PostgreSQL & ActiveRecord - slow query - ruby-on-rails

Question

I run simple query:
History.where(channel_id: 1).order('histories.id DESC').first
Result:
History Load (808.8ms) SELECT "histories".* FROM "histories" WHERE "histories"."channel_id" = 1 ORDER BY histories.id DESC LIMIT 1 [["channel_id", 1]]
808.8ms for 1 of 7 records with channel_id = 1. Total histories count is 2,110,443.
If I select all histories for channel_id = 1:
History.where(channel_id: 1)
History Load (0.5ms) SELECT "histories".* FROM "histories" WHERE "histories"."channel_id" = 1 [["channel_id", 1]]
It took only 0.5ms
And if we try to take one record with help of ruby Array:
History.where(channel_id: 1).order('histories.id DESC').to_a.first
History Load (0.5ms) SELECT "histories".* FROM "histories" WHERE "histories"."channel_id" = 1 ORDER BY id DESC [["channel_id", 1]]
Where I should find the problem?
PS: I already have an index on channel_id field.
UPD:
History.where(channel_id: 1).order('histories.id DESC').limit(1).explain
History Load (848.9ms) SELECT "histories".* FROM "histories" WHERE "histories"."channel_id" = 1 ORDER BY histories.id DESC LIMIT 1 [["channel_id", 1]]
=> EXPLAIN for: SELECT "histories".* FROM "histories" WHERE "histories"."channel_id" = 1 ORDER BY histories.id DESC LIMIT 1 [["channel_id", 1]]
QUERY PLAN
-------------------------------------------------------------------------------------------------------
Limit (cost=0.43..13.52 rows=1 width=42)
-> Index Scan Backward using histories_pkey on histories (cost=0.43..76590.07 rows=5849 width=42)
Filter: (channel_id = 1)
(3 rows)

Answer 1

There are two ways PostgreSQL can handle your query (with the ORDER BY and LIMIT clauses):
It can scan the table and order the found tuples, then limit your result. PostgreSQL will choose this plan if he thinks your table has a really low number of tuples, or if he thinks the index will be of no use;
It can use the index.
It seems that PostgreSQL did choose the first option, which can occur for only two reasons in my humble opinion:
Your table statistics are not accurate. I recommend you VACUUM your table and try again this query
The channel_id values are unevenly distributed (for example almost all of the tuples have channel_id=2), which is why PostgreSQL thinks that the index is of no use. Here I recommend the use of a partial index.