Why this query does't use index-only scan in PostgreSQL? - postgresql

I have a table with 28 columns and 7M records without primary key.
CREATE TABLE records (
direction smallint,
exporters_id integer,
time_stamp integer
...
)
I create index on this table and vacuum table after that (autovacuum is on)
CREATE INDEX exporter_dir_time_only_index ON sacopre_records
USING btree (exporters_id, direction, time_stamp);
and i want execute this query
SELECT count(exporters_id) FROM records WHERE exporters_id = 50
The table has 6982224 records with exporters_id = 50. I expected this query use index-only scan to get results but it used sequential scan.
This is "EXPLAIN ANALYZE" output:
Aggregate (cost=204562.25..204562.26 rows=1 width=4) (actual time=1521.862..1521.862 rows=1 loops=1)
-> Seq Scan on sacopre_records (cost=0.00..187106.88 rows=6982149 width=4) (actual time=0.885..1216.211 rows=6982224 loops=1)
Filter: (exporters_id = 50)
Rows Removed by Filter: 2663
Total runtime: 1521.886 ms
but when I change the exporters_id to another id, query use index-only scan
Aggregate (cost=46.05..46.06 rows=1 width=4) (actual time=0.321..0.321 rows=1 loops=1)
-> Index Only Scan using exporter_dir_time_only_index on sacopre_records (cost=0.43..42.85 rows=1281 width=4) (actual time=0.313..0.315 rows=4 loops=1)
Index Cond: (exporters_id = 47)
Heap Fetches: 0
Total runtime: 0.358 ms
Where is the problem?

The explain is telling you the reason. Look closer.
Aggregate (cost=204562.25..204562.26 rows=1 width=4) (actual time=1521.862..1521.862 rows=1 loops=1)
-> Seq Scan on sacopre_records (cost=0.00..187106.88 rows=6982149 width=4) (actual time=0.885..1216.211 rows=6982224 loops=1)
Filter: (exporters_id = 50)
Rows Removed by Filter: 2663
Total runtime: 1521.886 ms
Your filter is removing only 2663 rows out of the total amount of 6982149 rows in the table, hence doing a sequential scan should really be faster than using an index, as the disk head should pass through 6982149 - 2663 = 6979486 records anyway. The disk head is starting to read the entire table sequentially and on the way is removing that tiny fraction (0.000004 %) that does not match your criteria. While in the index scan case it should jump from the index file(s) and get back to the data file(s) 6979486 times, which for sure should be slower than these 1.5 seconds you are getting now!

Related

postgresql search is slow on type text[] column

I have product_details table with 30+ Million records. product attributes text type data is stored into column Value1.
Front end(web) users search for product details and it will be queried on column Value1.
create table product_details(
key serial primary key ,
product_key int,
attribute_key int ,
Value1 text[],
Value2 int[],
status text);
I created gin index on column Value1 to improve search query performance.
query execution improved a lot for many queries.
Tables and indexes are here
Below is one of query used by application for search.
select p.key from (select x.product_key,
x.value1,
x.attribute_key,
x.status
from product_details x
where value1 IS NOT NULL
) as pr_d
join attribute_type at on at.key = pr_d.attribute_key
join product p on p.key = pr_d.product_key
where value1_search(pr_d.value1) ilike '%B s%'
and at.type = 'text'
and at.status = 'active'
and pr_d.status = 'active'
and 1 = 1
and p.product_type_key=1
and 1 = 1
group by p.key
query is executed in 2 or 3 secs if we search %B % or any single or two char words and below is query plan
Group (cost=180302.82..180302.83 rows=1 width=4) (actual time=49.006..49.021 rows=65 loops=1)
Group Key: p.key
-> Sort (cost=180302.82..180302.83 rows=1 width=4) (actual time=49.005..49.009 rows=69 loops=1)
Sort Key: p.key
Sort Method: quicksort Memory: 28kB
-> Nested Loop (cost=0.99..180302.81 rows=1 width=4) (actual time=3.491..48.965 rows=69 loops=1)
Join Filter: (x.attribute_key = at.key)
Rows Removed by Join Filter: 10051
-> Nested Loop (cost=0.99..180270.15 rows=1 width=8) (actual time=3.396..45.211 rows=69 loops=1)
-> Index Scan using products_product_type_key_status on product p (cost=0.43..4420.58 rows=1413 width=4) (actual time=0.024..1.473 rows=1630 loops=1)
Index Cond: (product_type_key = 1)
-> Index Scan using product_details_product_attribute_key_status on product_details x (cost=0.56..124.44 rows=1 width=8) (actual time=0.026..0.027 rows=0 loops=1630)
Index Cond: ((product_key = p.key) AND (status = 'active'))
Filter: ((value1 IS NOT NULL) AND (value1_search(value1) ~~* '%B %'::text))
Rows Removed by Filter: 14
-> Seq Scan on attribute_type at (cost=0.00..29.35 rows=265 width=4) (actual time=0.002..0.043 rows=147 loops=69)
Filter: ((value_type = 'text') AND (status = 'active'))
Rows Removed by Filter: 115
Planning Time: 0.732 ms
Execution Time: 49.089 ms
But if i search for %B s%, query took 75 secs and below is query plan (second time query execution took 63 sec)
In below query plan, DB engine didn't consider index for scan as in above query plan indexes were used. Not sure why ?
Group (cost=8057.69..8057.70 rows=1 width=4) (actual time=62138.730..62138.737 rows=12 loops=1)
Group Key: p.key
-> Sort (cost=8057.69..8057.70 rows=1 width=4) (actual time=62138.728..62138.732 rows=14 loops=1)
Sort Key: p.key
Sort Method: quicksort Memory: 25kB
-> Nested Loop (cost=389.58..8057.68 rows=1 width=4) (actual time=2592.685..62138.710 rows=14 loops=1)
-> Hash Join (cost=389.15..4971.85 rows=368 width=4) (actual time=298.280..62129.956 rows=831 loops=1)
Hash Cond: (x.attribute_type = at.key)
-> Bitmap Heap Scan on product_details x (cost=356.48..4937.39 rows=681 width=8) (actual time=298.117..62128.452 rows=831 loops=1)
Recheck Cond: (value1_search(value1) ~~* '%B s%'::text)
Rows Removed by Index Recheck: 26168889
Filter: ((value1 IS NOT NULL) AND (status = 'active'))
Rows Removed by Filter: 22
Heap Blocks: exact=490 lossy=527123
-> Bitmap Index Scan on product_details_value1_gin (cost=0.00..356.31 rows=1109 width=0) (actual time=251.596..251.596 rows=2846970 loops=1)
Index Cond: (value1_search(value1) ~~* '%B s%'::text)
-> Hash (cost=29.35..29.35 rows=265 width=4) (actual time=0.152..0.153 rows=269 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 18kB
-> Seq Scan on attribute_type at (cost=0.00..29.35 rows=265 width=4) (actual time=0.010..0.122 rows=269 loops=1)
Filter: ((value_type = 'text') AND (status = 'active'))
Rows Removed by Filter: 221
-> Index Scan using product_pkey on product p (cost=0.43..8.39 rows=1 width=4) (actual time=0.009..0.009 rows=0 loops=831)
Index Cond: (key = x.product_key)
Filter: (product_type_key = 1)
Rows Removed by Filter: 1
Planning Time: 0.668 ms
Execution Time: 62138.794 ms
Any suggestions pls to improve query for search %B s%
thanks
ilike '%B %' has no usable trigrams in it. The planner knows this, and punishes the pg_trgm index plan so much that the planner then goes with an entirely different plan instead.
But ilike '%B s%' does have one usable trigram in it, ' s'. It turns out that this trigram sucks because it is extremely common in the searched data, but the planner currently has no way to accurately estimate how much it sucks.
Even worse, this large number matches means your full bitmap can't fit in work_mem so it goes lossy. Then it needs to recheck all the tuples in any page which contains even one tuple that has the ' s' trigram in it, which looks like it is most of the pages in your table.
The first thing to do is to increase your work_mem to the point you stop getting lossy blocks. If most of your time is spent in the CPU applying the recheck condition, this should help tremendously. If most of your time is spent reading the product_details from disk (so that the recheck has the data it needs to run) then it won't help much. If you had done EXPLAIN (ANALYZE, BUFFERS) with track_io_timing turned on, then we would already know which is which.
Another thing you could do is have the application inspect the search parameter, and if it looks like two letters (with or without a space between), then forcibly disable that index usage, or just throw an error if there is no good reason to do that type of search. For example, changing the part of the query to look like this will disable the index:
where value1_search(pr_d.value1)||'' ilike '%B s%'
Another thing would be to rethink your data representation. '%B s%' is a peculiar thing to search for. Why would anyone search for that? Does it have some special meaning within the context of your data, which is not obvious to the outside observer? Maybe you could represent it in a different way that gets along better with pg_trgm.
Finally, you could try to improve the planning for GIN indexes generally by explicitly estimating how many tuples are going to fail recheck (due to inherent lossiness of the index, not due to overrunning work_mem). This would be a major undertaking, and you would be unlikely to see it in production for at least a couple years, if ever.

PostgreSQL slow order

I have table (over 100 millions records) on PostgreSQL 13.1
CREATE TABLE report
(
id serial primary key,
license_plate_id integer,
datetime timestamp
);
Indexes (for test I create both of them):
create index report_lp_datetime_index on report (license_plate_id, datetime);
create index report_lp_datetime_desc_index on report (license_plate_id desc, datetime desc);
So, my question is why query like
select * from report r
where r.license_plate_id in (1,2,4,5,6,7,8,10,15,22,34,75)
order by datetime desc
limit 100
Is very slow (~10sec). But query without order statement is fast (milliseconds).
Explain:
explain (analyze, buffers, format text) select * from report r
where r.license_plate_id in (1,2,4,5,6,7,8,10,15,22,34, 75,374,57123)
limit 100
Limit (cost=0.57..400.38 rows=100 width=316) (actual time=0.037..0.216 rows=100 loops=1)
Buffers: shared hit=103
-> Index Scan using report_lp_id_idx on report r (cost=0.57..44986.97 rows=11252 width=316) (actual time=0.035..0.202 rows=100 loops=1)
Index Cond: (license_plate_id = ANY ('{1,2,4,5,6,7,8,10,15,22,34,75,374,57123}'::integer[]))
Buffers: shared hit=103
Planning Time: 0.228 ms
Execution Time: 0.251 ms
explain (analyze, buffers, format text) select * from report r
where r.license_plate_id in (1,2,4,5,6,7,8,10,15,22,34,75,374,57123)
order by datetime desc
limit 100
Limit (cost=44193.63..44193.88 rows=100 width=316) (actual time=4921.030..4921.047 rows=100 loops=1)
Buffers: shared hit=11455 read=671
-> Sort (cost=44193.63..44221.76 rows=11252 width=316) (actual time=4921.028..4921.035 rows=100 loops=1)
Sort Key: datetime DESC
Sort Method: top-N heapsort Memory: 128kB
Buffers: shared hit=11455 read=671
-> Bitmap Heap Scan on report r (cost=151.18..43763.59 rows=11252 width=316) (actual time=54.422..4911.927 rows=12148 loops=1)
Recheck Cond: (license_plate_id = ANY ('{1,2,4,5,6,7,8,10,15,22,34,75,374,57123}'::integer[]))
Heap Blocks: exact=12063
Buffers: shared hit=11455 read=671
-> Bitmap Index Scan on report_lp_id_idx (cost=0.00..148.37 rows=11252 width=0) (actual time=52.631..52.632 rows=12148 loops=1)
Index Cond: (license_plate_id = ANY ('{1,2,4,5,6,7,8,10,15,22,34,75,374,57123}'::integer[]))
Buffers: shared hit=59 read=4
Planning Time: 0.427 ms
Execution Time: 4921.128 ms
You seem to have rather slow storage, if reading 671 8kB-blocks from disk takes a couple of seconds.
The way to speed this up is to reorder the table in the same way as the index, so that you can find the required rows in the same or adjacent table blocks:
CLUSTER report_lp_id_idx USING report_lp_id_idx;
Be warned that rewriting the table in this way causes downtime – the table will not be available while it is being rewritten. Moreover, PostgreSQL does not maintain the table order, so subsequent data modifications will cause performance to gradually deteriorate, so that after a while you will have to run CLUSTER again.
But if you need this query to be fast no matter what, CLUSTER is the way to go.
Your two indices do exactly the same thing, so you can remove the second one, it's useless.
To optimize your query, the order of the fields inside the index must be reversed:
create index report_lp_datetime_index on report (datetime,license_plate_id);
BEGIN;
CREATE TABLE foo (d INTEGER, i INTEGER);
INSERT INTO foo SELECT random()*100000, random()*1000 FROM generate_series(1,1000000) s;
CREATE INDEX foo_d_i ON foo(d DESC,i);
COMMIT;
VACUUM ANALYZE foo;
EXPLAIN ANALYZE SELECT * FROM foo WHERE i IN (1,2,4,5,6,7,8,10,15,22,34,75) ORDER BY d DESC LIMIT 100;
Limit (cost=0.42..343.92 rows=100 width=8) (actual time=0.076..9.359 rows=100 loops=1)
-> Index Only Scan Backward using foo_d_i on foo (cost=0.42..40976.43 rows=11929 width=8) (actual time=0.075..9.339 rows=100 loops=1)
Filter: (i = ANY ('{1,2,4,5,6,7,8,10,15,22,34,75}'::integer[]))
Rows Removed by Filter: 9016
Heap Fetches: 0
Planning Time: 0.339 ms
Execution Time: 9.387 ms
Note the index is not used to optimize the WHERE clause. It is used here as a compact and fast way to store references to the rows ordered by date DESC, so the ORDER BY can do an index-only scan and avoid sorting. By adding column id to the index, an index-only scan can be performed to test the condition on id, without hitting the table for every row. Since there is a low LIMIT value it does not need to scan the whole index, it only scans it in date DESC order until it finds enough rows satisfying the WHERE condition to return the result.
It will be faster if you create the index in date DESC order, this could be useful if you use ORDER BY date DESC + LIMIT in other queries too.
You forget that OP's table has a third column, and he is using SELECT *. So that wouldn't be an index-only scan.
Easy to work around. The optimum way to do this query would be an index-only scan to filter on WHERE conditions, then LIMIT, then hit the table to get the rows. For some reason if "select *" is used postgres takes the id column from the table instead of taking it from the index, which results in lots of unnecessary heap fetches for rows whose id is rejected by the WHERE condition.
Easy to work around, by doing it manually. I've also added another bogus column to make sure the SELECT * hits the table.
EXPLAIN (ANALYZE,buffers) SELECT * FROM foo
JOIN (SELECT d,i FROM foo WHERE i IN (1,2,4,5,6,7,8,10,15,22,34,75) ORDER BY d DESC LIMIT 100) f USING (d,i)
ORDER BY d DESC LIMIT 100;
Limit (cost=0.85..1281.94 rows=1 width=17) (actual time=0.052..3.618 rows=100 loops=1)
Buffers: shared hit=453
-> Nested Loop (cost=0.85..1281.94 rows=1 width=17) (actual time=0.050..3.594 rows=100 loops=1)
Buffers: shared hit=453
-> Limit (cost=0.42..435.44 rows=100 width=8) (actual time=0.037..2.953 rows=100 loops=1)
Buffers: shared hit=53
-> Index Only Scan using foo_d_i on foo foo_1 (cost=0.42..51936.43 rows=11939 width=8) (actual time=0.037..2.935 rows=100 loops=1)
Filter: (i = ANY ('{1,2,4,5,6,7,8,10,15,22,34,75}'::integer[]))
Rows Removed by Filter: 9010
Heap Fetches: 0
Buffers: shared hit=53
-> Index Scan using foo_d_i on foo (cost=0.42..8.45 rows=1 width=17) (actual time=0.005..0.005 rows=1 loops=100)
Index Cond: ((d = foo_1.d) AND (i = foo_1.i))
Buffers: shared hit=400
Execution Time: 3.663 ms
Another option is to just add the primary key to the date,license_plate index.
SELECT * FROM foo JOIN (SELECT id FROM foo WHERE i IN (1,2,4,5,6,7,8,10,15,22,34,75) ORDER BY d DESC LIMIT 100) f USING (id) ORDER BY d DESC LIMIT 100;
Limit (cost=1357.98..1358.23 rows=100 width=17) (actual time=3.920..3.947 rows=100 loops=1)
Buffers: shared hit=473
-> Sort (cost=1357.98..1358.23 rows=100 width=17) (actual time=3.919..3.931 rows=100 loops=1)
Sort Key: foo.d DESC
Sort Method: quicksort Memory: 32kB
Buffers: shared hit=473
-> Nested Loop (cost=0.85..1354.66 rows=100 width=17) (actual time=0.055..3.858 rows=100 loops=1)
Buffers: shared hit=473
-> Limit (cost=0.42..509.41 rows=100 width=8) (actual time=0.039..3.116 rows=100 loops=1)
Buffers: shared hit=73
-> Index Only Scan using foo_d_i_id on foo foo_1 (cost=0.42..60768.43 rows=11939 width=8) (actual time=0.039..3.093 rows=100 loops=1)
Filter: (i = ANY ('{1,2,4,5,6,7,8,10,15,22,34,75}'::integer[]))
Rows Removed by Filter: 9010
Heap Fetches: 0
Buffers: shared hit=73
-> Index Scan using foo_pkey on foo (cost=0.42..8.44 rows=1 width=17) (actual time=0.006..0.006 rows=1 loops=100)
Index Cond: (id = foo_1.id)
Buffers: shared hit=400
Execution Time: 3.972 ms
Edit
After thinking about it... since the LIMIT restricts the output to 100 rows ordered by date desc, wouldn't it be nice if we could get the 100 most recent rows for each license_plate_id, put all that into a top-n sort, and only keep the best 100 for all license_plate_ids? That would avoid reading and throwing away a lot of rows from the index. Even if that's much faster than hitting the table, it will still load up these index pages in RAM and clog up your buffers with stuff you don't actually need to keep in cache. Let's use LATERAL JOIN:
EXPLAIN (ANALYZE,BUFFERS)
SELECT * FROM foo
JOIN (SELECT d,i FROM
(VALUES (1),(2),(4),(5),(6),(7),(8),(10),(15),(22),(34),(75)) idlist
CROSS JOIN LATERAL
(SELECT d,i FROM foo WHERE i=idlist.column1 ORDER BY d DESC LIMIT 100) f2
ORDER BY d DESC LIMIT 100
) f3 USING (d,i)
ORDER BY d DESC LIMIT 100;
It's even faster: 2ms, and it uses the index on (license_plate_id,date) instead of the other way around. Also, and this is important, since each subquery in the lateral hits only the index pages that contain rows that will actually be selected, while the previous queries hit much more index pages. So you save on RAM buffers.
If you don't need the index on (date,license_plate_id) and don't want to keep a useless index, that could be interesting since this query doesn't use it. On the other hand, if you need the index on (date,license_plate_id) for something else and want to keep it, then... maybe not.
Please post results for the winning query 🔥

PostgreSQL multi-column group by not using index when selecting minimum

When selecting MIN on a column in PostgreSQL (11, 12, 13) after a GROUP BY operation on multiple columns, any index created on the grouped columns is not used: https://dbfiddle.uk/?rdbms=postgres_13&fiddle=30e0f341940f4c1fa6013677643a0baf
CREATE TABLE tags (id serial, series int, index int, page int);
CREATE INDEX ON tags (page, series, index);
INSERT INTO tags (series, index, page)
SELECT
ceil(random() * 10),
ceil(random() * 100),
ceil(random() * 1000)
FROM generate_series(1, 100000);
EXPLAIN ANALYZE
SELECT tags.page, tags.series, MIN(tags.index)
FROM tags GROUP BY tags.page, tags.series;
HashAggregate (cost=2291.00..2391.00 rows=10000 width=12) (actual time=108.968..133.153 rows=9999 loops=1)
Group Key: page, series
Batches: 1 Memory Usage: 1425kB
-> Seq Scan on tags (cost=0.00..1541.00 rows=100000 width=12) (actual time=0.015..55.240 rows=100000 loops=1)
Planning Time: 0.257 ms
Execution Time: 133.771 ms
Theoretically, the index should allow the database to seek in steps of (tags.page, tags.series) instead of performing a full scan. This would result in 10,000 processed rows for above dataset instead of 100,000. This link describes the method with no grouped columns.
This answer (as well as this one) suggests using DISTINCT ON with an ordering instead of GROUP BY but that produces this query plan:
Unique (cost=0.42..5680.42 rows=10000 width=12) (actual time=0.066..268.038 rows=9999 loops=1)
-> Index Only Scan using tags_page_series_index_idx on tags (cost=0.42..5180.42 rows=100000 width=12) (actual time=0.064..227.219 rows=100000 loops=1)
Heap Fetches: 100000
Planning Time: 0.426 ms
Execution Time: 268.712 ms
While the index is now being used, it still appears to be scanning the full set of rows. When using SET enable_seqscan=OFF, the GROUP BY query degrades to the same behaviour.
How can I encourage PostgreSQL to use the multi-column index?
If you can pull the set of distinct page,series from another table then you can hack it with a lateral join:
CREATE TABLE pageseries AS SELECT DISTINCT page,series FROM tags ORDER BY page,series;
EXPLAIN ANALYZE SELECT p.*, minindex FROM pageseries p CROSS JOIN LATERAL (SELECT index minindex FROM tags t WHERE t.page=p.page AND t.series=p.series ORDER BY page,series,index LIMIT 1) x;
Nested Loop (cost=0.42..8720.00 rows=10000 width=12) (actual time=0.039..56.013 rows=10000 loops=1)
-> Seq Scan on pageseries p (cost=0.00..145.00 rows=10000 width=8) (actual time=0.012..1.872 rows=10000 loops=1)
-> Limit (cost=0.42..0.84 rows=1 width=12) (actual time=0.005..0.005 rows=1 loops=10000)
-> Index Only Scan using tags_page_series_index_idx on tags t (cost=0.42..4.62 rows=10 width=12) (actual time=0.004..0.004 rows=1 loops=10000)
Index Cond: ((page = p.page) AND (series = p.series))
Heap Fetches: 0
Planning Time: 0.168 ms
Execution Time: 57.077 ms
...but it is not necessarily faster:
EXPLAIN ANALYZE SELECT tags.page, tags.series, MIN(tags.index)
FROM tags GROUP BY tags.page, tags.series;
HashAggregate (cost=2291.00..2391.00 rows=10000 width=12) (actual time=56.177..58.923 rows=10000 loops=1)
Group Key: page, series
Batches: 1 Memory Usage: 1425kB
-> Seq Scan on tags (cost=0.00..1541.00 rows=100000 width=12) (actual time=0.010..12.845 rows=100000 loops=1)
Planning Time: 0.129 ms
Execution Time: 59.644 ms
It would be massively faster IF the number of iterations in the nested loop was small, in other words if there was a low number of distinct (page,series). I'll try with series alone, since that has only 10 distinct values:
CREATE TABLE series AS SELECT DISTINCT series FROM tags;
EXPLAIN ANALYZE SELECT p.*, minindex FROM series p CROSS JOIN LATERAL (SELECT index minindex FROM tags t WHERE t.series=p.series ORDER BY series,index LIMIT 1) x;
Nested Loop (cost=0.29..886.18 rows=2550 width=8) (actual time=0.081..0.264 rows=10 loops=1)
-> Seq Scan on series p (cost=0.00..35.50 rows=2550 width=4) (actual time=0.007..0.010 rows=10 loops=1)
-> Limit (cost=0.29..0.31 rows=1 width=8) (actual time=0.024..0.024 rows=1 loops=10)
-> Index Only Scan using tags_series_index_idx on tags t (cost=0.29..211.29 rows=10000 width=8) (actual time=0.023..0.023 rows=1 loops=10)
Index Cond: (series = p.series)
Heap Fetches: 0
Planning Time: 0.198 ms
Execution Time: 0.292 ms
In this case, definitely worth it, because the query hits only 10/100000 rows. The other queries hit 10000/100000 rows, or 10% of the table, which is above the threshold where an index would really help.
Note putting the column with lower cardinality first will result in a smaller index:
CREATE INDEX ON tags (series, page, index);
select pg_relation_size( 'tags_page_series_index_idx' );
4284416
select pg_relation_size( 'tags_series_page_index_idx' );
3104768
...but it doesn't make the query any faster.
If this type of stuff is really critical, perhaps try clickhouse or dolphindb.
To support that kind of thing PostgreSQL would have to have something like an index skip scan, and it is only efficient to use that if there are few groups.
If the speed of that query is essential, you could consider using a materialized view.

Postgres uses Hash Join with Seq Scan when Inner Select Index Cond is faster

Postgres is using a much heavier Seq Scan on table tracking when an index is available. The first query was the original attempt, which uses a Seq Scan and therefore has a slow query. I attempted to force an Index Scan with an Inner Select, but postgres converted it back to effectively the same query with nearly the same runtime. I finally copied the list from the Inner Select of query two to make the third query. Finally postgres used the Index Scan, which dramatically decreased the runtime. The third query is not viable in a production environment. What will cause postgres to use the last query plan?
(vacuum was used on both tables)
Tables
tracking (worker_id, localdatetime) total records: 118664105
project_worker (id, project_id) total records: 12935
INDEX
CREATE INDEX tracking_worker_id_localdatetime_idx ON public.tracking USING btree (worker_id, localdatetime)
Queries
SELECT worker_id, localdatetime FROM tracking t JOIN project_worker pw ON t.worker_id = pw.id WHERE project_id = 68475018
Hash Join (cost=29185.80..2638162.26 rows=19294218 width=16) (actual time=16.912..18376.032 rows=177681 loops=1)
Hash Cond: (t.worker_id = pw.id)
-> Seq Scan on tracking t (cost=0.00..2297293.86 rows=118716186 width=16) (actual time=0.004..8242.891 rows=118674660 loops=1)
-> Hash (cost=29134.80..29134.80 rows=4080 width=8) (actual time=16.855..16.855 rows=2102 loops=1)
Buckets: 4096 Batches: 1 Memory Usage: 115kB
-> Seq Scan on project_worker pw (cost=0.00..29134.80 rows=4080 width=8) (actual time=0.004..16.596 rows=2102 loops=1)
Filter: (project_id = 68475018)
Rows Removed by Filter: 10833
Planning Time: 0.192 ms
Execution Time: 18382.698 ms
SELECT worker_id, localdatetime FROM tracking t WHERE worker_id IN (SELECT id FROM project_worker WHERE project_id = 68475018 LIMIT 500)
Hash Semi Join (cost=6905.32..2923969.14 rows=27733254 width=24) (actual time=19.715..20191.517 rows=20530 loops=1)
Hash Cond: (t.worker_id = project_worker.id)
-> Seq Scan on tracking t (cost=0.00..2296948.27 rows=118698327 width=24) (actual time=0.005..9184.676 rows=118657026 loops=1)
-> Hash (cost=6899.07..6899.07 rows=500 width=8) (actual time=1.103..1.103 rows=500 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 28kB
-> Limit (cost=0.00..6894.07 rows=500 width=8) (actual time=0.006..1.011 rows=500 loops=1)
-> Seq Scan on project_worker (cost=0.00..28982.65 rows=2102 width=8) (actual time=0.005..0.968 rows=500 loops=1)
Filter: (project_id = 68475018)
Rows Removed by Filter: 4493
Planning Time: 0.224 ms
Execution Time: 20192.421 ms
SELECT worker_id, localdatetime FROM tracking t WHERE worker_id IN (322016383,316007840,...,285702579)
Index Scan using tracking_worker_id_localdatetime_idx on tracking t (cost=0.57..4766798.31 rows=21877360 width=24) (actual time=0.079..29.756 rows=22112 loops=1)
" Index Cond: (worker_id = ANY ('{322016383,316007840,...,285702579}'::bigint[]))"
Planning Time: 1.162 ms
Execution Time: 30.884 ms
... is in place of the 500 id entries used in the query
Same query ran on another set of 500 id's
Index Scan using tracking_worker_id_localdatetime_idx on tracking t (cost=0.57..4776714.91 rows=21900980 width=24) (actual time=0.105..5528.109 rows=117838 loops=1)
" Index Cond: (worker_id = ANY ('{286237712,286237844,...,216724213}'::bigint[]))"
Planning Time: 2.105 ms
Execution Time: 5534.948 ms
The distribution of "worker_id" within "tracking" seems very skewed. For one thing, the number of rows in one of your instances of query 3 returns over 5 times as many rows as the other instance of it. For another, the estimated number of rows is 100 to 1000 times higher than the actual number. This can certainly lead to bad plans (although it is unlikely to be the complete picture).
What is the actual number of distinct values for worker_id within tracking: select count(distinct worker_id) from tracking? What does the planner think this value is: select n_distinct from pg_stats where tablename='tracking' and attname='worker_id'? If those values are far apart and you force the planner to use a more reasonable value with alter table tracking alter column worker_id set (n_distinct = <real value>); analyze tracking; does that change the plans?
If you want to nudge PostgreSQL towards a nested loop join, try the following:
Create an index on tracking that can be used for an index-only scan:
CREATE INDEX ON tracking (worker_id) INCLUDE (localdatetime);
Make sure that tracking is VACUUMed often, so that an index-only scan is effective.
Reduce random_page_cost and increase effective_cache_size so that the optimizer prices index scans lower (but don't use insane values).
Make sure that you have good estimates on project_worker:
ALTER TABLE project_worker ALTER project_id SET STATISTICS 1000;
ANALYZE project_worker;

Nested Loop Left Join cost too much time?

This is the query:
EXPLAIN (analyze, BUFFERS, SETTINGS)
SELECT
operation.id
FROM
operation
RIGHT JOIN(
SELECT uid, did FROM (
SELECT uid, did FROM operation where id = 993754
) t
) parts ON (operation.uid = parts.uid AND operation.did = parts.did)
and EXPLAIN info:
Nested Loop Left Join (cost=0.85..29695.77 rows=100 width=8) (actual time=13.709..13.711 rows=1 loops=1)
Buffers: shared hit=4905
-> Unique (cost=0.42..8.45 rows=1 width=16) (actual time=0.011..0.013 rows=1 loops=1)
Buffers: shared hit=5
-> Index Only Scan using oi on operation operation_1 (cost=0.42..8.44 rows=1 width=16) (actual time=0.011..0.011 rows=1 loops=1)
Index Cond: (id = 993754)
Heap Fetches: 1
Buffers: shared hit=5
-> Index Only Scan using oi on operation (cost=0.42..29686.32 rows=100 width=24) (actual time=13.695..13.696 rows=1 loops=1)
Index Cond: ((uid = operation_1.uid) AND (did = operation_1.did))
Heap Fetches: 1
Buffers: shared hit=4900
Settings: max_parallel_workers_per_gather = '4', min_parallel_index_scan_size = '0', min_parallel_table_scan_size = '0', parallel_setup_cost = '0', parallel_tuple_cost = '0', work_mem = '256MB'
Planning Time: 0.084 ms
Execution Time: 13.728 ms
Why does Nested Loop cost more and more time than sum of childs cost? What can I do for that? The Execution Time should less than 1 ms right?
update:
Nested Loop Left Join (cost=5.88..400.63 rows=101 width=8) (actual time=0.012..0.012 rows=1 loops=1)
Buffers: shared hit=8
-> Index Scan using oi on operation operation_1 (cost=0.42..8.44 rows=1 width=16) (actual time=0.005..0.005 rows=1 loops=1)
Index Cond: (id = 993754)
Buffers: shared hit=4
-> Bitmap Heap Scan on operation (cost=5.45..391.19 rows=100 width=24) (actual time=0.004..0.005 rows=1 loops=1)
Recheck Cond: ((uid = operation_1.uid) AND (did = operation_1.did))
Heap Blocks: exact=1
Buffers: shared hit=4
-> Bitmap Index Scan on ou (cost=0.00..5.42 rows=100 width=0) (actual time=0.003..0.003 rows=1 loops=1)
Index Cond: ((uid = operation_1.uid) AND (did = operation_1.did))
Buffers: shared hit=3
Settings: max_parallel_workers_per_gather = '4', min_parallel_index_scan_size = '0', min_parallel_table_scan_size = '0', parallel_setup_cost = '0', parallel_tuple_cost = '0', work_mem = '256MB'
Planning Time: 0.127 ms
Execution Time: 0.028 ms
Thanks all of you, when I split the index to btree(id) and btree(uid, did), everything's going perfect, but what caused those can not be used together? Any details or rules?
BTW, the sql is used for Real-Time Calculation, there are some Window Functions code didn't show here.
The Nested Loop does not take much time actually. The actual time of 13.709..13.711 means that it took 13.709 ms until the first row was ready to be emitted from this node and it took 0.002 ms until it was finished.
Note that the startup cost of 13.709 ms includes the cost of its two child nodes. Both of the child nodes need to emit at least one row before the nested loop can start.
The Unique child began emitting its first (and only) row after 0.011 ms. The Index Only Scan child however only started to emit its first (and only) row after 13.695 ms. This means that most of your actual time spent is in this Index Only Scan.
There is a great answer here which explains the costs and actual times in depth.
Also there is a nice tool at https://explain.depesz.com which calculates an inclusive and exclusive time for each node. Here it is used for your query plan which clearly shows that most of the time is spent in the Index Only Scan.
Since the query is spending almost all of the time in this index only scan, optimizations there will have the most benefit. Creating a separate index for the columns uid and did on the operation table should improve query time a lot.
CREATE INDEX operation_uid_did ON operation(uid, did);
The current execution plan contains 2 index only scans.
A slow one:
-> Index Only Scan using oi on operation (cost=0.42..29686.32 rows=100 width=24) (actual time=13.695..13.696 rows=1 loops=1)
Index Cond: ((uid = operation_1.uid) AND (did = operation_1.did))
Heap Fetches: 1
Buffers: shared hit=4900
And a fast one:
-> Index Only Scan using oi on operation operation_1 (cost=0.42..8.44 rows=1 width=16) (actual time=0.011..0.011 rows=1 loops=1)
Index Cond: (id = 993754)
Heap Fetches: 1
Buffers: shared hit=5
Both of them use the index oi but have different index conditions. Note how the fast one, who uses the id as index condition only needs to load 5 pages of data (Buffers: shared hit=5). The slow one needs to load 4900 pages instead (Buffers: shared hit=4900). This indicates that the index is optimized to query for id but not so much for uid and did. Probably the index oi covers all 3 columns id, uid, did in this order.
A multi-column btree index can only be used efficently when there are constraints in the query on the leftmost columns. The official documentation about multi-column indexes explains this very well in depth.
Why does Nested Loop cost more and more time than sum of childs cost?
Based on your example, it doesn't. Can you elaborate on what makes you think it does?
Anyway, it seems extravagant to visit 4900 pages to fetch 1 tuple. I'm guessing your tables are not getting vacuumed enough.
Although now I prefer Florian's suggestion, that "uid" and "did" are not the leading columns of the index, and that is why it is slow. It is basically doing a full index scan, using the index as a skinny version of the table. It is a shame that EXPLAIN output doesn't make it clear when a index is being used in this fashion, rather than the traditional "jump to a specific part of the index"
So you have a missing index.