Skip to content

Latest commit

 

History

History
154 lines (105 loc) · 5 KB

README.md

File metadata and controls

154 lines (105 loc) · 5 KB
Raw

postgresql-replicant

A PostgreSQL streaming logical replication client for Haskell.

React to database changes in real-time.

This library is currently EXPERIMENTAL and is not ready for production yet. Developers are encouraged to test out the library, add issues, contribute and make it awesome.

For more, read the find documentation

Setup

In order to use a logical replication client you need to set the wal_level in Postgres to logical. You can set that, for example, in your postgres.conf file:

wal_level = logical

Then restart your server, log in as postgres or another super user and check:

SHOW wal_level

It should show logical.

You will also need a user who is allowed to use replication features.

Then add a database and your user with the REPLICATION trait to the database, grant all on it if you need to.

We haven't added authorization support yet to the library so make sure in pg_hba.conf you trust local connections:

host    all             all             127.0.0.1/32            trust

Caveat emptor.

You will also need to install the wal2json plugin for your PostgreSQL server. For example in Ubuntu-like distributions you can:

$ sudo apt install postgresql-12-wal2json

Assuming you have installed PostgreSQL from the package repositories.

You can then setup a basic "hello world" program using this library:

module Main where

import Control.Exception

import Database.PostgreSQL.Replicant

main :: IO ()
main = do
  let settings = PgSettings "my-user" (Just "my-password") "my-database" "localhost" "5432" "testing"
  withLogicalStream settings $ \change -> do
    print change
    return $ changeNextLSN change
      `catch` \err ->
      print err

Which should connect, create a testing replication slot, and start sending your callback the changes to your database as they arrive.

The type of the callback to withLogicalStream is:

Change -> IO LSN

Note the return type. The LSN or Log Sequence Number is used by replicant after running your callback to update its internal stream state and tell PostgreSQL that you have consumed this change. This means that if the connection fails and replicant reconnects to the same slot it will restart the stream at the last LSN replicant was able to successfully send to the server.

All you have to do is return changeNextLSN change at the end of your callback.

WAL Output Plugins

Presently we only support the wal2json WAL output plugin and only the v1 format. Support for the v2 format is planned as are more output plugins.

Example Program

Included is a simple program, replicant-example which should help you test your connection settings and make sure that you can connect and receive WAL changes.

You can change the connection settings through environment variables:

PG_USER=myuser PG_DATABASE=mydb replicant-example

The configuration settings are:

  • PG_USER: The replication user to connect as
  • PG_DATABASE: The database to connect to
  • PG_HOST: The host name of the database to connect to
  • PG_PORT: The port to connect on
  • PG_SLOTNAME: The replication slot to create or connect to
  • PG_UPDATEDELAY: The frequency (in ms) to update PostgreSQL on the stream status

Profiling

To enable profile builds, see stack.yaml

Build stack build

Then run stack exec --profile -- replicant-example +RTS -p (make sure to set the right database configs)

Run bin/bench.sh my-database create if this is the first time running the bench. This will create a table called, replicant_profile.

Run bin/bench.sh my-database, wait a bit, cancel it. It will insert a bunch of rows into replicant_profile. Feel free to run other SQL statements that modify the table while the bench is running. Cancel the script when you're done.

Then you can generate reports, see the GHC Manual on Profiling

FAQ

Why not use LISTEN/NOTIFY

You don't have to set up triggers for each table with logical replication. You also don't have the message size limitation. That limitation often forces clients to perform a query for each message to fetch the data and can use more than one connection. We only use one connection.

What are the trade offs?

This library uses logical replication slots. If your program loses the connection to the database and cannot reconnect the server will hold onto WAL files until your program can reconnect and resume the replication stream. On a busy database, depending on your configuration, this can eat up disk space quickly.

It's a good idea to monitor your replication slots. If postgresql-replicant cannot reconnect to the slot and resume the stream you may need to have a strategy for dropping the slot and recreating it once conditions return to normal. Have a strategy.