53

I have a table in a postgresql 9.4 database with a jsonb field called receivers. Some example rows:

[{"id": "145119603", "name": "145119603", "type": 2}]
[{"id": "1884595530", "name": "1884595530", "type": 1}]
[{"id": "363058213", "name": "363058213", "type": 1}]
[{"id": "1427965764", "name": "1427965764", "type": 1}]
[{"id": "193623800", "name": "193623800", "type": 0}, {"id": "419955814", "name": "419955814", "type": 0}]
[{"id": "624635532", "name": "624635532", "type": 0}, {"id": "1884595530", "name": "1884595530", "type": 1}]
[{"id": "791712670", "name": "791712670", "type": 0}]
[{"id": "895207852", "name": "895207852", "type": 0}]
[{"id": "144695994", "name": "144695994", "type": 0}, {"id": "384217055", "name": "384217055", "type": 0}]
[{"id": "1079725696", "name": "1079725696", "type": 0}]

I have a list of values for id and want to select any row that contains an object with any of the values from that list, within the array in the jsonb field.

Is that possible? Is there a GIN index I can make that will speed this up?

Shrayas
  • 6,784
  • 11
  • 37
  • 54
user3761100
  • 553
  • 1
  • 4
  • 5

2 Answers2

85

There is no single operation, which can help you, but you have a few options:

1. If you have a small (and fixed) number of ids to query, you can use multiple containment operators @> combined with or; f.ex.:

where data @> '[{"id": "1884595530"}]' or data @> '[{"id": "791712670"}]'

A simple gin index can help you on your data column here.

2. If you have variable number of ids (or you have a lot of them), you can use json[b]_array_elements() to extract each element of the array, build up an id list and then query it with the any-containment operator ?|:

select *
from   jsonbtest
where  to_json(array(select jsonb_array_elements(data) ->> 'id'))::jsonb ?|
         array['1884595530', '791712670'];

Unfortunately, you cannot index an expression, which has a sub-query in it. If you want to index it, you need to create a function for it:

create function idlist_jsonb(jsonbtest)
  returns jsonb
  language sql
  strict
  immutable
as $func$
  select to_json(array(select jsonb_array_elements($1.data) ->> 'id'))::jsonb
$func$;

create index on jsonbtest using gin (idlist_jsonb(jsonbtest));

After this, you can query ids like this:

select *, jsonbtest.idlist_jsonb
from   jsonbtest
where  jsonbtest.idlist_jsonb ?| array['193623800', '895207852'];

Note: I used dot notation / computed field here, but you don't have to.

3. But at this point, you don't have to stick with json[b]: you have a simple text array, which is supported by PostgreSQL too.

create function idlist_array(jsonbtest)
  returns text[]
  language sql
  strict
  immutable
as $func$
  select array(select jsonb_array_elements($1.data) ->> 'id')
$func$;

create index on jsonbtest using gin (idlist_array(jsonbtest));

And query this computed field with the overlap array operator &&:

select *, jsonbtest.idlist_array
from   jsonbtest
where  jsonbtest.idlist_array && array['193623800', '895207852'];

Note: From my internal testing, this latter solution is calculated with a higher cost than the jsonb variant, but in fact it is faster than that, a little. If performance really matters to you, you should test both.

pozs
  • 34,608
  • 5
  • 57
  • 63
  • 1
    Here's another approach: http://dba.stackexchange.com/questions/130699/postgresql-json-query-array-against-multiple-values – Chrizt0f Feb 01 '17 at 21:41
  • @Chrizt0f that is my 1. approach as well (`ANY` can be expressed by `OR`ed expressions). Usually it's hard to bind `jsonb[]` in client applications, but if OP (or anyone who's interested in this) can do that, it's a valid solution as well -- but note that it will use the index multiple times, so my original note still applies (*If you have a small (and fixed) (sic!) number of ids to query*) -- but with `ANY` it is more comfortable to supply non-fixed number of ids OFC. – pozs Feb 02 '17 at 09:03
11

I find workaround:
where data::text similar to '%("id": "145119603"|"id": "1884595530")%'

taiberium
  • 127
  • 1
  • 2