To split records in one column into multiple columns - sql-server

Question

I am not sure whether this is achievable in sql server or not. I have a table ADDRESS and its metadata is given below. SQ_ID is my unique key column
SQ_ID INT
ADDRESS_LINE_1 varchar(255)
ADDRESS_LINE_2 varchar(255)
ADDRESS_LINE_3 varchar(255)
ADDRESS_LINE_4 varchar(255)
REGION varchar(255)
POSTOCODE varchar(255)
COUNTRY_CODE varchar(255)
Now I have loaded around 2 millions of records into this table and the problem is that all the address details have been loaded into ADDRESS LINE 1 column in this table. Now I am trying to find a way to split this. Below given is some of the sample set of addresses.
So I want to split the data in Address line 1 in such a way that
value in addressLine1 in the given record should be populated to
addressline1 in the same table value in addressLine2 in the given
record should be populated to addressline2 in the same table value in
addressLine3 in the given record should be populated to addressline3
in the same table value in primaryTown in the given record should be
populated to addressline4 in the same table value in provinceOrState
in the given record should be populated to region in the same table
value in isoAlpha2Code in the given record should be populated to
COUNTRY_CODE in the same table value in zipOrPostalCode in the given
record should be populated to POSTOCODE in the same table
My Expected output is given below
Also I am attaching set of sample address values for testing
{'addressLine1': '67 xxxx Road', 'primaryTown': 'HOxxxCHxxCH',
'zipOrPostalCode': 'RM11 1EX', 'addressCountry': {'isoAlpha2Code':
'GB'}}
{'primaryTown': 'MünXXer', 'addressCountry': {'isoAlpha2Code': 'DE'}}
{'addressLine1': '28 THE EXCAC', 'primaryTown': 'PERTH',
'provinceOrState': 'WA', 'addressCountry': {'isoAlpha2Code': 'AU'}}
{'addressLine1': '28 THE ESPLANADE', 'primaryTown': 'PERTH',
'provinceOrState': 'WA', 'addressCountry': {'isoAlpha2Code': 'AU'}}
{'addressLine1': '76 XXX STREET', 'primaryTown': 'MAXDFOT',
'provinceOrState': 'NSW', 'addressCountry': {'isoAlpha2Code': 'AU'}}
{'addressLine1': 'UNIT 56', 'addressLine2': '22 XDFR STREET',
'primaryTown': 'MANLY VALE', 'provinceOrState': 'NSW',
'addressCountry': {'isoAlpha2Code': 'AU'}}
{'addressLine1': 'BjoXCDSaret 15', 'addressLine2': 'Jppsdeheim',
'addressLine3': '', 'primaryTown': 'AKERdwfUS', 'addressCountry':
{'isoAlpha2Code': 'NO'}}

Answer 1

You can't parse JSON by splitting. SQL Server 2016 and later supports JSON though, so you can extract the data you want using functions like JSON_VALUE.
You don't have to do that though, unless you want to index and filter those values. It may make sense to extract the country or zip code if you intend to filter or group results using them. Address lines on the other hand can't be queried and might as well remain in the JSON string. You can always extract them when necessary with JSON_VALUE.
Puttin primaryTown in AddressLine4 looks like a mistake though. That is a significant attribute that could be used in querying. It should go to its own field, or not extracted at all.
SQL Server 2016 and later
In any case, you can parse a JSON value with JSON_VALUE , eg :
declare #myTable table(json varchar(max))
insert into #myTable
values
('{"addressLine1": "BjoXCDSaret 15", "addressLine2": "Jppsdeheim","addressLine3": "", "primaryTown": "AKERdwfUS", "addressCountry":
{"isoAlpha2Code": "NO"}}'),
('{"addressLine1": "67 xxxx Road", "primaryTown": "HOxxxCHxxCH",
"zipOrPostalCode": "RM11 1EX", "addressCountry": {"isoAlpha2Code":
"GB"}}'),
('{"primaryTown": "MünXXer", "addressCountry": {"isoAlpha2Code": "DE"}}'),
('{"addressLine1": "28 THE EXCAC", "primaryTown": "PERTH",
"provinceOrState": "WA", "addressCountry": {"isoAlpha2Code": "AU"}}'),
('{"addressLine1": "28 THE ESPLANADE", "primaryTown": "PERTH",
"provinceOrState": "WA", "addressCountry": {"isoAlpha2Code": "AU"}}'),
('{"addressLine1": "76 XXX STREET", "primaryTown": "MAXDFOT",
"provinceOrState": "NSW", "addressCountry": {"isoAlpha2Code": "AU"}}'),
('{"addressLine1": "UNIT 56", "addressLine2": "22 XDFR STREET",
"primaryTown": "MANLY VALE", "provinceOrState": "NSW",
"addressCountry": {"isoAlpha2Code": "AU"}}')
select
JSON_VALUE(json,'$.zipOrPostalCode') as ZipCode,
JSON_VALUE(json,'$.primaryTown') as primaryTown,
JSON_VALUE(json,'$.addressCountry.isoAlpha2Code') as CountryCode
from #mytable
You can use the same expressions to create persisted computed columns on the table and index them, thus speeding queries that need to filter by country or zip code.
Older versions
In older versions, perhaps the simplest solution would be to create a SQLCLR function that uses JSON.NET to parse the data and return the values.
Another option is to convert the JSON string to XML using string replacements and use XPATH to retrieve values. This can get really tricky though as the replacements depend on the expected data, are sensitive to whitespace and can easily break when dealing with nested objects.
For example, this flat JSON object :
declare #json varchar(max)='{"addressLine1": "BjoXCDSaret 15", "addressLine2": "Jppsdeheim", "addressLine3": "", "primaryTown": "AKERdwfUS"}'
Can be converted to XML with a few replacements. Notice that I fixed whitespace differences to ensure there's always a space between tokens. Otherwise I'd have to replace both "," and ", ".
select
cast(
replace(
replace(
replace(
replace(#json,'{"','<it '),
'": "',' ="'),
'", "','" '),
'}',' />')
as xml)
The result is :
<it addressLine1="BjoXCDSaret 15" addressLine2="Jppsdeheim" addressLine3="" primaryTown="AKERdwfUS" />
This can now be queried with .value :
select
cast(replace(replace(replace(replace(#json,'{"','<it '),'": "',' ="'),'", "','" '),'}',' />') as xml)
.value('(/it/#primaryTown)[1]','varchar(20)')
This will return :
AKERdwfUS
This breaks with the nested addressCountry object though. If you know what the JSON text contains, you could cheat and replace specific attributes, not just separators, eg :
declare #json varchar(max)='{"addressLine1": "UNIT 56", "addressLine2": "22 XDFR STREET", "primaryTown": "MANLY VALE", "provinceOrState": "NSW", "addressCountry": {"isoAlpha2Code": "AU"}}'
select
cast(
replace(
replace(
replace(
replace(
replace(
replace(#json,'", "addressCountry": {"','"><addressCountry '),
'}}','/></it>'),
'{"','<it '),
'": "',' ="'),
'", "','" ')
,'}',' />')
as xml).value('(/it/addressCountry/#isoAlpha2Code)[1]','varchar(20)')
This returns AU.
That's some serious cheating though that can only work through trial and error. In this case, the addressCountry attribute and the following separators are converted to an element. }} is expected to appear only at the end of the string, so it gets special treatment.
Use a client script
It's probably easier to use a small .NET program to read the data using JSON.NET and extract the desired values. 2M rows aren't a lot so parsing the data once in a while won't be a big problem