Re: Can't get INSERT /*+ APPEND */ to indicate direct load in execution plan.....

  • From: "Radoulov, Dimitre" <cichomitiko@xxxxxxxxx>
  • To: Greg Rahn <greg@xxxxxxxxxxxxxxxxxx>
  • Date: Wed, 02 Nov 2011 16:47:26 +0100

Greg,
this is quite useful bit of information.

So now we know that Mark's insert was performed in serial :)


Thank you!
Dimitre

P.S. What's SQL Monitor and does it require a separate license?


On 02/11/2011 15:47, Greg Rahn wrote:
> The way to tell if the insert is using PDML or not is by the placement
> of the LOAD row source:  If it is above the PX COORDINATOR line, its
> all done by the QC.  If below the PX COORDINATOR row source, its done
> by the PX servers.  SQL Monitor will show this nicely on the Parallel
> tab, just expand the tree/nodes and look at the amount of work done.
>
> On Wed, Nov 2, 2011 at 12:38 AM, Radoulov, Dimitre
> <cichomitiko@xxxxxxxxx>  wrote:
>> Thanks Greg,
>> could you please clarify the following:
>>
>>>> It's still using PX ...
>> Yes, I see that parallel execution is involved.
>>
>> I wanted to know if the *insert* part of the statement
>> was executed in parallel and where is the evidence for that.
>> I believe that only the select part execution mode is visible
>> in the plan output, but I may be wrong, of course.
>>
>> My point was that as far as Mark's first mail is concerned,
>> we see that conventional load is used, but we can only guess
>> if the insert into table TEST is in serial mode or not,
>> am I missing something?
>>
>>
>> Regards
>> Dimitre
>>
>>
>> On Wed, Nov 2, 2011 at 7:58 AM, Greg Rahn<greg@xxxxxxxxxxxxxxxxxx>  wrote:
>>> The reason you see the NO_PARALLEL in the outline because you have not
>>> calibrated your IO for auto DOP to work as shown by the Notes section.
>>>   After doing so you should see the SHARED directive show up in its
>>> place.  It's still using PX and you can see this from looking at
>>> v$pq_tqstat from the same session you ran the query.  One can also
>>> look at the SQL Monitor Report as well.
>>>
>>> Using scott.emp might be a poor example because there is only 1
>>> granule for the table scan so only 1 PX server gets work.
>>>
>>>
>

--
//www.freelists.org/webpage/oracle-l

Other related posts:

  1. Home
  2. oracle-l
  3. Archive
  4. 11-2011