Re: MERGE statement with parallel DML enabled deadlocks itself
From: Thomas Kellerer <thomas.kellerer_at_mgm-tp.com>
Date: Wed, 06 May 2015 15:44:15 +0200
Message-ID: <554A1AAF.50905_at_mgm-tp.com>
Jonathan Lewis schrieb am 06.05.2015 um 13:58:
> Possibly a side effect of the degree of parallelism. I don't know how
> likely it is but with 33 partitions it's possible that 3, 11, and 33
> are the only degrees where pw joins might occur.
>
> Possibly it relates to the skew in the size of the partitions -
> especially if some don't exist to start with.
>
Date: Wed, 06 May 2015 15:44:15 +0200
Message-ID: <554A1AAF.50905_at_mgm-tp.com>
Jonathan Lewis schrieb am 06.05.2015 um 13:58:
> Possibly a side effect of the degree of parallelism. I don't know how
> likely it is but with 33 partitions it's possible that 3, 11, and 33
> are the only degrees where pw joins might occur.
>
> Possibly it relates to the skew in the size of the partitions -
> especially if some don't exist to start with.
>
Thanks. We will try and drop those partitions that are empty.
Can you imagine a reason why this would not be reproducable in a different environment?
On our development servers we have a similar situation (not all partitions actually have data) but regardless of which degree of parallelism I choose I never get a deadlock there.
Regards
Thomas
-- http://www.freelists.org/webpage/oracle-lReceived on Wed May 06 2015 - 15:44:15 CEST