Planner does not use btree index for LIKE 'prefix%' on text column, but does for equivalent range query (PostgreSQL 17.4)
Hello PostgreSQL team,
I have encountered what appears to be a planner regression or bug in
PostgreSQL 17.4 regarding the use of btree indexes for prefix searches with
LIKE 'prefix%'.
A btree index on a text column is not used for a query of the form WHERE
col LIKE 'prefix%', even though the index is healthy, statistics are
correct, and the query is highly selective.
The same index is used and performs well for an equivalent range query
(WHERE col >= 'prefix' AND col < 'nextprefix').
- This behavior persists even after VACUUM FULL, ANALYZE, and index
re-creation.
- The column and index use the default collation (en_US.UTF-8).
- The index is valid and ready.
- The planner does not use the index for LIKE 'prefix%', but does for the
equivalent range.
- Rewriting the query as a range (col >= 'prefix' AND col < 'nextprefix')
uses the index and is fast, but this should not be necessary.
Thank you for your attention!
Abdullah DURSUN <adursuns@gmail.com> writes:
A btree index on a text column is not used for a query of the form WHERE
col LIKE 'prefix%', even though the index is healthy, statistics are
correct, and the query is highly selective.
LIKE with a prefix condition can only use an index if the index's
collation is "C", or if it uses the "pattern" opclass. This isn't
new in v17, it's been true for a very long time.
d1=# create table foo (t text);
CREATE TABLE
d1=# create index on foo (t);
CREATE INDEX
d1=# explain select * from foo where t like 'prefix%';
QUERY PLAN
-----------------------------------------------------
Seq Scan on foo (cost=0.00..27.00 rows=7 width=32)
Filter: (t ~~ 'prefix%'::text)
(2 rows)
d1=# create index on foo (t collate "C");
CREATE INDEX
d1=# explain select * from foo where t like 'prefix%';
QUERY PLAN
-------------------------------------------------------------------------
Bitmap Heap Scan on foo (cost=4.22..14.37 rows=7 width=32)
Filter: (t ~~ 'prefix%'::text)
-> Bitmap Index Scan on foo_t_idx1 (cost=0.00..4.22 rows=7 width=0)
Index Cond: ((t >= 'prefix'::text) AND (t < 'prefiy'::text))
(4 rows)
regards, tom lane