April 28, 2024
Summary: wal2json is a JSON output plugin for changeset extraction in PostgreSQL.
Table of Contents
Introduction
wal2json is an output plugin for logical decoding. It means that the plugin have access to tuples produced by INSERT and UPDATE. Also, UPDATE/DELETE old row versions can be accessed depending on the configured replica identity. Changes can be consumed using the streaming protocol (logical replication slots) or by a special SQL API.
format version 1 produces a JSON object per transaction. All of the new/old tuples are available in the JSON object. Also, there are options to include properties such as transaction timestamp, schema-qualified, data types, and transaction ids.
format version 2 produces a JSON object per tuple. Optional JSON object for beginning and end of transaction. Also, there are a variety of options to include properties.
Configuration
postgresql.conf
You need to set up the wal_level parameter at postgresql.conf:
wal_level = logical
After changing this parameter, a restart is needed.
Parameters
include-xids: add xid to each changeset. Default is false.include-timestamp: add timestamp to each changeset. Default is false.include-schemas: add schema to each change. Default is true.include-types: add type to each change. Default is true.include-typmod: add modifier to types that have it (eg. varchar(20) instead of varchar). Default is true.include-type-oids: add type oids. Default is false.include-domain-data-type: replace domain name with the underlying data type. Default is false.include-column-positions: add column position (pg_attribute.attnum). Default is false.include-origin: add origin of a piece of data. Default is false.include-not-null: add not null information as columnoptionals. Default is false.include-default: add default expression. Default is false.include-pk: add primary key information as pk. Column name and data type is included. Default is false.numeric-data-types-as-string: use string for numeric data types. JSON specification does not recognizeInfinityandNaNas valid numeric values. There might be potential interoperability problems for double precision numbers. Default is false.pretty-print: add spaces and indentation to JSON structures. Default is false.write-in-chunks: write after every change instead of every changeset. Only used whenformat-versionis1. Default is false.include-lsn: add nextlsn to each changeset. Default is false.include-transaction: emit records denoting the start and end of each transaction. Default is true.include-unchanged-toast(deprecated): Don’t use it. It is deprecated.filter-origins: exclude changes from the specified origins. Default is empty which means that no origin will be filtered. It is a comma separated value.filter-tables: exclude rows from the specified tables. Default is empty which means that no table will be filtered. It is a comma separated value. The tables should be schema-qualified.*.foomeans table foo in all schemas andbar.*means all tables in schema bar. Special characters (space, single quote, comma, period, asterisk) must be escaped with backslash. Schema and table are case-sensitive. Table"public"."Foo bar"should be specified aspublic.Foo\ bar.add-tables: include only rows from the specified tables. Default is all tables from all schemas. It has the same rules fromfilter-tables.filter-msg-prefixes: exclude messages if prefix is in the list. Default is empty which means that no message will be filtered. It is a comma separated value.add-msg-prefixes: include only messages if prefix is in the list. Default is all prefixes. It is a comma separated value.wal2jsonappliesfilter-msg-prefixesbefore this parameter.format-version: defines which format to use. Default is 1.actions: define which operations will be sent. Default is all actions (insert, update, delete, and truncate). However, if you are usingformat-version1, truncate is not enabled (backward compatibility).
Examples
There are two ways to obtain the changes (JSON objects) from wal2json plugin: calling functions via SQL or pg_recvlogical.
pg_recvlogical
Besides the configuration above, it is necessary to configure a replication connection to use pg_recvlogical. A logical replication connection in version 9.4, 9.5, and 9.6 requires replication keyword in the database column. Since version 10, logical replication matches a normal entry with a database name or keywords such as all.
First, add a replication connection rule at pg_hba.conf (9.4, 9.5, and 9.6):
local replication myuser trust
If you are using version 10 or later:
local mydatabase myuser trust
Also, set max_wal_senders at postgresql.conf:
max_wal_senders = 1
A restart is necessary if you changed max_wal_senders.
You are ready to try wal2json. In one terminal:
$ pg_recvlogical -d postgres --slot test_slot --create-slot -P wal2json
$ pg_recvlogical -d postgres --slot test_slot --start -o pretty-print=1 -o add-msg-prefixes=wal2json -f -
In another terminal:
$ cat /tmp/example1.sql
CREATE TABLE table1_with_pk (a SERIAL, b VARCHAR(30), c TIMESTAMP NOT NULL, PRIMARY KEY(a, c));
CREATE TABLE table1_without_pk (a SERIAL, b NUMERIC(5,2), c TEXT);
BEGIN;
INSERT INTO table1_with_pk (b, c) VALUES('Backup and Restore', now());
INSERT INTO table1_with_pk (b, c) VALUES('Tuning', now());
INSERT INTO table1_with_pk (b, c) VALUES('Replication', now());
SELECT pg_logical_emit_message(true, 'wal2json', 'this message will be delivered');
SELECT pg_logical_emit_message(true, 'pgoutput', 'this message will be filtered');
DELETE FROM table1_with_pk WHERE a < 3;
SELECT pg_logical_emit_message(false, 'wal2json', 'this non-transactional message will be delivered even if you rollback the transaction');
INSERT INTO table1_without_pk (b, c) VALUES(2.34, 'Tapir');
-- it is not added to stream because there isn't a pk or a replica identity
UPDATE table1_without_pk SET c = 'Anta' WHERE c = 'Tapir';
COMMIT;
DROP TABLE table1_with_pk;
DROP TABLE table1_without_pk;
The script above produces the output below:
$ psql -At -f /tmp/example1.sql postgres
CREATE TABLE
CREATE TABLE
BEGIN
INSERT 0 1
INSERT 0 1
INSERT 0 1
3/78BFC828
3/78BFC880
DELETE 2
3/78BFC990
INSERT 0 1
UPDATE 1
COMMIT
DROP TABLE
DROP TABLE
The output in the first terminal is:
{
"change": [
]
}
{
"change": [
]
}
{
"change": [
{
"kind": "message",
"transactional": false,
"prefix": "wal2json",
"content": "this non-transactional message will be delivered even if you rollback the transaction"
}
]
}
WARNING: table "table1_without_pk" without primary key or replica identity is nothing
{
"change": [
{
"kind": "insert",
"schema": "public",
"table": "table1_with_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "character varying(30)", "timestamp without time zone"],
"columnvalues": [1, "Backup and Restore", "2018-03-27 11:58:28.988414"]
}
,{
"kind": "insert",
"schema": "public",
"table": "table1_with_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "character varying(30)", "timestamp without time zone"],
"columnvalues": [2, "Tuning", "2018-03-27 11:58:28.988414"]
}
,{
"kind": "insert",
"schema": "public",
"table": "table1_with_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "character varying(30)", "timestamp without time zone"],
"columnvalues": [3, "Replication", "2018-03-27 11:58:28.988414"]
}
,{
"kind": "message",
"transactional": true,
"prefix": "wal2json",
"content": "this message will be delivered"
}
,{
"kind": "delete",
"schema": "public",
"table": "table1_with_pk",
"oldkeys": {
"keynames": ["a", "c"],
"keytypes": ["integer", "timestamp without time zone"],
"keyvalues": [1, "2018-03-27 11:58:28.988414"]
}
}
,{
"kind": "delete",
"schema": "public",
"table": "table1_with_pk",
"oldkeys": {
"keynames": ["a", "c"],
"keytypes": ["integer", "timestamp without time zone"],
"keyvalues": [2, "2018-03-27 11:58:28.988414"]
}
}
,{
"kind": "insert",
"schema": "public",
"table": "table1_without_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "numeric(5,2)", "text"],
"columnvalues": [1, 2.34, "Tapir"]
}
]
}
{
"change": [
]
}
{
"change": [
]
}
Dropping the slot in the first terminal:
Ctrl+C
$ pg_recvlogical -d postgres --slot test_slot --drop-slot
SQL functions
$ cat /tmp/example2.sql
CREATE TABLE table2_with_pk (a SERIAL, b VARCHAR(30), c TIMESTAMP NOT NULL, PRIMARY KEY(a, c));
CREATE TABLE table2_without_pk (a SERIAL, b NUMERIC(5,2), c TEXT);
SELECT 'init' FROM pg_create_logical_replication_slot('test_slot', 'wal2json');
BEGIN;
INSERT INTO table2_with_pk (b, c) VALUES('Backup and Restore', now());
INSERT INTO table2_with_pk (b, c) VALUES('Tuning', now());
INSERT INTO table2_with_pk (b, c) VALUES('Replication', now());
SELECT pg_logical_emit_message(true, 'wal2json', 'this message will be delivered');
SELECT pg_logical_emit_message(true, 'pgoutput', 'this message will be filtered');
DELETE FROM table2_with_pk WHERE a < 3;
SELECT pg_logical_emit_message(false, 'wal2json', 'this non-transactional message will be delivered even if you rollback the transaction');
INSERT INTO table2_without_pk (b, c) VALUES(2.34, 'Tapir');
-- it is not added to stream because there isn't a pk or a replica identity
UPDATE table2_without_pk SET c = 'Anta' WHERE c = 'Tapir';
COMMIT;
SELECT data FROM pg_logical_slot_get_changes('test_slot', NULL, NULL, 'pretty-print', '1', 'add-msg-prefixes', 'wal2json');
SELECT 'stop' FROM pg_drop_replication_slot('test_slot');
DROP TABLE table2_with_pk;
DROP TABLE table2_without_pk;
The script above produces the output below:
$ psql -At -f /tmp/example2.sql postgres
CREATE TABLE
CREATE TABLE
init
BEGIN
INSERT 0 1
INSERT 0 1
INSERT 0 1
3/78C2CA50
3/78C2CAA8
DELETE 2
3/78C2CBD8
INSERT 0 1
UPDATE 1
COMMIT
{
"change": [
{
"kind": "message",
"transactional": false,
"prefix": "wal2json",
"content": "this non-transactional message will be delivered even if you rollback the transaction"
}
]
}
psql:/tmp/example2.sql:17: WARNING: table "table2_without_pk" without primary key or replica identity is nothing
{
"change": [
{
"kind": "insert",
"schema": "public",
"table": "table2_with_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "character varying(30)", "timestamp without time zone"],
"columnvalues": [1, "Backup and Restore", "2018-03-27 12:05:29.914496"]
}
,{
"kind": "insert",
"schema": "public",
"table": "table2_with_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "character varying(30)", "timestamp without time zone"],
"columnvalues": [2, "Tuning", "2018-03-27 12:05:29.914496"]
}
,{
"kind": "insert",
"schema": "public",
"table": "table2_with_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "character varying(30)", "timestamp without time zone"],
"columnvalues": [3, "Replication", "2018-03-27 12:05:29.914496"]
}
,{
"kind": "message",
"transactional": true,
"prefix": "wal2json",
"content": "this message will be delivered"
}
,{
"kind": "delete",
"schema": "public",
"table": "table2_with_pk",
"oldkeys": {
"keynames": ["a", "c"],
"keytypes": ["integer", "timestamp without time zone"],
"keyvalues": [1, "2018-03-27 12:05:29.914496"]
}
}
,{
"kind": "delete",
"schema": "public",
"table": "table2_with_pk",
"oldkeys": {
"keynames": ["a", "c"],
"keytypes": ["integer", "timestamp without time zone"],
"keyvalues": [2, "2018-03-27 12:05:29.914496"]
}
}
,{
"kind": "insert",
"schema": "public",
"table": "table2_without_pk",
"columnnames": ["a", "b", "c"],
"columntypes": ["integer", "numeric(5,2)", "text"],
"columnvalues": [1, 2.34, "Tapir"]
}
]
}
stop
DROP TABLE
DROP TABLE
Let’s repeat the same example with format-version 2:
$ cat /tmp/example3.sql
CREATE TABLE table3_with_pk (a SERIAL, b VARCHAR(30), c TIMESTAMP NOT NULL, PRIMARY KEY(a, c));
CREATE TABLE table3_without_pk (a SERIAL, b NUMERIC(5,2), c TEXT);
SELECT 'init' FROM pg_create_logical_replication_slot('test_slot', 'wal2json');
BEGIN;
INSERT INTO table3_with_pk (b, c) VALUES('Backup and Restore', now());
INSERT INTO table3_with_pk (b, c) VALUES('Tuning', now());
INSERT INTO table3_with_pk (b, c) VALUES('Replication', now());
SELECT pg_logical_emit_message(true, 'wal2json', 'this message will be delivered');
SELECT pg_logical_emit_message(true, 'pgoutput', 'this message will be filtered');
DELETE FROM table3_with_pk WHERE a < 3;
SELECT pg_logical_emit_message(false, 'wal2json', 'this non-transactional message will be delivered even if you rollback the transaction');
INSERT INTO table3_without_pk (b, c) VALUES(2.34, 'Tapir');
-- it is not added to stream because there isn't a pk or a replica identity
UPDATE table3_without_pk SET c = 'Anta' WHERE c = 'Tapir';
COMMIT;
SELECT data FROM pg_logical_slot_get_changes('test_slot', NULL, NULL, 'format-version', '2', 'add-msg-prefixes', 'wal2json');
SELECT 'stop' FROM pg_drop_replication_slot('test_slot');
DROP TABLE table3_with_pk;
DROP TABLE table3_without_pk;
The script above produces the output below:
$ psql -At -f /tmp/example3.sql postgres
CREATE TABLE
CREATE TABLE
init
BEGIN
INSERT 0 1
INSERT 0 1
INSERT 0 1
3/78CB8F30
3/78CB8F88
DELETE 2
3/78CB90B8
INSERT 0 1
UPDATE 1
COMMIT
psql:/tmp/example3.sql:20: WARNING: no tuple identifier for UPDATE in table "public"."table3_without_pk"
{"action":"M","transactional":false,"prefix":"wal2json","content":"this non-transactional message will be delivered even if you rollback the transaction"}
{"action":"B"}
{"action":"I","schema":"public","table":"table3_with_pk","columns":[{"name":"a","type":"integer","value":1},{"name":"b","type":"character varying(30)","value":"Backup and Restore"},{"name":"c","type":"timestamp without time zone","value":"2019-12-29 04:58:34.806671"}]}
{"action":"I","schema":"public","table":"table3_with_pk","columns":[{"name":"a","type":"integer","value":2},{"name":"b","type":"character varying(30)","value":"Tuning"},{"name":"c","type":"timestamp without time zone","value":"2019-12-29 04:58:34.806671"}]}
{"action":"I","schema":"public","table":"table3_with_pk","columns":[{"name":"a","type":"integer","value":3},{"name":"b","type":"character varying(30)","value":"Replication"},{"name":"c","type":"timestamp without time zone","value":"2019-12-29 04:58:34.806671"}]}
{"action":"M","transactional":true,"prefix":"wal2json","content":"this message will be delivered"}
{"action":"D","schema":"public","table":"table3_with_pk","identity":[{"name":"a","type":"integer","value":1},{"name":"c","type":"timestamp without time zone","value":"2019-12-29 04:58:34.806671"}]}
{"action":"D","schema":"public","table":"table3_with_pk","identity":[{"name":"a","type":"integer","value":2},{"name":"c","type":"timestamp without time zone","value":"2019-12-29 04:58:34.806671"}]}
{"action":"I","schema":"public","table":"table3_without_pk","columns":[{"name":"a","type":"integer","value":1},{"name":"b","type":"numeric(5,2)","value":2.34},{"name":"c","type":"text","value":"Tapir"}]}
{"action":"C"}
stop
DROP TABLE
DROP TABLE