How do I modify fields inside the new PostgreSQL JSON datatype?

Question

With postgresql 9.3 I can SELECT specific fields of a JSON data type, but how do you modify them using UPDATE? I can't find any examples of this in the postgresql documentation, or anywhere online. I have tried the obvious:

postgres=# create table test (data json);
CREATE TABLE
postgres=# insert into test (data) values ('{"a":1,"b":2}');
INSERT 0 1
postgres=# select data->'a' from test where data->>'b' = '2';
 ?column?
----------
 1
(1 row)
postgres=# update test set data->'a' = to_json(5) where data->>'b' = '2';
ERROR:  syntax error at or near "->"
LINE 1: update test set data->'a' = to_json(5) where data->>'b' = '2...

pozs pozs · Accepted Answer · 2014-05-06T17:00:31

Update: With PostgreSQL 9.5, there are some jsonb manipulation functionality within PostgreSQL itself (but none for json; casts are required to manipulate json values).

Merging 2 (or more) JSON objects (or concatenating arrays):

SELECT jsonb '{"a":1}' || jsonb '{"b":2}', -- will yield jsonb '{"a":1,"b":2}'
       jsonb '["a",1]' || jsonb '["b",2]'  -- will yield jsonb '["a",1,"b",2]'

So, setting a simple key can be done using:

SELECT jsonb '{"a":1}' || jsonb_build_object('<key>', '<value>')

Where <key> should be string, and <value> can be whatever type to_jsonb() accepts.

For setting a value deep in a JSON hierarchy, the jsonb_set() function can be used:

SELECT jsonb_set('{"a":[null,{"b":[]}]}', '{a,1,b,0}', jsonb '{"c":3}')
-- will yield jsonb '{"a":[null,{"b":[{"c":3}]}]}'

Full parameter list of jsonb_set():

jsonb_set(target         jsonb,
          path           text[],
          new_value      jsonb,
          create_missing boolean default true)

path can contain JSON array indexes too & negative integers that appear there count from the end of JSON arrays. However, a non-existing, but positive JSON array index will append the element to the end of the array:

SELECT jsonb_set('{"a":[null,{"b":[1,2]}]}', '{a,1,b,1000}', jsonb '3', true)
-- will yield jsonb '{"a":[null,{"b":[1,2,3]}]}'

For inserting into JSON array (while preserving all of the original values), the jsonb_insert() function can be used (in 9.6+; this function only, in this section):

SELECT jsonb_insert('{"a":[null,{"b":[1]}]}', '{a,1,b,0}', jsonb '2')
-- will yield jsonb '{"a":[null,{"b":[2,1]}]}', and
SELECT jsonb_insert('{"a":[null,{"b":[1]}]}', '{a,1,b,0}', jsonb '2', true)
-- will yield jsonb '{"a":[null,{"b":[1,2]}]}'

Full parameter list of jsonb_insert():

jsonb_insert(target       jsonb,
             path         text[],
             new_value    jsonb,
             insert_after boolean default false)

Again, negative integers that appear in path count from the end of JSON arrays.

So, f.ex. appending to an end of a JSON array can be done with:

SELECT jsonb_insert('{"a":[null,{"b":[1,2]}]}', '{a,1,b,-1}', jsonb '3', true)
-- will yield jsonb '{"a":[null,{"b":[1,2,3]}]}', and

However, this function is working slightly differently (than jsonb_set()) when the path in target is a JSON object's key. In that case, it will only add a new key-value pair for the JSON object when the key is not used. If it's used, it will raise an error:

SELECT jsonb_insert('{"a":[null,{"b":[1]}]}', '{a,1,c}', jsonb '[2]')
-- will yield jsonb '{"a":[null,{"b":[1],"c":[2]}]}', but
SELECT jsonb_insert('{"a":[null,{"b":[1]}]}', '{a,1,b}', jsonb '[2]')
-- will raise SQLSTATE 22023 (invalid_parameter_value): cannot replace existing key

Deleting a key (or an index) from a JSON object (or, from an array) can be done with the - operator:

SELECT jsonb '{"a":1,"b":2}' - 'a', -- will yield jsonb '{"b":2}'
       jsonb '["a",1,"b",2]' - 1    -- will yield jsonb '["a","b",2]'

Deleting, from deep in a JSON hierarchy can be done with the #- operator:

SELECT '{"a":[null,{"b":[3.14]}]}' #- '{a,1,b,0}'
-- will yield jsonb '{"a":[null,{"b":[]}]}'

For 9.4, you can use a modified version of the original answer (below), but instead of aggregating a JSON string, you can aggregate into a json object directly with json_object_agg().

Original answer: It is possible (without plpython or plv8) in pure SQL too (but needs 9.3+, will not work with 9.2)

CREATE OR REPLACE FUNCTION "json_object_set_key"(
  "json"          json,
  "key_to_set"    TEXT,
  "value_to_set"  anyelement
)
  RETURNS json
  LANGUAGE sql
  IMMUTABLE
  STRICT
AS $function$
SELECT concat('{', string_agg(to_json("key") || ':' || "value", ','), '}')::json
  FROM (SELECT *
          FROM json_each("json")
         WHERE "key" <> "key_to_set"
         UNION ALL
        SELECT "key_to_set", to_json("value_to_set")) AS "fields"
$function$;

SQLFiddle

Edit:

A version, which sets multiple keys & values:

CREATE OR REPLACE FUNCTION "json_object_set_keys"(
  "json"          json,
  "keys_to_set"   TEXT[],
  "values_to_set" anyarray
)
  RETURNS json
  LANGUAGE sql
  IMMUTABLE
  STRICT
AS $function$
SELECT concat('{', string_agg(to_json("key") || ':' || "value", ','), '}')::json
  FROM (SELECT *
          FROM json_each("json")
         WHERE "key" <> ALL ("keys_to_set")
         UNION ALL
        SELECT DISTINCT ON ("keys_to_set"["index"])
               "keys_to_set"["index"],
               CASE
                 WHEN "values_to_set"["index"] IS NULL THEN 'null'::json
                 ELSE to_json("values_to_set"["index"])
               END
          FROM generate_subscripts("keys_to_set", 1) AS "keys"("index")
          JOIN generate_subscripts("values_to_set", 1) AS "values"("index")
         USING ("index")) AS "fields"
$function$;

Edit 2: as @ErwinBrandstetter noted these functions above works like a so-called UPSERT (updates a field if it exists, inserts if it does not exist). Here is a variant, which only UPDATE:

CREATE OR REPLACE FUNCTION "json_object_update_key"(
  "json"          json,
  "key_to_set"    TEXT,
  "value_to_set"  anyelement
)
  RETURNS json
  LANGUAGE sql
  IMMUTABLE
  STRICT
AS $function$
SELECT CASE
  WHEN ("json" -> "key_to_set") IS NULL THEN "json"
  ELSE (SELECT concat('{', string_agg(to_json("key") || ':' || "value", ','), '}')
          FROM (SELECT *
                  FROM json_each("json")
                 WHERE "key" <> "key_to_set"
                 UNION ALL
                SELECT "key_to_set", to_json("value_to_set")) AS "fields")::json
END
$function$;

Edit 3: Here is recursive variant, which can set (UPSERT) a leaf value (and uses the first function from this answer), located at a key-path (where keys can only refer to inner objects, inner arrays not supported):

CREATE OR REPLACE FUNCTION "json_object_set_path"(
  "json"          json,
  "key_path"      TEXT[],
  "value_to_set"  anyelement
)
  RETURNS json
  LANGUAGE sql
  IMMUTABLE
  STRICT
AS $function$
SELECT CASE COALESCE(array_length("key_path", 1), 0)
         WHEN 0 THEN to_json("value_to_set")
         WHEN 1 THEN "json_object_set_key"("json", "key_path"[l], "value_to_set")
         ELSE "json_object_set_key"(
           "json",
           "key_path"[l],
           "json_object_set_path"(
             COALESCE(NULLIF(("json" -> "key_path"[l])::text, 'null'), '{}')::json,
             "key_path"[l+1:u],
             "value_to_set"
           )
         )
       END
  FROM array_lower("key_path", 1) l,
       array_upper("key_path", 1) u
$function$;

Updated: Added function for replacing an existing json field's key by another given key. Can be in handy for updating data types in migrations or other scenarios like data structure amending.

CREATE OR REPLACE FUNCTION json_object_replace_key(
    json_value json,
    existing_key text,
    desired_key text)
  RETURNS json AS
$BODY$
SELECT COALESCE(
(
    SELECT ('{' || string_agg(to_json(key) || ':' || value, ',') || '}')
    FROM (
        SELECT *
        FROM json_each(json_value)
        WHERE key <> existing_key
        UNION ALL
        SELECT desired_key, json_value -> existing_key
    ) AS "fields"
    -- WHERE value IS NOT NULL (Actually not required as the string_agg with value's being null will "discard" that entry)

),
    '{}'
)::json
$BODY$
  LANGUAGE sql IMMUTABLE STRICT
  COST 100;

Update: functions are compacted now.

How do I modify fields inside the new PostgreSQL JSON datatype?

20 Answers