Re: Prefix LIKE search and indexes issue.

On 23 Jul 2010, at 23:22, Marcelo de Moraes Serpa wrote:
> The following query works:
>
> SELECT * FROM users WHERE 'com.app.mycompany' LIKE reversed_domain || %
>
> However, it does sequential search, meaning it doesn't use any index.

The database may choose to use a seqscan for several reasons, not necessarily related to how you write your query. Is it a problem in your case? An EXPLAIN ANALYSE would give us more insight.

I would expect the planner to pick an index scan if there's sufficient data in that table, as a suffix-search like that suits a btree index just fine.

> What I would like to know is, how could I make it use an index? I've

If you really have to, for testing purposes you can temporarily disable sequential scans (SET enable_seqscan TO 'off') to see if the plan resulting from that is indeed more efficient.

If it is, that probably means your database statistics aren't up to date ([VACUUM ]ANALYSE) or the statistics target for that specific column is too small (ALTER TABLE users ALTER COLUMN reversed_domain SET STATISTICS ....).

> done some research and asked around #postgres but things are still not
> clear to me. Some good souls hinted me at the prefix extension, but
> how would I use it? Is there any other simpler / extension-free way to
> solve this issue?

I'm not familiar with said extension, but I think that one's aimed at LIKE searches where the search term _starts_ with a wildcard instead of _ending_ with one. That's a situation where a btree index is in trouble.

Alban Hertroys

--
If you can't see the forest for the trees,
cut the trees and you'll see there is no forest.

!DSPAM:737,4c4abfcd286214416410229!