Re: [HACKERS] Help with pl/pgsql, triggers, and foreign keys

Ed Loehr wrote:

> At least in 6.5.2, you can definitely implement referential integrity
> (RI) via pl/pgsql. As someone noted earlier, RI is to be released in
> 7.0, but I suspect it will take a subsequent release or two to
> stabilize before it's fit for consumption by the more conservative
> reliability-focused users among us...

I hope that this isn't true.

First, because FOREIGN KEY is implemented as builtin triggers
written in C. BETA should turn out most of the bugs, which
could still be in it.

Second, RI cannot get implemented reliable with regular
triggers. You can easily violate the semantics with
concurrently running transactions. Have first transaction
inserting a reference, the trigger checks for key existence
and finds it. Now second transaction deletes the key, and an
eventually existing ON DELETE CASCADE trigger fired on that
wouldn't find the reference, because it isn't committed yet.
Second transaction commits, what finally removes the key. Now
first transaction commits, making the reference visible, but
referencing a non existing key - inconsistency.

So anyone who needs referential integrity is asked to stress
the code as far as he can, at least during BETA.

Jan

--

#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#========================================= wieck(at)debis(dot)com (Jan Wieck) #