Remove a known subset of nodes from JSON, in SQLAzure - sql

Question

Once upon a time, there was one row of data (massively simplified, the actual json data is 10KB+) thus:
ID, json
1, '{
"a1.arr": [1,2,3],
"a1.boo": true,
"a1.str": "hello",
"a1.num": 123
}'
A process was supposed to write another record with predominantly different data:
ID, json
2, '{
"a1.arr": [1,2,3], //from ID 1
"a2.arr": [4,5,6], //new (and so are all below)
"a2.boo": false,
"a2.str": "goodbye",
"a2.num": 456
}'
But due to some external error, the original set of json from ID 1 ended up also being represented in ID 2, so now the table looks like this:
ID, json
1, '{
"a1.arr": [1,2,3],
"a1.boo": true,
"a1.str": "hello",
"a1.num": 123
}'
2, '{
"a1.arr": [1,2,3],
"a1.boo": true, //extraneous
"a1.str": "hello", //extraneous
"a1.num": 123, //extraneous
"a2.arr": [4,5,6],
"a2.boo": false,
"a2.str": "goodbye",
"a2.num": 456
}'
I'd like to know if there's a way to remove the extraneous lines from the ID 2 record.
I believe that the entire JSON string from ID 1 is represented in ID 2 as a contiguous block, so string replacement could work but there's a chance that some reordering has taken place. Gets a bit messy with the element that is supposed to remain, though
There's also a chance that some of the a1.* nodes' values have been changed slightly, (I didn't do a diff) but I'm happy to use just the node names, not their values, in assessing whether an node should be removed. One of the nodes (a1.arr) should be kept in ID 2. The resultset should hence look like:
ID, json
1, '{
"a1.arr": [1,2,3],
"a1.boo": true,
"a1.str": "hello",
"a1.num": 123
}'
2, '{
"a1.arr": [1,2,3],
"a2.arr": [4,5,6],
"a2.boo": false,
"a2.str": "goodbye",
"a2.num": 456
}'
I've started playing about with https://dba.stackexchange.com/questions/168303/can-sql-server-2016-extract-node-names-from-json to get the list of node names from ID 1 that I want to remove from ID 2, just not sure how I then strip them out of ID 2's JSON - presumably a deserialize, reduce and reserialize sequence?

Answer 1

You can follow this approach:
get the keys you want to replace with an openjson on row with id=1 value
use cross apply to filter keys in rows with id>1
rebuild the json string without unwanted keys using STRING_AGG and group by
This code should work:
declare #tmp table ([id] int, jsonValue nvarchar(max))
declare #source_json nvarchar(max)
insert into #tmp values
(1, '{"a1.arr":[1,2,3],"a1.boo":true,"a1.str":"hello","a1.num":123}')
,(2, '{"a1.arr":[1,2,3],"a1.boo":true,"a1.str":"hello","a1.num":123,"a2.arr":[4,5,6],"a2.boo":false,"a2.str":"goodbye","a2.num":456}')
,(3, '{"a1.arr":[1,2,3],"a1.boo":true,"a1.str":"hello","a1.num":123,"a3.arr":[4,5,6],"a3.boo":false,"a3.str":"goodbye","a3.num":456}')
--get json string from id=1
select #source_json = jsonValue from #tmp where [id] = 1
--rebuild json string for id > 1 removing keys from id = 1
select t.[id],
'{' + STRING_AGG( '"' + g.[key] + '":"' + STRING_ESCAPE(g.[value], 'json') + '"', ',') + '}' as [json]
from #tmp t cross apply openjson(jsonValue) g
where t.id > 1
and g.[key] not in (select [key] from openjson(#source_json) where [key] <> 'a1.arr')
group by t.id
Result: