Tom Lane wrote:
Misestimated hash aggregation, perhaps? What is the query and what does
EXPLAIN show for it? What have you got work_mem set to?
oom_test=> \d oom_tab
Table "public.oom_tab"
Column | Type | Modifiers
--------+---------+-----------
val | integer |
oom_test=> explain select val,count(*) from oom_tab group by val;
QUERY PLAN
-------------------------------------------------------------------------
HashAggregate (cost=1163446.13..1163448.63 rows=200 width=4)
-> Seq Scan on oom_tab (cost=0.00..867748.42 rows=59139542 width=4)
The row estimitate for oom_tab is close to the actual value. Most of
the values are unique, however, so the result should have around 59M
rows too.
I've tried it with work_mem set to 32M, 512M, 1G, and 2G. It fails in
all cases, but it hits the failure point quicker with work_mem=32M.
-- todd
