Christopher Kings-Lynne <chriskl@familyhealth.com.au> writes:
> What is going on here? Surely getting a FOR UPDATE row lock should
> prevent another process getting an update lock?
I could not duplicate your results. I did
regression=# create table tab(id int , blah int);
CREATE TABLE
regression=# insert into tab values(1,1);
INSERT 320558 1
regression=# insert into tab values(1,2);
INSERT 320559 1
regression=# insert into tab values(2,3);
INSERT 320560 1
regression=# BEGIN;
BEGIN
regression=# SELECT * FROM tab WHERE id=1 FOR UPDATE;id | blah
----+------ 1 | 1 1 | 2
(2 rows)
<< in another window >>
regression=# UPDATE tab SET blah=1 WHERE id=1;
[waits]
<< back to first window >>
regression=# UPDATE tab SET blah=1 WHERE id=1;
UPDATE 2
regression=# end;
COMMIT
<< second window now reports >>
UPDATE 2
regression=#
The behavior you describe would certainly be a bug, but you'll have to
show a reproducible example to convince me it wasn't pilot error. One
idea that springs to mind is that maybe additional rows with id=1 were
inserted (by some other transaction) between the SELECT FOR UPDATE and
the UPDATE?
regards, tom lane