Received: from malur.postgresql.org ([217.196.149.56]) by arkaria.postgresql.org with esmtp (Exim 4.84_2) (envelope-from ) id 1cDH8U-0004ik-8V for pgsql-performance@arkaria.postgresql.org; Sat, 03 Dec 2016 20:42:18 +0000 Received: from localhost ([127.0.0.1] helo=postgresql.org) by malur.postgresql.org with smtp (Exim 4.84_2) (envelope-from ) id 1cDH8S-0001hI-Vb for pgsql-performance@arkaria.postgresql.org; Sat, 03 Dec 2016 20:42:17 +0000 Received: from makus.postgresql.org ([2001:4800:1501:1::229]) by malur.postgresql.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_CBC_SHA384:256) (Exim 4.84_2) (envelope-from ) id 1cDH8S-0001hA-51 for pgsql-performance@postgresql.org; Sat, 03 Dec 2016 20:42:16 +0000 Received: from pecan2-mail.exetel.com.au ([220.233.0.71] helo=pecan2.exetel.com.au) by makus.postgresql.org with esmtp (Exim 4.84_2) (envelope-from ) id 1cDH8M-0003hP-H7 for pgsql-performance@postgresql.org; Sat, 03 Dec 2016 20:42:14 +0000 Received: from 131.95.233.220.static.exetel.com.au ([220.233.95.131] helo=gandalf.measday.com) by pecan2.exetel.com.au with esmtp (Exim 4.84) (envelope-from ) id 1cDH8G-0000vR-Pr; Sun, 04 Dec 2016 07:42:04 +1100 Received: from localhost (localhost [127.0.0.1]) by gandalf.measday.com (Postfix) with ESMTP id 95D0B80ABC1B; Sun, 4 Dec 2016 07:42:04 +1100 (AEDT) X-Virus-Scanned: amavisd-new at measday.com Received: from gandalf.measday.com ([IPv6:::1]) by localhost (gandalf.measday.com [IPv6:::1]) (amavisd-new, port 10024) with ESMTP id CybbpEbjzNG3; Sun, 4 Dec 2016 07:42:03 +1100 (AEDT) Received: from [10.26.68.210] (android-d33517d905c49d4e.measday.com [10.26.68.210]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by gandalf.measday.com (Postfix) with ESMTPSA id E1376809F2F1; Sun, 4 Dec 2016 07:42:03 +1100 (AEDT) Subject: Re: Substantial different index use between 9.5 and 9.6 To: Daniel Blanch Bataller References: <21406.1480636092@sss.pgh.pa.us> Cc: Tom Lane , pgsql-performance@postgresql.org From: Bill Measday Message-ID: <6586dbf1-3c1e-e7b1-bdae-c2a3d2914f49@measday.com> Date: Sun, 4 Dec 2016 07:42:02 +1100 User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:45.0) Gecko/20100101 Thunderbird/45.5.0 MIME-Version: 1.0 In-Reply-To: Content-Type: multipart/alternative; boundary="------------173B9B9D4CF61659B25DAD41" X-Pg-Spam-Score: 1.0 (+) List-Archive: List-Help: List-ID: List-Owner: List-Post: List-Subscribe: List-Unsubscribe: X-Mailing-List: pgsql-performance Precedence: bulk Sender: pgsql-performance-owner@postgresql.org This is a multi-part message in MIME format. --------------173B9B9D4CF61659B25DAD41 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 8bit Seems to be a replicable issue in PostGis - ticket raised at their end, so I'll wait for a resolution of the root cause. Thanks for your help/thoughts. Rgds Bill On 3/12/2016 2:41 AM, Daniel Blanch Bataller wrote: > ANALYZE takes samples at random, so statistics might be different even > with same postgresql version: > > https://www.postgresql.org/docs/current/static/sql-analyze.html > > For large tables, ANALYZE takes a random sample of the table > contents, rather than examining every row. This allows even very > large tables to be analyzed in a small amount of time. Note, > however, that the statistics are only approximate, and will change > slightly each time ANALYZE is run, even if the actual table > contents did not change. This might result in small changes in the > planner's estimated costs shown by EXPLAIN > . > In rare situations, this non-determinism will cause the planner's > choices of query plans to change after ANALYZE is run. To avoid > this, raise the amount of statistics collected by ANALYZE, as > described below. > > > Though, having that round (x 1000) difference, my bet is that you have > different statistics target whether on database, table or columns, see: > > The extent of analysis can be controlled by adjusting the > default_statistics_target > configuration > variable, or on a column-by-column basis by setting the per-column > statistics target with ALTER TABLE ... ALTER COLUMN ... SET > STATISTICS (see ALTER TABLE > ). > The target value sets the maximum number of entries in the > most-common-value list and the maximum number of bins in the > histogram. The default target value is 100, but this can be > adjusted up or down to trade off accuracy of planner estimates > against the time taken for ANALYZE and the amount of space > occupied in pg_statistic. In particular, setting the statistics > target to zero disables collection of statistics for that column. > It might be useful to do that for columns that are never used as > part of the WHERE, GROUP BY, or ORDER BY clauses of queries, since > the planner will have no use for statistics on such columns. > > > Here is some help on how to see statistics per column: > > http://stackoverflow.com/questions/15034622/check-statistics-targets-in-postgresql > > Check if this is the case. > > > > > > >> El 2 dic 2016, a las 1:26, Bill Measday > > escribió: >> >> Thanks Tom. >> >> First, this wasn't a migration but new db loaded from scratch (if >> that matters). >> >> As per the end of the original post "I have vacuum analysed both >> tables". I assume this is what you meant? >> >> My gut feel was that it isn't a postgis issue since the third example >> I gave uses the index, but I will take it up with them too. >> >> Rgds >> >> >> Bill >> >> On 2/12/2016 10:48 AM, Tom Lane wrote: >>> Bill Measday > writes: >>>> Substantial different index use between 9.5 and 9.6 >>> Maybe you missed an ANALYZE after migrating? The plan difference >>> seems to be due to a vast difference in rowcount estimate for the >>> m_elevations condition: >>> >>>> -> Bitmap Heap Scan on m_elevations e >>>> (cost=282802.21..37401439.43 rows=3512160 width=8) >>>> -> Seq Scan on m_elevations e >>>> (cost=10000000000.00..13296950520.12 rows=3512159563 width=8) >>> If you don't know where that factor-of-1000 came from, maybe take >>> it up with the postgis folk. It'd mostly be coming out of their >>> selectivity estimation routines. >>> >>> regards, tom lane >> >> >> >> -- >> Sent via pgsql-performance mailing list >> (pgsql-performance@postgresql.org >> ) >> To make changes to your subscription: >> http://www.postgresql.org/mailpref/pgsql-performance > --------------173B9B9D4CF61659B25DAD41 Content-Type: text/html; charset=utf-8 Content-Transfer-Encoding: 8bit

Seems to be a replicable issue in PostGis - ticket raised at their end, so I'll wait for a resolution of the root cause.


Thanks for your help/thoughts.


Rgds


Bill


On 3/12/2016 2:41 AM, Daniel Blanch Bataller wrote:
ANALYZE takes samples at random, so statistics might be different even with same postgresql version:


For large tables, ANALYZE takes a random sample of the table contents, rather than examining every row. This allows even very large tables to be analyzed in a small amount of time. Note, however, that the statistics are only approximate, and will change slightly each time ANALYZE is run, even if the actual table contents did not change. This might result in small changes in the planner's estimated costs shown by EXPLAIN. In rare situations, this non-determinism will cause the planner's choices of query plans to change after ANALYZE is run. To avoid this, raise the amount of statistics collected by ANALYZE, as described below.

Though, having that round (x 1000) difference, my bet is that you have different statistics target whether on database, table or columns, see:

The extent of analysis can be controlled by adjusting the default_statistics_target configuration variable, or on a column-by-column basis by setting the per-column statistics target with ALTER TABLE ... ALTER COLUMN ... SET STATISTICS (see ALTER TABLE). The target value sets the maximum number of entries in the most-common-value list and the maximum number of bins in the histogram. The default target value is 100, but this can be adjusted up or down to trade off accuracy of planner estimates against the time taken for ANALYZE and the amount of space occupied in pg_statistic. In particular, setting the statistics target to zero disables collection of statistics for that column. It might be useful to do that for columns that are never used as part of the WHERE, GROUP BY, or ORDER BY clauses of queries, since the planner will have no use for statistics on such columns.

Here is some help on how to see statistics per column:


Check if this is the case.





El 2 dic 2016, a las 1:26, Bill Measday <bill@measday.com> escribió:

Thanks Tom.

First, this wasn't a migration but new db loaded from scratch (if that matters).

As per the end of the original post "I have vacuum analysed both tables".  I assume this is what you meant?

My gut feel was that it isn't a postgis issue since the third example I gave uses the index, but I will take it up with them too.

Rgds


Bill

On 2/12/2016 10:48 AM, Tom Lane wrote:
Bill Measday <bill@measday.com> writes:
Substantial different index use between 9.5 and 9.6
Maybe you missed an ANALYZE after migrating?  The plan difference
seems to be due to a vast difference in rowcount estimate for the
m_elevations condition:

      ->  Bitmap Heap Scan on m_elevations e
(cost=282802.21..37401439.43 rows=3512160 width=8)
      ->  Seq Scan on m_elevations e
(cost=10000000000.00..13296950520.12 rows=3512159563 width=8)
If you don't know where that factor-of-1000 came from, maybe take
it up with the postgis folk.  It'd mostly be coming out of their
selectivity estimation routines.

regards, tom lane



--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance


--------------173B9B9D4CF61659B25DAD41--