I have table with above 100mln rows. I have to count, and extract rows as in following query. The query runs verly long. Explain shows that query doesn't use b-tree index which is created on "created_date" column. I found on stackoverflow some explanation, that b-trees indicies are useless to filter when table has many rows.
There is an advice to Cluster Index. Should i Cluster table on "created_date" index, if i also use often query, where i ORDER BY id?
What would you advice me to faster queries? Maybe should i read more about sharding?
explain SELECT count(r.id) FROM results_new r
WHERE r.searches_id = 4351940 AND (created_date between '2008-01-01' and '2012-12-13')
Limit (cost=1045863.78..1045863.79 rows=1 width=4)
-> Aggregate (cost=1045863.78..1045863.79 rows=1 width=4)
-> Index Scan using results_new_searches_id_idx on results_new r (cost=0.00..1045012.38 rows=340560 width=4)"
Index Cond: (searches_id = 4351940)"
Filter: ((created_date >= '2008-01-01 00:00:00'::timestamp without time zone) AND (created_date <= '2012-12-13 00:00:00'::timestamp without time zone))
From the look of it, the database has decided that a lookup for one searches_id will produce fewer rows to go through than a lookup for the created_date range. (and that combining the result of two index scans with a bitmap isn't worthwhile...)
If you need this query often, then consider creating an index on searches_id, created_date and then both conditions should go into the index condition.