JSON in 18c and 19c

JSON in Database 18c/19c
JSON as it was meant to be
Stew Ashton
OUG Ireland 2019
Can you read the following line? If not, please move closer.
It's much better when you can read the code ;)

Who am I?
• 1981-2015:
– Developer / application architect
– From Mainframe to client / server to Web
• 2005-present: focus on Oracle DB development
– Advocate of data-centric application architecture
– Contribute to asktom & ODC/OTN SQL forum
– Presented at OOW, UKOUG tech, DOAG, OUGN, ilOUG
– Author of in-house CSV and XML handler
2

JSON is Everywhere
• APEX_JSON
• ORDS : REST-enabled SQL services
• SQL Developer / sqlcl JSON output
• GeoJSON
• Storage of JSON in database tables
– JSON Data Guide
• PL/SQL Object Types for JSON
• SODA (Simple Oracle Document Access):
NoSQL type development in the database
• SQL/JSON : query (JSONSQL) or generate (JSONSQL)
3

Why "JSON as it was meant to be"?
• json.org :
"JSON is a lightweight data-interchange format."
– Not meant to store data
– Deals in data structures, not "documents"
– Interchange between programming languages
• Many advantages over CSV or XML
• So: use JSON to exchange data to or from SQL 👍
4

Data Exchange Headaches
CSV
Encoding ✗ Who knows?
Decimal ✗ NLS settings
Datetime
formats
✗ NLS settings
Field terminator ✗ Open bar
Field enclosure ✗ Open bar
Escaping ✗ field enclosure * 2
Record delimiter ✗ Open bar
5

CSV XML
Encoding ✗ Who knows? ✗ Open bar
Decimal ✗ NLS settings ✗ NLS settings
Datetime
formats
✗ NLS settings ISO 8601
✗ except decimal!
Field terminator ✗ Open bar ✓ </tag>
Field enclosure ✗ Open bar ✓ <tag></tag>
Escaping ✗ field enclosure * 2 ✓ Standard
Record delimiter ✗ Open bar Root element?
6

CSV XML JSON
Encoding ✗ Who knows? ✗ Open bar ✓ UTF-8 (AL32UTF8)
Decimal ✗ NLS settings ✗ NLS settings ✓ Period, period!
Datetime
formats
✗ NLS settings ISO 8601
✗ except decimal!
✓ ISO 8601 (Oracle)
Field terminator ✗ Open bar ✓ </tag> ✓ comma
Field enclosure ✗ Open bar ✓ <tag></tag> ✓ Double quotes
Escaping ✗ field enclosure * 2 ✓ Standard ✓ Standard
Record delimiter ✗ Open bar Root element? ✓ {} or []
7
JSON alone is independent of database parameters

JSON is a data structure
• Object
• Array
8

• Object: unordered collection of name/value pairs
– {"Empno":7369,"Ename":"SMITH"}
{"Ename":"SMITH","Empno":7369}
– XML: <Empno>7369</Empno><Ename>SMITH</Ename>
– Path expression: '$.Ename' , '$.Empno'
• Array: ordered list of values
– [7369,"SMITH","CLERK"]
– CSV: 7369,"SMITH","CLERK"
– Path expression: '$[0]' , '$[1]' , '$[2]'
9
JSON_EQUAL

• Object: unordered collection of name/value pairs
– {"Empno":7369,"Ename":"SMITH"}
{"Ename":"SMITH","Empno":7369}
– XML: <Empno>7369</Empno><Ename>SMITH</Ename>
– Path expression: '$.Ename' , '$.Empno'
• Array: ordered list of values
– [7369,"SMITH","CLERK"]
– CSV: 7369,"SMITH","CLERK"
– Path expression: '$[0]' , '$[1]' , '$[2]'
• "value" can be another object or array
10

JSON Data in SQL Statements
• It’s not a special data type
– VARCHAR2
– CLOB
– BLOB in AL32UTF8 (18c)
• "Oracle recommends that you use AL32UTF8 as the database character set if at all
possible."
• When does Oracle consider data to be JSON?
'{ }' could be an empty JSON object or a JSON string value "{ }"
– Stored data: column with IS JSON constraint
– Output: Result of JSON generation function
– Input: FORMAT JSON in JSON generation function
– TREAT(… AS JSON) (18c)
11

SQL <> JSON Transformation
Direction Source Function Returns Prefer for
Generate 1 row
JSON_OBJECT object OLTP
JSON_ARRAY array ETL
12

Generate
1 row
N rows
JSON_OBJECTAGG object OLTP
JSON_ARRAYAGG array ETL
13

Generate
1 row
N rows
Query JSON
JSON_QUERY 1 JSON part PRETTY
JSON_VALUE 1 Scalar
JSON_EXISTS true/false
14

Generate
1 row
N rows
Query JSON
JSON_QUERY 1 JSON part PRETTY
JSON_VALUE 1 Scalar
JSON_EXISTS true/false
JSON_TABLE Multiple ETL / OLTP
15

SQL Data Type to JSON
16
Input data types JSON result Example
VARCHAR2
string
"varchar2"
NVARCHAR2 "nvarchar2"
CLOB "clob"

17
VARCHAR2
string
"varchar2"
CLOB "clob"
RAW
hex string
"1F2F3F4F"
BLOB "1F2F3F4F5F6F7F8F"

18
VARCHAR2
string
"varchar2"
CLOB "clob"
RAW
hex string
"1F2F3F4F"
NUMBER
number
6
BINARY_DOUBLE 7
BINARY_FLOAT 8

19
VARCHAR2
string
"varchar2"
CLOB "clob"
RAW
hex string
"1F2F3F4F"
NUMBER
number
6
BINARY_DOUBLE 7
BINARY_FLOAT 8
DATE
ISO 8601
string
"2018-10-08T20:00:07"
TIMESTAMP "2018-10-08T20:00:07.795339"
TIMESTAMP WITH LOCAL TZ "2018-10-08T18:00:07.795339Z"
TIMESTAMP WITH TZ "2018-10-08T18:00:07.795339Z"
INTERVAL DAY TO SECOND "P13D"
INTERVAL YEAR TO MONTH "P1Y2M"
Converted to UTC

create table t (
C_VARCHAR2 VARCHAR2(16),
C_NVARCHAR2 NVARCHAR2(16),
C_CLOB CLOB,
C_RAW RAW(16),
C_BLOB BLOB,
C_NUMBER NUMBER(3,1),
C_BINARY_DOUBLE BINARY_DOUBLE,
C_BINARY_FLOAT BINARY_FLOAT,
C_DATE DATE,
C_TIMESTAMP TIMESTAMP,
C_TIMESTAMP_WITH_LOCAL_TIME_ZO TIMESTAMP WITH LOCAL TIME ZONE,
C_TIMESTAMP_WITH_TIME_ZONE TIMESTAMP WITH TIME ZONE,
C_INTERVAL_DAY_TO_SECOND INTERVAL DAY TO SECOND,
C_INTERVAL_YEAR_TO_MONTH INTERVAL YEAR TO MONTH
);
20
Table

select C_VARCHAR2||','||
C_NVARCHAR2||',’||
C_CLOB||','||
rawtohex(C_RAW)||',’||
<blob is complicated>||','||
to_char(C_NUMBER, 'TM', 'nls_numeric_characters=''. ''')||','||
to_char(C_BINARY_DOUBLE, 'TM', 'nls_numeric_characters=''. ''')||','||
to_char(C_BINARY_FLOAT, 'TM', 'nls_numeric_characters=''. ''')||','||
to_char(C_DATE, 'yyyy-mm-dd"T"hh24:mi:ss')||','||
to_char(C_TIMESTAMP, 'yyyy-mm-dd"T"hh24:mi:ssxff')||','||
to_char(C_TIMESTAMP_WITH_LOCAL_TIME_ZO, 'yyyy-mm-dd"T"hh24:mi:ssxff')||','||
to_char(C_TIMESTAMP_WITH_TIME_ZONE, 'yyyy-mm-
dd"T"hh24:mi:ssxffTZH:TZM')||','||
C_INTERVAL_DAY_TO_SECOND||','||
C_INTERVAL_YEAR_TO_MONTH
csv_text from T;
21
CSV

select xmlforest(
C_VARCHAR2,
C_NVARCHAR2,
C_CLOB,
rawtohex(C_RAW),
<blob is complicated>,
to_char(C_NUMBER, 'TM', 'nls_numeric_characters=''. ''') as C_NUMBER,
to_char(C_BINARY_DOUBLE, 'TM', 'nls_numeric_characters=''. ''') as
C_BINARY_DOUBLE,
to_char(C_BINARY_FLOAT, 'TM', 'nls_numeric_characters=''. ''') as C_BINARY_FLOAT,
to_char(C_DATE, 'yyyy-mm-dd"T"hh24:mi:ss') as C_DATE,
to_char(C_TIMESTAMP, 'yyyy-mm-dd"T"hh24:mi:ssxff', 'nls_numeric_characters=''.
''') as C_TIMESTAMP,
to_char(to_timestamp_tz(C_TIMESTAMP_WITH_LOCAL_TIME_ZO), 'yyyy-mm-
dd"T"hh24:mi:ssxffTZH:TZM', 'nls_numeric_characters=''. ''') as
C_TIMESTAMP_WITH_LOCAL_TIME_ZO,
to_char(C_TIMESTAMP_WITH_TIME_ZONE, 'yyyy-mm-dd"T"hh24:mi:ssxffTZH:TZM',
'nls_numeric_characters=''. ''') as C_TIMESTAMP_WITH_TIME_ZONE,
C_INTERVAL_DAY_TO_SECOND,
) T_XML
from T; 22
XML

select json_array(
C_VARCHAR2,
C_NVARCHAR2,
C_CLOB,
C_RAW,
C_BLOB,
C_NUMBER,
C_BINARY_DOUBLE,
C_BINARY_FLOAT,
C_DATE,
C_TIMESTAMP,
C_TIMESTAMP_WITH_TIME_ZONE,
) sql_to_json
from T;
[
"varchar2",
"nvarchar2",
"clob",
"1F2F3F4F",
"1F2F3F4F5F6F7F8F",
6,
7,
8,
"2018-10-08T20:00:07",
"2018-10-08T20:00:07.795339",
"2018-10-08T18:00:07.795339Z",
"2018-10-08T18:00:07.795339Z",
"P13D",
"P1Y2M"
]
23
What if C_BINARY_FLOAT is null?

select json_array(
C_VARCHAR2,
C_NVARCHAR2,
C_CLOB,
C_RAW,
C_BLOB,
C_NUMBER,
C_BINARY_DOUBLE,
C_BINARY_FLOAT, -- disappeared 😧
C_DATE,
C_TIMESTAMP,
) sql_to_json
from T;
[
"varchar2",
"nvarchar2",
"clob",
"1F2F3F4F",
"1F2F3F4F5F6F7F8F",
6,
7,
"2018-10-08T20:00:07",
"2018-10-08T20:00:07.795339",
"2018-10-08T18:00:07.795339Z",
"2018-10-08T18:00:07.795339Z",
"P13D",
"P1Y2M"
]
24
Why?

select json_array(
C_VARCHAR2,
C_NVARCHAR2,
C_CLOB,
C_RAW,
C_BLOB,
C_NUMBER,
C_BINARY_DOUBLE,
C_BINARY_FLOAT,
C_DATE,
C_TIMESTAMP,
ABSENT ON NULL) sql_to_json
from T;
[
"varchar2",
"nvarchar2",
"clob",
"1F2F3F4F",
"1F2F3F4F5F6F7F8F",
6,
7,
"2018-10-08T20:00:07",
"2018-10-08T20:00:07.795339",
"2018-10-08T18:00:07.795339Z",
"2018-10-08T18:00:07.795339Z",
"P13D",
"P1Y2M"
]
25
ABSENT ON NULL is the default

select json_array(
C_VARCHAR2,
C_NVARCHAR2,
C_CLOB,
C_RAW,
C_BLOB,
C_NUMBER,
C_BINARY_DOUBLE,
C_BINARY_FLOAT, -- it's back 🤗
C_DATE,
C_TIMESTAMP,
NULL ON NULL) sql_to_json
from T;
[
"varchar2",
"nvarchar2",
"clob",
"1F2F3F4F",
"1F2F3F4F5F6F7F8F",
6,
7,
null,
"2018-10-08T20:00:07",
"2018-10-08T20:00:07.795339",
"2018-10-08T18:00:07.795339Z",
"2018-10-08T18:00:07.795339Z",
"P13D",
"P1Y2M"
]
26
Always add this for arrays.

27
select json_object( {
key 'empno' value empno, "empno" : 7788,
'ename' value ename, "ename" : "SCOTT",
'job' : job, -- 19c "job" : "ANALYST",
DEPTNO, -- 19c "DEPTNO" : 20,
sal, -- 19c "sal" : 3000,
comm -- 19c "comm" : null ← NULL ON NULL
) jo }
from scott.emp
where ename = 'SCOTT';
Select json_object(*) -- 19c
from scott.emp
where ename = 'SCOTT';

JSON to SQL Data Type
28
Original SQL data type JSON Target SQL data type
VARCHAR2
string
VARCHAR2
NVARCHAR2
CLOB CLOB
RAW
(hex) string
BLOB
NUMBER
number
NUMBER
BINARY_DOUBLE
BINARY_FLOAT
DATE
ISO 8601
string
DATE
TIMESTAMP TIMESTAMP
TIMESTAMP WITH LOCAL TZ
TIMESTAMP WITH TZ TIMESTAMP WITH TZ
INTERVAL DAY TO SECOND
INTERVAL YEAR TO MONTH
These are the
data types that
can be returned
by JSON_VALUE
and JSON_TABLE.

select jt.* from json_data,
json_table(sql_to_json, '$' columns (
"Varchar2" VARCHAR2(16 BYTE) path '$[0]',
"Clob" CLOB path '$[2]',
"N" NUMBER path '$[5]',
"Date" DATE path '$[8]',
"Timestamp" TIMESTAMP path '$[9]',
"Timestamp with time zone" TIMESTAMP WITH TIME ZONE path '$[11]'
)) jt;
29
Varchar2 ClobN Date Timestamp Timestamp with time zone
varchar2 clob 6 2018-10-08 00:00 2018-10-08 23:26:22,67459 2018-10-08 21:26:22,67459 GMT
Time portion lost
during query
Time zone lost
during generation
<= if 18c, NLS_NUMERIC_CHARACTERS='.,’
Round trip

JSON to SQL Data Type
Input data types JSON result Return type For explicit conversion
VARCHAR2
string
VARCHAR2
NVARCHAR2 VARCHAR2 to_nchar()
CLOB CLOB
RAW
hex string
VARCHAR2(nn CHAR) hextoraw()
BLOB CLOB must write function
NUMBER
number
NUMBER
BINARY_DOUBLE NUMBER to_binary_double()
BINARY_FLOAT NUMBER to_binary_float()
DATE
ISO 8601
string
TIMESTAMP cast(<value> as date)
TIMESTAMP TIMESTAMP
TIMESTAMP WITH LOCAL TZ TIMESTAMP WITH TZ cast(<value> as timestamp with local TZ)
TIMESTAMP WITH TZ TIMESTAMP WITH TZ
INTERVAL DAY TO SECOND VARCHAR2(16 BYTE) to_dsinterval()
INTERVAL YEAR TO MONTH VARCHAR2(16 BYTE) to_yminterval()
30

Hex_to_blob function
create or replace FUNCTION hex_to_blob (hex CLOB)
RETURN BLOB AUTHID CURRENT_USER IS
b BLOB := NULL;
s VARCHAR2(32766 BYTE) := NULL;
l NUMBER := 32766;
BEGIN
if hex is not null then
dbms_lob.createtemporary(b, FALSE);
FOR i IN 0 .. floor(LENGTH(hex) / l) LOOP
dbms_lob.read(hex, l, i * l + 1, s);
dbms_lob.append(b, hextoraw(s));
END LOOP;
end if;
RETURN b;
END hex_to_blob;
31

So Far so Good
• Oracle can generate JSON from all scalar data types
(within reason).
• Oracle (with our help) can extract all scalar data types
from JSON.
• 19c: support for SQL objects and collections
• Data can go from SQL to JSON to SQL without loss
– But for time zones (and careful with time part of date)
• Note: except for "proof of concept", do not use JSON to
exchange data between Oracle DBs.
32

ETL: external table, 1 array per row
Extract
Select
json_array(<columns>)
from <table>;
(write to file)
Load
select <converted columns>
from <external table>,
json_table(json_data, '$'
columns (
a varchar2(99) path '$[0]',
b number path '$[1]',
...
));
33
• Not "standard" but scalable
• Must agree on
what a newline is

ETL: CLOB & array of arrays
Extract
select json_arrayagg(
json_array(<columns>)
returning clob
)
returning clob
from <table>;
(write clob to file)
Load
select <converted columns>
from json_table(
bfilename('<dir>', '<filename>'),
'$[*]'
columns (
a varchar2(99) path '$[0]',
b varchar2(99) path '$[1]', ...
));
34
Standard, but memory hog

Just Join?
with m as (
SELECT employee_id manager_id,
last_name
FROM employees
WHERE employee_id = 103
)
select MANAGER_ID, m.LAST_NAME, d.DEPARTMENT_ID,
d.DEPARTMENT_NAME, e.EMPLOYEE_ID, e.FIRST_NAME
from m
join departments d using(manager_id)
join employees e using(manager_id);
36

37
MANAGER_ID LAST_NAME DEPARTMENT_ID DEPARTMENT_NAME EMPLOYEE_ID FIRST_NAME
103Hunold 10Administration 104Bruce
103Hunold 20Marketing 104Bruce
103Hunold 30Purchasing 104Bruce
103Hunold 60IT 104Bruce
103Hunold 10Administration 105David
103Hunold 20Marketing 105David
103Hunold 30Purchasing 105David
103Hunold 60IT 105David
103Hunold 10Administration 106Valli
103Hunold 20Marketing 106Valli
103Hunold 30Purchasing 106Valli
103Hunold 60IT 106Valli
103Hunold 10Administration 107Diana
103Hunold 20Marketing 107Diana
103Hunold 30Purchasing 107Diana
103Hunold 60IT 107Diana

with m as (
SELECT employee_id manager_id,
last_name
FROM employees
WHERE employee_id = 103
)
SELECT json_object(
'manager_id' value manager_id,
'last_name' value last_name,
'departments' value ( ? ),
'employees' value ( ? ) )
) json_data FROM m;
38
SELECT JSON_ARRAYAGG(
JSON_OBJECT(
'department_id' value department_id,
'department_name'
value department_name
)
)
FROM departments
WHERE manager_id = m.manager_id
SELECT JSON_ARRAYAGG(
JSON_OBJECT(
'employee_id' value employee_id,
'first_name' value first_name
)
)
FROM employees
WHERE manager_id = m.manager_id

{
"manager_id" : 103,
"last_name" : "Hunold",
"departments" :
[
{
"department_id" : 10,
"department_name" : "Administration"
},
...
{
"department_id" : 60,
"department_name" : "IT"
}
],
"employees" :
[
{
"employee_id" : 104,
"first_name" : "Bruce"
},
...
{
"employee_id" : 107,
"first_name" : "Diana"
}
]
}
39
Now let's go from JSON to SQL

Just the Scalar values
select j.* from mgr_json,
json_table(json_data, '$' columns
manager_id, last_name
) j;
40
MANAGER_ID LAST_NAME
103 Hunold

Nested path: parent / child
manager_id, last_name,
nested path '$.departments[*]' columns (
department_id, department_name
)
) j;
41
MANAGER_ID LAST_NAME DEPARTMENT_ID DEPARTMENT_NAME
103 Hunold 10 Administration
103 Hunold 20 Marketing
103 Hunold 30 Purchasing
103 Hunold 60 IT

Nested path: Ordinality
manager_id, last_name,
d_ord for ordinality,
department_id, department_name
)
) j;
42
MANAGER_ID LAST_NAME D_ORD DEPARTMENT_ID DEPARTMENT_NAME
103 Hunold 110 Administration
103 Hunold 220 Marketing
103 Hunold 330 Purchasing
103 Hunold 460 IT

Nested paths: Siblings
manager_id number, last_name,
d_ord for ordinality,
department_id number, department_name
),
nested path '$.employees[*]' columns (
e_ord for ordinality,
employee_id number, first_name
)
ABSENT ON NULL) j;
43

Nested paths: Siblings
MANAGER_ID LAST_NAME D_ORD DEPT_ID DEPARTMENT_NAME E_ORD EMP_ID FIRST_NAME
103Hunold 1 10Administration
103Hunold 2 20Marketing
103Hunold 3 30Purchasing
103Hunold 4 60IT
103Hunold 1 104Bruce
103Hunold 2 105David
103Hunold 3 106Valli
103Hunold 4 107Diana
44

Advantages for OLTP
• Out with ORM, in with database development
– Provide application-specific APIs that add value
– Use SQL*Net or REST
– Data access layer and DB are less tightly coupled
– Generate / interpret complex hierarchical data
• Read consistency for all data
– One SELECT per API
– (there are other ways…)
45

JSON in Database 18c/19c
JSON as it was meant to be
Stew Ashton
OUG Ireland 2019
Blog: stewashton.wordpress.com
Twitter: @stewashton

If we had time…
• ETL: multi-table insert with nested JSON
• OLTP modifications
– JSON_MERGEPATCH (19c)
• String manipulation
– Splitting strings
stewashton.wordpress.com/splitting-strings-a-new-champion/
– Listagg() for CLOBs
47

JSON in 18c and 19c

Recommended

More Related Content

What's hot (19)

Similar to JSON in 18c and 19c (20)

More from stewashton (6)

Recently uploaded (20)

JSON in 18c and 19c