RE: Still the case that rolling back to savepoint does not restart blocked sessions?
Date: Thu, 29 May 2014 21:17:12 +0000
Message-ID: <CE70217733273F49A8A162EE074F64D901DF6DE3_at_exmbx05.thus.corp>
No change in 12c.
There is a special case (in all versions) for pl/sql blocks -
if the first modification of the pl/sql block is the first modification of a transaction, and the savepoint is created before the first change, then the rollback to savepoint becomes a rollback of the complete transaction, viz: the following does a complete rollback:
commit;
begin
savepoint a;
update t1 set v1 = upper(v1) where id = 1;
rollback to savepoint;
end;
/
But this one leaves a transaction and lock in place:
commit;
begin
update t1 set v1 = lower(v1) where id = 1;
savepoint a;
update t1 set v1 = upper(v1) where id = 1;
rollback to savepoint;
end;
/
I recall a conversation many years ago describing some plans to allow the blocked session to check the locked row every 3 seconds to see if the lock had been released - but it looks as if that hasn't happened yet. (It's an interesting quirk of this situation that a row change that has been rolled back by a rollback to savepoint can be updated by a third session while the second session is still waiting for it (though really it's underlying transaction) to commit or rollback.)
Regards
Jonathan Lewis
http://jonathanlewis.wordpress.com
_at_jloracle
From: oracle-l-bounce_at_freelists.org [oracle-l-bounce_at_freelists.org] on behalf of Sayan Malakshinov [xt.and.r_at_gmail.com] Sent: 29 May 2014 19:30
To: Matt McPeak
Cc: oracle-l_at_freelists.org
Subject: Re: Still the case that rolling back to savepoint does not restart blocked sessions?
Oh, sorry, I have understood and even tested it on 12c: the same behaviour persists on 12.1 too:
Fixed test:
--first session:
drop table xt_test purge;
create table xt_test(id primary key, x) as select level,level from dual connect by level<=10;
begin
update xt_test set x=-x where id=1;
savepoint sp1;
update xt_test set x=-x where id=2;
dbms_lock.sleep(10);
rollback to sp1;
dbms_lock.sleep(30);
end;
/
-- lock in second until full commit/rollback:
update xt_test set x=x*10 where id=2;
On Thu, May 29, 2014 at 10:16 PM, Sayan Malakshinov <xt.and.r_at_gmail.com<mailto:xt.and.r_at_gmail.com>> wrote: Matt,
could you provide your example? Because I never seen such behaviour(or maybe i didn't understand you), and my test that I've made now, showed normal unlock.
My test:
--First session:
create table xt_test(id primary key, x) as select level,level from dual connect by level<=10;
begin
update xt_test set x=-x where id=1;
savepoint sp1;
update xt_test set x=-x where id=2;
rollback to sp1;
dbms_lock.sleep(30);
end;
/
-- and now in second:
update xt_test set x=x*10 where id=2;
On Thu, May 29, 2014 at 10:02 PM, McPeak, Matt <vxsmimmcp_at_subaru.com<mailto:vxsmimmcp_at_subaru.com>> wrote: If transaction A is waiting on a lock held by transaction B and transaction B rolls back to a savepoint (releasing the lock), transaction A remains blocked until transaction B ends completely (with either a full rollback or a commit).
Does anyone know if this is still the case in 12c? I really hate this behavior.
Thanks,
Matt
--
Best regards,
Sayan Malakshinov
Senior performance tuning engineer
PSBank
http://orasql.org
--
Best regards,
Sayan Malakshinov
Senior performance tuning engineer
PSBank
http://orasql.org
--
http://www.freelists.org/webpage/oracle-l
Received on Thu May 29 2014 - 23:17:12 CEST