Re: Why query plan is different? - Mailing list pgsql-performance

On 11.10.2016 03:47, Pavel Stehule wrote:

2016-10-10 23:17 GMT+02:00 Andrzej Zawadzki <zawadaa@gmail.com>:

On 10.10.2016 17:31, Andrzej Zawadzki wrote:

Hi,
Today, I noticed strange situation:

The same query run on different servers has very different plan:

Q: SELECT b.* FROM kredytob b  WHERE pesel = '22222222222'  ORDER BY b.id DESC LIMIT 1

Slow plan:

"Limit  (cost=0.43..28712.33 rows=1 width=4) (actual time=2574.041..2574.044 rows=1 loops=1)"
"  Output: id"
"  Buffers: shared hit=316132 read=110001"
"  ->  Index Scan Backward using kredytob_pkey on public.kredytob b  (cost=0.43..3244444.80 rows=113 width=4) (actual time=2574.034..2574.034 rows=1 loops=1)"
"        Output: id"
"        Filter: (b.pesel = '22222222222'::bpchar)"
"        Rows Removed by Filter: 433609"
"        Buffers: shared hit=316132 read=110001"
"Planning time: 0.414 ms"
"Execution time: 2574.139 ms"


Fast plan:
"Limit  (cost=115240.66..115240.66 rows=1 width=4) (actual time=463.275..463.276 rows=1 loops=1)"
"  Output: id"
"  Buffers: shared hit=14661 read=4576"
"  ->  Sort  (cost=115240.66..115240.94 rows=112 width=4) (actual time=463.271..463.271 rows=1 loops=1)"
"        Output: id"
"        Sort Key: b.id DESC"
"        Sort Method: top-N heapsort  Memory: 25kB"
"        Buffers: shared hit=14661 read=4576"
"        ->  Index Scan using kredytob_pesel_typkred_opclass_idx on public.kredytob b  (cost=0.43..115240.10 rows=112 width=4) (actual time=311.347..463.183 rows=5 loops=1)"
"              Output: id"
"              Index Cond: (b.pesel = '22222222222'::bpchar)"
"              Buffers: shared hit=14661 read=4576"
"Planning time: 0.383 ms"
"Execution time: 463.324 ms"

Data is almost equal - "slow" has a few more rows in table. ("Fast" is a copy from 1 am today).
Why runtime is slower?

I made another INDEX, without opclass:

CREATE INDEX kredytob_pesel_typkred_idx
  ON public.kredytob
  USING btree
  (pesel COLLATE pg_catalog."default", typkred);

after that: analyze kredytob;

And now:
"Limit  (cost=333.31..333.31 rows=1 width=4) (actual time=0.100..0.102 rows=1 loops=1)"
"  Output: id"
"  Buffers: shared hit=8"
"  ->  Sort  (cost=333.31..333.59 rows=114 width=4) (actual time=0.095..0.095 rows=1 loops=1)"
"        Output: id"
"        Sort Key: b.id DESC"
"        Sort Method: top-N heapsort  Memory: 25kB"
"        Buffers: shared hit=8"
"        ->  Index Scan using kredytob_pesel_typkred_idx on public.kredytob b  (cost=0.43..332.74 rows=114 width=4) (actual time=0.046..0.065 rows=5 loops=1)"
"              Output: id"
"              Index Cond: (b.pesel = '22222222222'::bpchar)"
"              Buffers: shared hit=8"
"Planning time: 0.438 ms"
"Execution time: 0.154 ms"

So, what is a reason that "SLOW" server doesn't like opclass index?

what is default locales?

LATIN2 - that's why I use opclass.

--
Andrzej