Results for the global reflog seem to be non-deterministic in transcripts :'(
It seems unlikely, but the only reason I could think of for this to happen with the current ORDER BY is if two rows are inserted with an identical timestamp; maybe getCurrentTime returns the same time if called within a ms or something.
It's extremely unlikely this would ever occur outside of a transcript environment, but either way, just adding the ROWID as an additional sorting column guarantees they'll sort in insert order.
If we see this again we should assume there's some concurrency within the transcript runner (I did a quick scan and couldn't see any, even though we do use TVars for some reason) and look at sorting that out instead 👀
Implementation notes
Sort by RowID in addition to time.
Test coverage
Unfortunately I can't force flakiness, so we'll just have to see if it comes back.
aryairani
commented
3 months ago
Overview
fixes #5271
Results for the global reflog seem to be non-deterministic in transcripts :'( It seems unlikely, but the only reason I could think of for this to happen with the current ORDER BY is if two rows are inserted with an identical timestamp; maybe
getCurrentTimereturns the same time if called within a ms or something.It's extremely unlikely this would ever occur outside of a transcript environment, but either way, just adding the ROWID as an additional sorting column guarantees they'll sort in insert order.
If we see this again we should assume there's some concurrency within the transcript runner (I did a quick scan and couldn't see any, even though we do use TVars for some reason) and look at sorting that out instead 👀
Implementation notes
Sort by RowID in addition to time.
Test coverage
Unfortunately I can't force flakiness, so we'll just have to see if it comes back.