Re: JSON Indexes

On Tue, Jun 24, 2014 at 3:15 PM, CS_DBA <cs_dba(at)consistentstate(dot)com> wrote:
> Hi All;
>
> We're struggling to get the planner to use a json index.
>
> I have this table :
>
> Table "public.mytest"
>
> Column | Type | Modifiers
>
> --------+-------------------+-----------------------------------------------------
>
> id | bigint | not null default
> nextval('events_id_seq'::regclass)
>
> task | json |
>
>
> I added a PK constraint on the id column and created this json index:
>
> create index mytest_json_col_idx on mytest ((task->'name'));
>
> However the planner never uses the index...
>
>
> EXPLAIN SELECT (mytest.task->>'name') as name,
>
> COUNT((mytest.task->>'name')) AS task_count
>
> FROM mytest
>
> GROUP BY (mytest.task->>'name')
>
> ORDER BY 2 DESC;
>
>
> QUERY PLAN
>
> -----------------------------------------------------------------------------
>
> Sort (cost=155097.84..155098.34 rows=200 width=32)
>
> Sort Key: (count(((task ->> 'name'::text))))
>
> -> HashAggregate (cost=155087.70..155090.20 rows=200 width=32)
>
> -> Seq Scan on mytab (cost=0.00..149796.94 rows=705435 width=32)
>
>
> Am I missing something?

yes.

first of all, your create index doesn't work for me: I get:

postgres=# create index mytest_json_col_idx on mytest ((task->'name'));
ERROR: data type json has no default operator class for access method "btree"

now, if you change it to:
create index mytest_json_col_idx on mytest ((task->>'name'));

it works. Next, try disabling seq_scan if you want to force an index
scan. It is not a given that a full table count/group by is better
done via an index.

merlin