One of the most exciting enhancements that comes with the new PostgreSQL 10 release, is Logical Replication. The functionality, although not yet as extensive as pgLogical which it is based on, provides a powerful replication option when you want control over table backups at a finite level – allowing all kinds of ETL goodness like:

  • replicating only certain tables, columns, or rows
  • consolidating multiple databases into a single one
  • sharing a subset of the database between multiple databases
  • replicating between different major versions of PostgreSQL

Want to try it without the fuss?

Installing multiple PostgreSQL 10 sandboxes on your localhost in 4 easy steps

You want to try it but can’t face installation hell? The BigSQL distribution has your back. We make it easy.

  1. Make 2 directories:

    # the logical replication publisher
    mkdir pub_test     
    
    # the logical replication subscriber
    mkdir sub_test
  2. Navigate to your pub_test directory:

    cd pub_test
  3. Install the a postgresql 10 instance (Windows users don’t prefix the pgc command with ./ as shown in the following commands):

    # use this for MAC / Linux:
    python -c "$(curl -fsSL http://s3.amazonaws.com/pgcentral/install.py)" 
    
    # use this for Windows:
    @powershell -NoProfile -ExecutionPolicy unrestricted -Command "iex ((new-object net.webclient).DownloadString('http://s3.amazonaws.com/pgcentral/install.ps1'))"
    
    # run pgc commands to install pg10
    cd bigsql           
    ./pgc install pg10
    ./pgc start pg10
    
    # make sure the instance is running
    ./pgc status
  4. Repeat all of the commands in step 3 after navigating to your sub_test directory:

    cd sub_test

Take note that:

  • The first installation, pub_test, will default to port 5432.
  • The second installation, sub_test, will default to port 5433.

Edit postgresql.conf

For logical replication to work, you will first need to change the configuration parameter wal_level to ‘logical’.

So, you will need to edit the file postgresql.conf for BOTH of your PostgreSQL (pub_test AND sub_test). Navigate to the directory where postgresql.conf is located:

cd {pub_test,sub_test}/bigsql/data/pg10

Open the file with your favorite editor and change wal_level from ‘hot_standby’ to ‘logical’

#------------------------------------------------------------------------------
# WRITE AHEAD LOG
#------------------------------------------------------------------------------

# - Settings -

wal_level = logical         # (change requires restart)

Navigate to the bigsql directory and restart pg10:

cd ../..
./pgc restart pg10

Exercise: Rock Legends Database

For this exercise, we will create a rock legends database and setup logical replication on the rocker girls table.

Create the publisher

  1. Connect as a superuser to the PostgreSQL instance that will serve as the publisher:

    #publisher instance in running on port 5432 - the default       
    psql -h localhost -p 5432 -U postgres
  2. Create rock_legends_db database:

    postgres=# CREATE DATABASE rock_legends_db;
    postgres=# \c rock_legends_db;
        You are now connected to database "rock_legends_db" as user "postgres". 
  3. Create table rocker_girls:

    CREATE TABLE rocker_girls (
         name text PRIMARY KEY,
         band text NOT NULL,
         record_creation_date timestamptz NOT NULL DEFAULT now()
        );
  4. Add record to rocker_girls table:

    INSERT INTO rocker_girls(name, band) VALUES('Debbie Harry', 'Blondie');
  5. Check to see that rocker_girls table with data was created:

    rock_legends_db=# select * from rocker_girls; 
    
    name         | band    | record_creation_date          
    ------------ | ------- | ----------------------------- 
    Debbie Harry | Blondie | 2017-05-30 12:52:12.177873-04 
    (1 row)
  6. Setup the database as the publisher for the rocker_girls table:

    CREATE PUBLICATION pub_rock FOR TABLE rocker_girls;
  7. Exit psql:

    rock_legends_db=# \q

Create the subscriber

  1. Connect as a superuser to the PostgreSQL instance that will serve as the subscriber:

    #subscriber instance in running on port 5433 - the default      
    psql -h localhost -p 5433 -U postgres
  2. Repeat steps 2 – 3 from above. DO NOT add any records to the table!

  3. Setup the database as the subscriber:

    CREATE SUBSCRIPTION sub_rock 
    CONNECTION 'host=localhost port=5432 dbname=rock_legends_db' 
    PUBLICATION pub_rock;
  4. Verify that the rocker_girls “Debbie Harry” row was replicated:

    rock_legends_db=# select * from rocker_girls; 
    
    name         | band    | record_creation_date          
    ------------ | ------- | ----------------------------- 
    Debbie Harry | Blondie | 2017-05-30 12:52:12.177873-04 
    (1 row)

Debbie Harry has been replicated!

Test the replication by adding another record

  1. Connect to the publisher:

    psql -h localhost -p 5432 -U postgres -d rock_legends_db
  2. Add a new record:

    INSERT INTO rocker_girls(name, band) VALUES('Janis Joplin', 'Big Brother and the Holding Company');
  3. Disconnect from publisher and connect to subscriber:

    rock_legends_db=# \q
    psql -h localhost -p 5433 -U postgres -d rock_legends_db
  4. Run select all on rocker_girls table:

    rock_legends_db=# select * from rocker_girls;
    
    name         | band                                | record_creation_date          
    ------------ | ----------------------------------- | ----------------------------- 
    Debbie Harry | Blondie                             | 2017-05-30 12:52:12.177873-04 
    Janis Joplin | Big Brother and the Holding Company | 2017-05-31 08:22:57.708333-04
    (2 rows)

Janis has been replicated!

What’s next

In upcoming posts, I’ll show more ways to manage your replications. In the meantime, you can continue to test in your sandboxes. And in case you need to blow away your logical replication …

How to delete the subscriber and publisher

rock_legends_db=# drop subscription testsub;
NOTICE:  dropped replication slot "rock_legends_db" on publisher
DROP SUBSCRIPTION

rock_legends_db=# drop publication alltables;
DROP PUBLICATION