I worked on this code as a PoC 4 years ago. I have not actively maintained it and I've moved on to other projects. You're still free to use it, but keep this in mind. If you're looking for something more actively maintained check out Debezium's fork.
A PostgreSQL logical decoder output plugin to deliver data as Protocol Buffers
Version: 0.2.0
decoderbufs is a PostgreSQL logical decoder output plugin to deliver data as Protocol Buffers.
decoderbufs is released under the MIT license (See LICENSE file).
Shoutouts:
- The PostgreSQL Team for adding logical decoding support. This is a immensely useful feature.
- Michael Paquier for his decoder_raw project and blog posts as a guide to teach myself how to write a PostgreSQL logical decoder output plugin.
- Martin Kleppmann for making me aware that PostgreSQL was working on logical decoding.
- Magnus Edenhill for letting me know that protobuf-c existed and just in general for writing librdkafka.
This code is built with the following assumptions. You may get mixed results if you deviate from these versions.
- PostgreSQL 9.4+
- Protocol Buffers 3.x
- protobuf-c 1.2.x
- PostGIS 2.1.x
- PostgreSQL
- PostGIS
- Protocol Buffers
- protobuf-c
To build you will need to install PostgreSQL (for pg_config) and PostgreSQL server development packages. On Debian based distributions you can usually do something like this:
apt-get install -y postgresql postgresql-server-dev-9.6
You will also need to make sure that protobuf-c and it's header files have been installed. See their Github page for further details.
If you have all of the prerequisites installed you should be able to just:
make && make install
Once the extension has been installed you just need to enable it and logical replication in postgresql.conf:
# MODULES
shared_preload_libraries = 'decoderbufs'
# REPLICATION
wal_level = logical # minimal, archive, hot_standby, or logical (change requires restart)
max_wal_senders = 8 # max number of walsender processes (change requires restart)
wal_keep_segments = 4 # in logfile segments, 16MB each; 0 disables
#wal_sender_timeout = 60s # in milliseconds; 0 disables
max_replication_slots = 4 # max number of replication slots (change requires restart)
In addition, permissions will have to be added for the user that connects to the DB to be able to replicate. This can be modified in pg_hba.conf like so:
local replication <youruser> trust
host replication <youruser> 127.0.0.1/32 trust
host replication <youruser> ::1/128 trust
And restart PostgreSQL.
-- can use SQL for demo purposes
select * from pg_create_logical_replication_slot('decoderbufs_demo', 'decoderbufs');
-- DO SOME TABLE MODIFICATIONS (see below about UPDATE/DELETE)
-- peek at WAL changes using decoderbufs debug mode for SQL console
select data from pg_logical_slot_peek_changes('decoderbufs_demo', NULL, NULL, 'debug-mode', '1');
-- get WAL changes using decoderbufs to update the WAL position
select data from pg_logical_slot_get_changes('decoderbufs_demo', NULL, NULL, 'debug-mode', '1');
-- check the WAL position of logical replicators
select * from pg_replication_slots where slot_type = 'logical';
If you're performing an UPDATE/DELETE on your table and you don't see results for those operations from logical decoding, make sure you have set REPLICA IDENTITY appropriately for your use case.
For consuming the binary format, I have implemented a custom PostgreSQL logical replication client that publishes to Apache Kafka. I'm hoping to clean that up a bit and open source the project.
The following table shows how current PostgreSQL type OIDs are mapped to which decoderbuf fields:
| PostgreSQL Type OID | Decoderbuf Field |
|---|---|
| BOOLOID | datum_boolean |
| INT2OID | datum_int32 |
| INT4OID | datum_int32 |
| INT8OID | datum_int64 |
| OIDOID | datum_int64 |
| FLOAT4OID | datum_float |
| FLOAT8OID | datum_double |
| NUMERICOID | datum_double |
| CHAROID | datum_string |
| VARCHAROID | datum_string |
| BPCHAROID | datum_string |
| TEXTOID | datum_string |
| JSONOID | datum_string |
| XMLOID | datum_string |
| UUIDOID | datum_string |
| TIMESTAMPOID | datum_string |
| TIMESTAMPTZOID | datum_string |
| BYTEAOID | datum_bytes |
| POINTOID | datum_point |
| PostGIS geometry | datum_point |
| PostGIS geography | datum_point |
File bug reports, feature requests and questions using GitHub Issues
This approach is the one we wanted when we first started kicking around ideas for how to replicate our Postgres DBs in near-realtime. It should provide much better resiliency in the face of network outages and unplanned downtime than a push mechanism (like using pg_kafka with a trigger) would.
The PostgreSQL docs are pretty good and are definitely worth a read.
NOT ALL OID TYPES ARE SUPPORTED CURRENTLY. I really want to iterate this point. There are lots of OIDs in Postgres. Right now I'm translating the ones that I see used a lot, but it is by no means comprehensive. I hope to update this project to support most if not all types at some point.
