Re: Strange VACUUM behaviour

"Florian G. Pflug" <fgp(at)phlo(dot)org> writes:
> INFO: index "pg_toast_2144146_index" now contains 1971674 row versions
> in 10018 pages
> DETAIL: 4 index row versions were removed.
> 2489 index pages have been deleted, 0 are currently reusable.

> Just for the archives - I finally solved the problem - and of course
> it was me being a dumbass. I _wanted to set vacuum mem to 256mb, but
> overlooked that the setting is in kb, not in bytes - so the value
> I set was actually 256GB - which seems to have overflowed to some
> awefully small value, and caused the vacuum to run in really small steps...

Ah-hah, I should have thought of that :-(. I thought the 4-at-a-time
thing was pretty odd ...

> Might be worth an overflow check and/or some notice in the docs, though

There is an overflow check for this and other memory-size parameters in
PG 8.1.

regression=# set maintenance_work_mem = 256000000;
ERROR: 256000000 is outside the valid range for parameter "maintenance_work_mem" (1024 .. 2097151)

regards, tom lane