Обсуждение: Understanding the output of EXPLAIN ANALYSE

On Thu, Jul 23, 2009 at 6:14 PM, Thurber, Fred<Fred.Thurber@gdit.com> wrote:
> "Nested Loop  (cost=78.63..101.14 rows=1 width=256) (actual
> time=0.000..16.000 rows=2818 loops=1)"


Well it was expecting 1 row and got 2,818 so the costs are going to be
pretty wildly wrong. But since you cut out all the details it's pretty
hard to guess what happened.

--
greg
http://mit.edu/~gsstark/resume.pdf

On Thu, Jul 23, 2009 at 12:14 PM, Thurber, Fred<Fred.Thurber@gdit.com> wrote:
>
> So the cost was 101.  We put indexes on the two tables that were being
> joined and re-ran EXPLAIN ANALYZE.  The cost went way down, but the
> execution time went way up.  What am I missing?
>
> "Nested Loop  (cost=0.00..15.28 rows=1 width=256) (actual
> time=0.000..110.000 rows=2818 loops=1)"
> …
> "Total runtime: 110.000 ms"
>

run ANALYZE and rerun "explain analyze your_query"

--
Atentamente,
Jaime Casanova
Soporte y capacitación de PostgreSQL
Asesoría y desarrollo de sistemas
Guayaquil - Ecuador
Cel. +59387171157

Here is the full output:


"Nested Loop  (cost=78.63..101.14 rows=1 width=256) (actual time=0.000..16.000 rows=2818 loops=1)"
"  ->  Nested Loop  (cost=78.63..95.12 rows=1 width=148) (actual time=0.000..0.000 rows=1 loops=1)"
"        ->  Merge Join  (cost=78.63..89.24 rows=1 width=100) (actual time=0.000..0.000 rows=1 loops=1)"
"              Merge Cond: (("outer".gridid = "inner".gridid) AND ("outer".subcenterid = "inner".subcenterid) AND
("outer".centerid= "inner".centerid))" 
"              ->  Sort  (cost=43.69..45.19 rows=600 width=80) (actual time=0.000..0.000 rows=1 loops=1)"
"                    Sort Key: datasetdir.gridid, datasetdir.subcenterid, datasetdir.centerid"
"                    ->  Seq Scan on mdgriddatasetdir datasetdir  (cost=0.00..16.00 rows=600 width=80) (actual
time=0.000..0.000rows=1 loops=1)" 
"              ->  Sort  (cost=34.94..36.09 rows=460 width=26) (actual time=0.000..0.000 rows=50 loops=1)"
"                    Sort Key: reg.gridid, reg.subcenterid, reg.centerid"
"                    ->  Seq Scan on mdgridregistrtions reg  (cost=0.00..14.60 rows=460 width=26) (actual
time=0.000..0.000rows=50 loops=1)" 
"        ->  Index Scan using mdgridaois_pkey on mdgridaois aois  (cost=0.00..5.87 rows=1 width=56) (actual
time=0.000..0.000rows=1 loops=1)" 
"              Index Cond: (aois.aoiid = "outer".aoiid)"
"              Filter: ((northlat > 18.0) AND (southlat < 30.0) AND (round((((distance(((eastlon - westlon))::double
precision)+ distance(((116.0 - eastlon))::double precision)) + distance(12::double precision)) + distance(((westlon -
128.0))::doublepreci (..)" 
"  ->  Index Scan using bobs_grid_index on g580000266058 grids  (cost=0.00..6.01 rows=1 width=116) (actual
time=0.000..0.000rows=2818 loops=1)" 
"        Index Cond: (("outer".datasetid = grids.datasetid) AND (grids.validtime >= 1197867600) AND (grids.validtime <=
1197954000))"
"Total runtime: 16.000 ms"


Now we added a couple of indexes, and it seemingly slowed down, but the estimated cost was lower:

"Nested Loop  (cost=0.00..15.28 rows=1 width=256) (actual time=0.000..110.000 rows=2818 loops=1)"
"  ->  Nested Loop  (cost=0.00..9.40 rows=1 width=208) (actual time=0.000..32.000 rows=2818 loops=1)"
"        ->  Nested Loop  (cost=0.00..3.38 rows=1 width=100) (actual time=0.000..0.000 rows=1 loops=1)"
"              Join Filter: (("outer".centerid = "inner".centerid) AND ("outer".subcenterid = "inner".subcenterid) AND
("outer".gridid= "inner".gridid))" 
"              ->  Seq Scan on mdgriddatasetdir datasetdir  (cost=0.00..1.01 rows=1 width=80) (actual time=0.000..0.000
rows=1loops=1)" 
"              ->  Seq Scan on mdgridregistrtions reg  (cost=0.00..1.50 rows=50 width=26) (actual time=0.000..0.000
rows=50loops=1)" 
"        ->  Index Scan using bobs_grid_index on g580000266058 grids  (cost=0.00..6.01 rows=1 width=116) (actual
time=0.000..0.000rows=2818 loops=1)" 
"              Index Cond: (("outer".datasetid = grids.datasetid) AND (grids.validtime >= 1197867600) AND
(grids.validtime<= 1197954000))" 
"  ->  Index Scan using mdgridaois_pkey on mdgridaois aois  (cost=0.00..5.87 rows=1 width=56) (actual time=0.022..0.028
rows=1loops=2818)" 
"        Index Cond: (aois.aoiid = "outer".aoiid)"
"        Filter: ((northlat > 18.0) AND (southlat < 30.0) AND (round((((distance(((eastlon - westlon))::double
precision)+ distance(((116.0 - eastlon))::double precision)) + distance(12::double precision)) + distance(((westlon -
128.0))::doubleprecision)) (..)" 

"Total runtime: 110.000 ms"

Thurber, Fred schrieb:
>
> Now we added a couple of indexes, and it seemingly slowed down, but the estimated cost was lower:

which indexes did you add? As far as I can see, there is only bobs_grid_index used in both
plans. And just to mention it - did you run VACUUM and / or ANAYLZE?

Cheers

Andy