PostgreSQL: sequential table scan on Hstore column despite having a GiN index

Question

I have a table with a hstore column and roughly 22 mio records (the ways table from a partial osm-database).

Despite having a GIN index on the hstore column, queries for a specific tag result in a sequential table scan that takes > 60 sec to return a single column.

What i have been doing so far.

1. I created the GIN index using pgAdminIII.
2. Executing vacuum analayze
3. Executing a query of the kind: select id from table where tags->'name'='foo'
4. Deleting index and starting from 1. again ...
5. [Edit] As suggested by the user a_horse_with_no_name I updated the table statistics by executing analyze on the table. But that had no effect.

You can see the query plan here. For some reason the explain analyze takes only ~20 sec to complete.

How can I properly index a hstore column on a large table like this, to reduce query execution cost significantly?

Thank you for your help!

Answer 1

I see two possible solutions:

If you always query that key value for equality you can use an a B-Tree index on the expression (`tags -> 'name')

create index idx_name on ways ( (tags -> 'name') );

A quick test has shown that Postgres does use the index to find if a key value is present in the hstore column, but apparently not for finding the associated value.

So you could try to add a condition to test for that key value as well:

select id
from ways
where tags ? 'name' 
  and tags -> 'name' = 'Wiehbergpark';

If all rows contain that key, it might not help though.