Thread: inputs into query planner costing

On 3/31/2015 10:31 AM, Mike Roest wrote:
> Hi there,
>     I'm having an issue with query performance between 2 different pgsql
> environments.
>
> Ther first is our current production postgres server with is running
> 9.3.5 on Centos 5 x64.  The second system is Amazon's RDS postgres as a
> service.  On our local DB server we have a query that executes in a
> reasonable amount of time (600 msec).  On RDS the query will run for
> more then 10 minutes on a similarly CPU specced systems.  I've been
> working through with Amazon support and I'm looking for more suggestions
> on where to look (both for me and to direct Amazon).  The RDS system
> does use a network filesystem while our production server is a local
> RAID10 array, I can see that effecting the actual performance of the
> query but not the query planner costing (unless there's an input to
> query planner costing that I can't find)
>
> The Query plan costs generated by the 2 systems are vastly different,
> while the plans themselves are basically identical other then the
> materialization that RDS is doing (if I disable the materialization then
> they are almost the same other then a seq scan/heap scan on one small
> <2000 row table).  All the tables in the query have been analyzed on
> each server without any impact
>
> Current Production
> Explain:
> http://explain.depesz.com/s/Tkyc
> Explain Analyze
> http://explain.depesz.com/s/UnQt
>
> RDS: (with enable_material=off)
> http://explain.depesz.com/s/vDiV
>
> (with enable_material=on)
> http://explain.depesz.com/s/HUjx
>
> I have validated that all the query planning configuration variables on
> this page
> http://www.postgresql.org/docs/9.3/static/runtime-config-query.html are
> the same between the 2 environments.  If I modify the local production
> system values for things like random_page_cost and seq_page_cost to
> absurd values like 60000 I can get it to generate a similar planner
> cost.  Similarly if I lower the RDS values to absurdly low values like
> .00000001 I can get it to generate a similarly costed plan (while still
> performing horridly).
>
> I've reached the end of things I can think about (I'm also working on
> rewriting the query but it's a generated query out of a infrastructure
> component so it's not a simple change).
>
> Just looking for any ideas on additional things to look into.
>
> The query is available here:
> https://www.dropbox.com/s/m31ct6k0mod0576/simplifiedquery.sql?dl=0
>
> --
> Data's inconvienient when people have opinions.

Seems like there is no useable index on table timesheet.  It always
seems to table scan all 99K rows, several times.

I'll bet on RDS that table scan is super slow.

The RDS plans seem to be just explain?  I assume its too slow to run an
explain analyze on?  Would be neat to see explain analyze from RDS.
(any way to add a little extra where magic to cut the rows down to a
useable, but still slow, sample?)


On this one:
http://explain.depesz.com/s/UnQt

Line 11 table scans 99K rows, then all those rows are carried up the
chain (lines 10, 9, 8, 7 and 6).  Any way to reduce the row count
earlier?   Line 5 finally seems to filter out 94K rows.

Would be neat to see if these are buffered reads or are actually hitting
disk too. (something like  EXPLAIN (ANALYZE, BUFFERS) select...)

-Andy

Mike Roest <mike.roest@replicon.com> writes:
>    I'm having an issue with query performance between 2 different pgsql
> environments.

> Ther first is our current production postgres server with is running 9.3.5
> on Centos 5 x64.  The second system is Amazon's RDS postgres as a service.
> On our local DB server we have a query that executes in a reasonable amount
> of time (600 msec).  On RDS the query will run for more then 10 minutes on
> a similarly CPU specced systems.  I've been working through with Amazon
> support and I'm looking for more suggestions on where to look (both for me
> and to direct Amazon).  The RDS system does use a network filesystem while
> our production server is a local RAID10 array, I can see that effecting the
> actual performance of the query but not the query planner costing (unless
> there's an input to query planner costing that I can't find)

> The Query plan costs generated by the 2 systems are vastly different, while
> the plans themselves are basically identical other then the materialization
> that RDS is doing (if I disable the materialization then they are almost
> the same other then a seq scan/heap scan on one small <2000 row table).
> All the tables in the query have been analyzed on each server without any
> impact

Last I checked, there was not any magic pixie dust in the planner ;-).
Your results have to be explained by one or more of these things:

1. Not same version of Postgres between the two systems.

2. Not same planner parameter settings.

3. Different physical table sizes.

4. Different ANALYZE statistics.

As for #1, I have no idea whether Amazon RDS runs a purely stock Postgres
release or has some custom modifications of their own, but it'd be worth
asking about that.

As for #2, you say you checked that, but I'm dubious.  In particular this
discrepancy:

Index Scan using uix2pdas_userpolicy on policydataaccessscope policydataaccessscope31 (cost=0.28..8.30 rows=1 width=16)

Index Scan using uix2pdas_userpolicy on policydataaccessscope policydataaccessscope31 (cost=0.28..4.30 rows=1 width=16)

is hard to explain unless the second system is using a smaller
random_page_cost than the first.  Maybe somebody used ALTER ROLE SET
or ALTER DATABASE SET to adjust parameters in a way that only affects
some roles/databases?

I suspect that the large differences in some of the seqscan costs might be
explainable by #3, ie those tables are bloated with lots of empty space on
one system but not the other.  Comparing pg_relation_size() would be the
way to find out.

I mention #4 for completeness.  ANALYZE uses random sampling, so it's
expectable that the data distribution stats would be a bit different on
the two systems, but large differences that persist across multiple
ANALYZE attempts are unlikely.  (Although ... you do have the
same default_statistics_target on both systems, no?  Table-specific
statistics targets could be a gotcha as well.)

            regards, tom lane