Re: BUG #4596: information_schema.table_privileges is way too slow
- From: Kirill Simonov <xi(at)gamma(dot)dn(dot)ua>
- To: Pavel Stehule <pavel(dot)stehule(at)gmail(dot)com>
- Cc: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>, pgsql-bugs(at)postgresql(dot)org
- Subject: Re: BUG #4596: information_schema.table_privileges is way too slow
- Date: Thu, 25 Dec 2008 15:36:48 +0200
- Message-id: <49538C70.5030403@gamma.dn.ua> <text/plain>
Pavel Stehule wrote:
2008/12/25 Kirill Simonov <xi(at)gamma(dot)dn(dot)ua>:
Tom Lane wrote:
"Kirill Simonov" <xi(at)gamma(dot)dn(dot)ua> writes:
It takes about 5 minutes to perform the query
SELECT * FROM information_schema.table_privileges
on an empty database (i.e. with system tables only).
Not here. What non-default settings might you be using?
Indeed, it is slow because there are a lot of rows in pg_authid (about 700).
Is there a possibility to make table_privileges faster with a large number
of roles?
Thanks,
Kirill
two years ago I tested 50000 users without problems. Try to vacuum and
reindex your system tables
Neither VACUUM nor REINDEX SYSTEM did help. The problem could be
reproduced on a freshly installed Postgres:
-- add a function to generate dummy roles.
create language plpgsql;
create function create_dummy_role(start int, finish int) returns void as $$
begin
for i in start..finish loop
execute 'create role dummy_' || cast(i as text);
end loop;
end;
$$ language plpgsql;
-- no extra roles
select count(*) from information_schema.table_privileges;
>>> Time: 11.467 ms
-- 10 roles
select create_dummy_role(1, 10);
select count(*) from information_schema.table_privileges;
>>> Time: 161.539 ms
-- 100 roles
select create_dummy_role(11, 100);
select count(*) from information_schema.table_privileges;
>>> Time: 7807.675 ms
-- 1000 roles
select create_dummy_role(101, 1000);
select count(*) from information_schema.table_privileges;
>>> Time: 543030.948 ms
Thanks,
Kirill
Home |
Main Index |
Thread Index