Query for array elements inside JSON type

pacothelovetaco picture pacothelovetaco · Mar 29, 2014 · Viewed 168.5k times · Source

I'm trying to test out the json type in PostgreSQL 9.3.
I have a json column called data in a table called reports. The JSON looks something like this:

{
  "objects": [
    {"src":"foo.png"},
    {"src":"bar.png"}
  ],
  "background":"background.png"
}

I would like to query the table for all reports that match the 'src' value in the 'objects' array. For example, is it possible to query the DB for all reports that match 'src' = 'foo.png'? I successfully wrote a query that can match the "background":

SELECT data AS data FROM reports where data->>'background' = 'background.png'

But since "objects" has an array of values, I can't seem to write something that works. Is it possible to query the DB for all reports that match 'src' = 'foo.png'? I've looked through these sources but still can't get it:

I've also tried things like this but to no avail:

SELECT json_array_elements(data->'objects') AS data from reports
WHERE  data->>'src' = 'foo.png';

I'm not an SQL expert, so I don't know what I am doing wrong.

Answer

Erwin Brandstetter picture Erwin Brandstetter · Mar 29, 2014

json in Postgres 9.3+

Unnest the JSON array with the function json_array_elements() in a lateral join in the FROM clause and test for its elements:

WITH reports(data) AS (
   VALUES ('{"objects":[{"src":"foo.png"}, {"src":"bar.png"}]
           , "background":"background.png"}'::json)
   ) 
SELECT *
FROM   reports r, json_array_elements(r.data#>'{objects}') obj
WHERE  obj->>'src' = 'foo.png';

The CTE (WITH query) just substitutes for a table reports.
Or, equivalent for just a single level of nesting:

SELECT *
FROM   reports r, json_array_elements(r.data->'objects') obj
WHERE  obj->>'src' = 'foo.png';

->>, -> and #> operators are explained in the manual.

Both queries use an implicit JOIN LATERAL.

SQL Fiddle.

Closely related answer:

jsonb in Postgres 9.4+

Use the equivalent jsonb_array_elements().

Better yet, use the new "contains" operator @> (best in combination with a matching GIN index on the expression data->'objects'):

CREATE INDEX reports_data_gin_idx ON reports
USING gin ((data->'objects') jsonb_path_ops);

SELECT * FROM reports WHERE data->'objects' @> '[{"src":"foo.png"}]';

Since the key objects holds a JSON array, we need to match the structure in the search term and wrap the array element into square brackets, too. Drop the array brackets when searching a plain record.

Detailed explanation and more options: