occasionally, the test will loop continuously in the second transaction as the first transaction has not properly aborted its identical write server-side. I believe this is because we do not run AbortTransaction correctly when CommitTransaction encounters a context cancelation, but I'm not exactly sure. This requires further investigation and even if the aforementioned theory is correct, we need a more consistent regression test.