Erik Jones <erik@myemma.com> writes:
> Well, if Postgres had killed the proc itself it would have written
> out a nicely formatted Postgres-style memory context report along
> with an ERROR message along the lines of OUT OF MEMORY and the
> request size and Postgres would not have bounced. Since the
> postmaster dropped into recovery mode when the proc received the
> SIGABRT and died, that means that the signal came from somewhere
> else, OOM killer?
No, an abort() is expected when glibc's malloc code detects a problem,
and all that other junk is stuff that malloc helpfully prints on stderr
before committing hara-kiri.
This seems clearly a memory-stomp bug of some kind (although there's
a very small probability that it was a transient RAM glitch). Not much
we can do about it without a test case, though.
regards, tom lane