Re: Strange locking choices in pg_shdepend.c

Alvaro Herrera wrote:

> Hmm, unless revoking privileges concurrently, for two different users on
> the same object could cause a problem? I don't see us grabbing a lock
> on the object itself -- does this matter?

I tried a simple test: a process in a loop calling GRANT and REVOKE on random
users on a given table, and another process calling DROP OWNED BY
another set of users.

Prepare the test:

psql -c "create table foo()"
for i in `seq 0 100`; do psql -c "create user u$i"; done
for i in `seq 0 100`; do psql -c "create user v$i"; done
for i in `seq 0 100`; do psql -c "grant select on table foo to u$i"; done

Then, on one terminal
while true
do
r=$((RANDOM * 100 / 32764))
s=$((RANDOM * 100 / 32764))
psql -c "grant select on table foo to v$r"
psql -c "revoke select on table foo from v$s"
done

And another terminal

for i in `seq 1 100`; do psql -c "drop owned by u$i"; done

I get a lot of
ERREUR: tuple concurrently updated

So, yeah, I think our GRANT/REVOKE code has a race condition, which
probably isn't very critical at all but it's still there.

--
Alvaro Herrera Valdivia, Chile Geotag: -39,815 -73,257
"God is real, unless declared as int"