Re: slow query - only uses indices

On Dec 24, 2007 7:46 PM, Marc <pgsql-general@mbreslow.net> wrote:
> Hey Folks,
>
> This query is running really slowly.  Sometimes much slower then others.  I
> have a feeling that there may be contention on one of the indices it is
> using.  In the explain plan, it looks like it estimates 2 rows but actually
> finds 228 rows?  Is that really bad?
>

You should do an ANALYZE if actual rows are significantly different.
And they are.

>
>
> Query and explain plan are below.  Seems like it spend the most time doing
> Index Scan using i_tablea_atextfield on tablea ru  (cost=0.00..2265.28
> rows=2 width=12) (actual time=0.624..881.313 rows=228 loops=1)
>
> Any suggestions?
>
> SELECT z.atextfield,
>        z.btextfield,
>        z.abigintfield,
>        p.achar255field, p.ptextfield,
>        z.achar1field,     u.aboolfield,
>        z.textfield1,
>        z.achar8field,
>        z.achar16field
> FROM tablea ru
> INNER JOIN tableb u ON ( u.atextfield = ru.anothertextfield )
> INNER JOIN tablec z ON u.atextfield = z.atextfield
> INNER JOIN tabled p ON p.id = z.pid
> LEFT JOIN tablee m ON u.atextfield = m.atextfield  AND m.boolcol5
>  WHERE ru.atextfield = 'thelookupval'
>   AND u.boolcol1 IS TRUE
>   AND u.boolcol2 IS FALSE
>   AND ru.achar1field <> 'N'
>   AND ru.boolcol3 IS FALSE
>   AND NOT EXISTS ( SELECT 'x' FROM tablea fru WHERE fru.atextfield =
> ru.anothertextfield AND fru.boolcol3 IS TRUE )
>   AND EXISTS ( SELECT 'x' FROM tablef s WHERE s.atextfield = ru.atextfield
> AND s.boolcol4 IS TRUE )
>   ORDER by ru.anothertextfield asc
>
>
>
> Sort  (cost=2341.96..2341.97 rows=2 width=146) (actual
> time=1118.810..1119.098 rows=228 loops=1)
>   Sort Key: ru.anothertextfield
>   ->  Nested Loop Left Join  (cost=0.00..2341.95 rows=2 width=146) (actual
> time= 0.930..1117.258 rows=228 loops=1)
>         ->  Nested Loop  (cost=0.00..2313.36 rows=2 width=131) (actual
> time=0.842..914.554 rows=228 loops=1)
>               ->  Nested Loop  (cost=0.00..2296.65 rows=2 width=93) (actual
> time= 0.765..901.916 rows=228 loops=1)
>                     ->  Nested Loop  (cost=0.00..2281.98 rows=2 width=72)
> (actual time=0.690..893.648 rows=228 loops=1)
>                           ->  Index Scan using i_tablea_atextfield on tablea
> ru  (cost= 0.00..2265.28 rows=2 width=12) (actual time=0.624..881.313
> rows=228 loops=1)
>                                 Index Cond: (atextfield =
> 'thelookupval'::text)
>                                 Filter: ((achar1field <> 'N'::bpchar) AND
> (boolcol3 IS FALSE) AND (NOT (subplan)) AND (subplan))
>                                 SubPlan
>                                   ->  Index Scan using tablef_pkey on tablef
> s  (cost=0.00..8.34 rows=1 width=0) (actual time=0.016..0.016 rows=1
> loops=228)
>                                         Index Cond: (atextfield = $1)
>                                         Filter: (boolcol4 IS TRUE)
>                                   ->  Bitmap Heap Scan on tablea fru
> (cost=4.61..90.41 rows=1 width=0) (actual time=3.590..3.590 rows=0
> loops=243)
>                                         Recheck Cond: (atextfield = $0)
>                                         Filter: (boolcol3 IS TRUE)
>                                         ->  Bitmap Index Scan on
> i_tablea_atextfield  (cost= 0.00..4.61 rows=22 width=0) (actual
> time=0.044..0.044 rows=17 loops=243)
>                                               Index Cond: (atextfield = $0)
>                           ->  Index Scan using tablec_pkey on tablec z
> (cost= 0.00..8.34 rows=1 width=60) (actual time=0.047..0.049 rows=1
> loops=228)
>                                 Index Cond: (z.atextfield =
> ru.anothertextfield)
>                     ->  Index Scan using tabled_pkey on tabled p  (cost=
> 0.00..7.32 rows=1 width=29) (actual time=0.030..0.031 rows=1 loops=228)
>                           Index Cond: (p.id = z.pid)
>               ->  Index Scan using tableb_pkey on tableb u  (cost=
> 0.00..8.34 rows=1 width=38) (actual time=0.049..0.051 rows=1 loops=228)
>                     Index Cond: (u.atextfield = ru.anothertextfield)
>                     Filter: ((boolcol1 IS TRUE) AND (boolcol2" IS FALSE))"
>         ->  Index Scan using tablee_atextfield_idx on tablee m
> (cost=0.00..14.28 rows=1 width=39) (actual time=0.883..0.883 rows=0
> loops=228)
>               Index Cond: (u.atextfield = m.atextfield)
>
>
>
>


I see a lot of Nested Nested Nested Loops on text fields which means a
lot of string comparisons.
Try using HashJoin here.

ANALYZE your database and explain again. Or try forcing a hashjoin if you still
dont see a difference.


--Imad