Re: Clustering with minimal locking

On Wed, Jun 18, 2008 at 9:26 AM, Decibel! <decibel(at)decibel(dot)org> wrote:

> On Jun 17, 2008, at 11:37 AM, Scott Ribe wrote:
>
>> BOOM! Deadlock.
>>>
>>
>> No more likely than with the current cluster command. Acquiring the lock
>> is
>> the same risk; but it is held for much less time.
>>
>
>
> Actually, no (at least in 8.2). CLUSTER grabs an exclusive lock before it
> does any work meaning that it can't deadlock by itself. Of course you could
> always do something like
>
> BEGIN;
> SELECT * FROM a;
> CLUSTER .. ON a;
> COMMIT;
>
> Which does introduce the risk of a deadlock

Really!!? Am I missing something? How can a single transaction, running
synchronous commands, deadlock itself!

Best regards,
--
gurjeet[(dot)singh](at)EnterpriseDB(dot)com
singh(dot)gurjeet(at){ gmail | hotmail | indiatimes | yahoo }.com

EnterpriseDB http://www.enterprisedb.com

Mail sent from my BlackLaptop device