Thread: How to mark a transaction as SERIALIZABLE?
In version 9.0, I've been reading http://www.postgresql.org/docs/9.0/interactive/sql-begin.html. I've been using the following :
CREATE OR REPLACE FUNCTION test(character varying)
RETURNS SETOF integer AS
$BODY$
BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE;
--SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;
RETURN 1;
END;
$BODY$
LANGUAGE plpgsql VOLATILE
which returns with the error :
ERROR: syntax error at or near "TRANSACTION"
LINE 1: TRANSACTION ISOLATION LEVEL SERIALIZABLE
if the BEGIN line is empty with no semi-colon, the function will work :
CREATE OR REPLACE FUNCTION test(character varying)
RETURNS integer AS
$BODY$
BEGIN
--TRANSACTION ISOLATION LEVEL SERIALIZABLE;
--SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;
RETURN 1;
END;
$BODY$
LANGUAGE plpgsql VOLATILE
but if I place a semi-colon after BEGIN (e.g. BEGIN;), I get the error :
ERROR: syntax error at or near ";"
LINE 5: BEGIN;
So any help you may be able to provide in the BEGIN syntax (or whatever I'm doing wrong) is greatly appreciated. Thanks!
-Joe
On 2 November 2010 01:51, Joe Carr <joe.carr@gmail.com> wrote:
In version 9.0, I've been reading http://www.postgresql.org/docs/9.0/interactive/sql-begin.html. I've been using the following :CREATE OR REPLACE FUNCTION test(character varying)RETURNS SETOF integer AS$BODY$BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE;--SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;RETURN 1;END;$BODY$LANGUAGE plpgsql VOLATILEwhich returns with the error :ERROR: syntax error at or near "TRANSACTION"LINE 1: TRANSACTION ISOLATION LEVEL SERIALIZABLEif the BEGIN line is empty with no semi-colon, the function will work :CREATE OR REPLACE FUNCTION test(character varying)RETURNS integer AS$BODY$BEGIN--TRANSACTION ISOLATION LEVEL SERIALIZABLE;--SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;RETURN 1;END;$BODY$LANGUAGE plpgsql VOLATILEbut if I place a semi-colon after BEGIN (e.g. BEGIN;), I get the error :ERROR: syntax error at or near ";"LINE 5: BEGIN;So any help you may be able to provide in the BEGIN syntax (or whatever I'm doing wrong) is greatly appreciated. Thanks!-Joe
BEGIN in your function isn't the same as BEGIN in an SQL statement block. It merely marks where the function begins, not the transaction. Note that you also shouldn't use a semi-colon after the BEGIN in a function. See this page for more info: http://www.postgresql.org/docs/9.0/interactive/plpgsql-structure.html
--
Thom Brown
Twitter: @darkixion
IRC (freenode): dark_ixion
Registered Linux user: #516935
Thanks for the link. I've done some more reading, and now it appears impossible to set the isolation level of a transaction from within a function.
CREATE OR REPLACE FUNCTION test(character varying)
RETURNS integer AS
$BODY$
BEGIN
SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;
RETURN 1;
END;
$BODY$
LANGUAGE plpgsql VOLATILE
Creates successfully, but when executed reports : "ERROR: SET TRANSACTION ISOLATION LEVEL must be called before any query". I believe this is happening because I'm exercising the function with :
select * from test('test');
so that select is executed prior to the SET TRANSACTION. If I remove the SET TRANSACTION command from the function, and then call it by :
SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;
select * from test('test');
that works. So, is that the correct way to set the isolation level for a function? Thanks again for your help.
On Tue, Nov 2, 2010 at 5:13 AM, Thom Brown <thom@linux.com> wrote:
BEGIN in your function isn't the same as BEGIN in an SQL statement block. It merely marks where the function begins, not the transaction. Note that you also shouldn't use a semi-colon after the BEGIN in a function. See this page for more info: http://www.postgresql.org/docs/9.0/interactive/plpgsql-structure.htmlOn 2 November 2010 01:51, Joe Carr <joe.carr@gmail.com> wrote:In version 9.0, I've been reading http://www.postgresql.org/docs/9.0/interactive/sql-begin.html. I've been using the following :CREATE OR REPLACE FUNCTION test(character varying)RETURNS SETOF integer AS$BODY$BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE;--SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;RETURN 1;END;$BODY$LANGUAGE plpgsql VOLATILEwhich returns with the error :ERROR: syntax error at or near "TRANSACTION"LINE 1: TRANSACTION ISOLATION LEVEL SERIALIZABLEif the BEGIN line is empty with no semi-colon, the function will work :CREATE OR REPLACE FUNCTION test(character varying)RETURNS integer AS$BODY$BEGIN--TRANSACTION ISOLATION LEVEL SERIALIZABLE;--SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;RETURN 1;END;$BODY$LANGUAGE plpgsql VOLATILEbut if I place a semi-colon after BEGIN (e.g. BEGIN;), I get the error :ERROR: syntax error at or near ";"LINE 5: BEGIN;So any help you may be able to provide in the BEGIN syntax (or whatever I'm doing wrong) is greatly appreciated. Thanks!-Joe
--
Thom Brown
Twitter: @darkixion
IRC (freenode): dark_ixion
Registered Linux user: #516935
Joe Carr <joe.carr@gmail.com> writes: > Thanks for the link. I've done some more reading, and now it appears > impossible to set the isolation level of a transaction from within a > function. That's correct. By the time the function gets control, you're already inside the transaction; it's too late to change its isolation level. regards, tom lane
joe.carr@gmail.com (Joe Carr) writes: > so that select is executed prior to the SET TRANSACTION. If I remove the SET > TRANSACTION command from the function, and then call it by : > > SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; > select * from test('test'); > > that works. So, is that the correct way to set the isolation level for a > function? Thanks again for your help. In effect, it's an illusion that you "set the isolation level for a function." Functions run in the pre-existing context of an existing transaction. (Yes, "existing" is pretty redundant there :-).) You already set the isolation level for the transaction - the function just uses the isolation already defined. If a function expects a particular isolation level, one could mandate this by checking the isolation level in the function, and raising an error if it doesn't meet up with expectations. The Postgres TODO list has an item, "Implement stored procedures" where the work suggested indicates: "This might involve the control of transaction state and the return of multiple result sets" That control of transaction state is what you're trying to do; it's not done, so that functionality doesn't exist at this time. -- wm(X,Y):-write(X),write('@'),write(Y). wm('cbbrowne','gmail.com'). http://www3.sympatico.ca/cbbrowne/postgresql.html "If you give a man a fish, he will eat for a day. If you teach him how to fish, he will sit in a boat and drink beer all day."