Possible race condition in the publish log causing a forked feed
Closed this issue · 1 comments
Basically an internalized version of the problem encountered in #289:
Line 91 in ab468b1
When publishing a new message, the previous sequence number query relies on the author log being in sync with the main log. The problem is that this seems to not always be the case. I suspect this is related to #251, although I don't know the details of how multilogs are implemented internally to know for sure.
One way or another, it's difficult but likely not impossible (as we saw in #289) for this to trigger a wrong sequence number result, which would in turn result in a wrong query of the main log for the previous message, which would in turn cause a forked feed.
This should be possible to trigger by publishing a huge number of messages posted by the same author in quick succession so that the main feed gets ahead of pl.byAuthor.
I've got a fix for this, but we still need a test that tests for this specifically so we can prove that it stays fixed.