Jonathan Foy <thefoy@gmail.com> writes:
> I was wondering if that was the problem. So I'm correct in thinking that
> the failure occurred when the vacuum tried to pull its 256 MB as defined in
> the maintenance_work_mem value, and the system just did not have enough
> available...any idea why that would suddenly start happening? The indexes I
> created shouldn't have affected that, should they?
Not directly, AFAICS, but they could stretch out the time required to
vacuum their tables, thus possibly leading to vacuums overlapping that
didn't overlap before. Just a guess though. Another likely bet is
that this is just an effect of the overall system load increasing
over time (more backends == more memory needed).
regards, tom lane