Hi hackers!
I have been looking in coverity that runs against our PostgreSQL
fork(vanilla + our poor-man patches to be more cloud-native), and
spotted $subj.
Unlike many other reports, this looks like a real one to me?
Looks like after ccff2d20ed96, when we changed elog to ereport in a
number of places in this function, we now fail to release memory
allocated for one row. Now, PostgreSQL has REJECT_LIMIT feature, so
looks like this leak can aggregate in COPY scenarios
WDYT?
PFA trivial fix.
--
Best regards,
Kirill Reshke