Thread: foreign key check makes a big LOCK

Szima Gabor (sygma@marr.irisz.hu) reports a bug with a severity of 3
The lower the number the more severe it is.

Short Description
foreign key check makes a big LOCK

Long Description
in: src/backend/utils/adt/ri_triggers.c

RI_FKey_check(), RI_FKey_noaction_upd(), RI_FKey_noaction_del(), etc..
checking the referential with SELECT FOR UPDATE.

After BEGIN TRANSACTION: the INSERT/DELETE/UPDATE calling foreign-key checks, and the SELECT FOR UPDATE locking ALL
matchedrows in referential table. 

I modify ri_triggers.c (remove "FOR UPDATE"). This working.. but is correct?


Sample Code


No file was uploaded with this report

> Short Description
> foreign key check makes a big LOCK
> 
> Long Description
> in: src/backend/utils/adt/ri_triggers.c
> 
> RI_FKey_check(), RI_FKey_noaction_upd(), RI_FKey_noaction_del(), etc..
> checking the referential with SELECT FOR UPDATE.
> 
> After BEGIN TRANSACTION: the INSERT/DELETE/UPDATE calling 
> foreign-key checks, and the SELECT FOR UPDATE locking ALL 
> matched rows in referential table.
> 
> I modify ri_triggers.c (remove "FOR UPDATE"). This working.. 
> but is correct?

It's not. If one transaction inserts FK 1 and another one deletes
PK 1 at the same time both will succeed.

RI triggers should perform dirty reads (and test if returned tuples
alive/dead/being updated by concurrent transaction) instead of
SELECT FOR UPDATE but dirty reads are not implemented, yet.

Vadim