I'm attempting to parse json data from zendesk using v: structure - arrays

Question

With standard fields, like id, this works perfectly. But I am not finding a way to parse the custom fields where the structure is
"custom_fields": [
{
"id": 57852188,
"value": ""
},
{
"id": 57522467,
"value": ""
},
{
"id": 57522487,
"value": ""
}
]
The general format that I have been using is:
Select v:id,v:updatedat
from zd_tickets
updated data:
{
"id":151693,
"brand_id": 36000,
"created_at": "2022-0523T19:26:35Z",
"custom_fields": [
{ "id": 57866008, "value": false },
{ "id": 360022282754, "value": "" },
{ "id": 80814087, "value": "NC" } ],
"group_id": 36000770
}

Answer 1

If you want to select all repeating elements you will need to use FLATTEN, otherwise you can use standard notation. This is all documented here: https://docs.snowflake.com/en/user-guide/querying-semistructured.html#retrieving-a-single-instance-of-a-repeating-element

Answer 2

So using this CTE to access the data in a way that look like a table:
with data(json) as (
select parse_json(column1) from values
('{
"id":151693,
"brand_id": 36000,
"created_at": "2022-0523T19:26:35Z",
"custom_fields": [
{ "id": 57866008, "value": false },
{ "id": 360022282754, "value": "" },
{ "id": 80814087, "value": "NC" } ],
"group_id": 36000770
} ')
)
SQL to unpack the top level items, as you have shown you have working:
select
json:id::number as id
,json:brand_id::number as brand_id
,try_to_timestamp(json:created_at::text, 'yyyy-mmddThh:mi:ssZ') as created_at
,json:custom_fields as custom_fields
from data;
gives:
ID
BRAND_ID
CREATED_AT
CUSTOM_FIELDS
151693
36000
2022-05-23 19:26:35.000
[ { "id": 57866008, "value": false }, { "id": 360022282754, "value": "" }, { "id": 80814087, "value": "NC" } ]
So now how to tackle that json/array of custom_fields..
Well if you only ever have 3 values, and the order is always the same..
select
to_array(json:custom_fields) as custom_fields_a
,custom_fields_a[0] as field_0
,custom_fields_a[1] as field_1
,custom_fields_a[2] as field_2
from data;
gives:
CUSTOM_FIELDS_A
FIELD_0
FIELD_1
FIELD_2
[ { "id": 57866008, "value": false }, { "id": 360022282754, "value": "" }, { "id": 80814087, "value": "NC" } ]
{ "id": 57866008, "value": false }
{ "id": 360022282754, "value": "" }
{ "id": 80814087, "value": "NC" }
so we can use flatten to access those objects, which makes "more rows"
select
d.json:id::number as id
,d.json:brand_id::number as brand_id
,try_to_timestamp(d.json:created_at::text, 'yyyy-mmddThh:mi:ssZ') as created_at
,f.*
from data as d
,table(flatten(input=>json:custom_fields)) f
ID
BRAND_ID
CREATED_AT
SEQ
KEY
PATH
INDEX
VALUE
THIS
151693
36000
2022-05-23 19:26:35.000
1
[0]
0
{ "id": 57866008, "value": false }
[ { "id": 57866008, "value": false }, { "id": 360022282754, "value": "" }, { "id": 80814087, "value": "NC" } ]
151693
36000
2022-05-23 19:26:35.000
1
[1]
1
{ "id": 360022282754, "value": "" }
[ { "id": 57866008, "value": false }, { "id": 360022282754, "value": "" }, { "id": 80814087, "value": "NC" } ]
151693
36000
2022-05-23 19:26:35.000
1
[2]
2
{ "id": 80814087, "value": "NC" }
[ { "id": 57866008, "value": false }, { "id": 360022282754, "value": "" }, { "id": 80814087, "value": "NC" } ]
So we can pull out know values (a manual PIVOT)
select
d.json:id::number as id
,d.json:brand_id::number as brand_id
,try_to_timestamp(d.json:created_at::text, 'yyyy-mmddThh:mi:ssZ') as created_at
,max(iff(f.value:id=80814087, f.value:value::text, null)) as v80814087
,max(iff(f.value:id=360022282754, f.value:value::text, null)) as v360022282754
,max(iff(f.value:id=57866008, f.value:value::text, null)) as v57866008
from data as d
,table(flatten(input=>json:custom_fields)) f
group by 1,2,3, f.seq
grouping by the f.seq means if you have many "rows" of input these will be kept apart, even if they share common values for 1,2,3
gives:
ID
BRAND_ID
CREATED_AT
V80814087
V360022282754
V57866008
151693
36000
2022-05-23 19:26:35.000
NC
<empty string>
false
Now if you do not know the names of the values, there is no way short of dynamic SQL and double parsing to turns rows into columns.

Answer 3

I ended up doing the following, with 2 different CTEs (CTE and UCF):
Used to_array to gather my custom fields
Unioned the custom fields together twice; once for the id of the field and once for the value (and used combinations of substring, position and replace to clean up data as needed (same setup for all fields)
Joined the resulting data to a Custom Fields Table (contains the id and a name) to include the name of the custom field in my result set.
WITH UCF AS (--Union Gathered Array into 2 fields (an id field and a value field)
WITH CTE AS( ---Gather array of custom fields
SELECT v:id as id,
to_array(v:custom_fields) as cf
,cf[0] as f0,cf[1] as f1,cf[2] as f2
FROM ZD_TICKETS)
SELECT id,
substring(f0,7,position(',',f0)-7) AS cf_id, REPLACE(substring(f0,position('value":',f0)+8,position('"',f0,position('value":',f0)+8)),'"}') AS cf_value
FROM CTE c
WHERE f0 not like '%null%'
UNION
SELECT id,
substring(f1,7,position(',',f1)-7) AS cf_id,
REPLACE(substring(f1,position('value":',f1)+8,position('"',f1,position('value":',f1)+8)),'"}') AS cf_value
FROM CTE c
WHERE f1 not like '%null%'
-- field 3
UNION
SELECT id,
substring(f2,7,position(',',f2)-7) AS cf_id,
REPLACE(substring(f2,position('value":',f2)+8,position('"',f2,position('value":',f2)+8)),'"}') AS cf_value
FROM CTE c
WHERE f2 not like '%null%' --this removes records where the value is null
)
SELECT UCF.*,CFD.name FROM UCF
LEFT OUTER JOIN "FLBUSINESS_DB"."STAGING"."FILE_ZD_CUSTOM_FIELD_IDS" CFD
ON CFD.id=UCF.cf_id
WHERE cf_value<>'' --this removes records where the value is blank
The result set looks like: