Generic PostgreSQL Setup Guide

Logical replication with the test_decoding plugin

IMPORTANT: You can only enable logical replication with the test_decoding plugin if your PostgreSQL server version is 9.4.15 or later. Prior minor versions have breaking bugs.

Logical replication is based on logical decoding of the PostgreSQL write-ahead log (WAL). Fivetran reads the WAL using the test_decoding plugin to detect any new or changed data.

Learn more in our logical replication documentation.

Logical replication with the pgoutput plugin Beta

IMPORTANT: You can only enable logical replication with the pgoutput plugin if your PostgresSQL server version is 10+. Prior minor versions have breaking bugs.

Logical replication is based on logical decoding of the PostgreSQL write-ahead log (WAL). Fivetran reads the WAL using the pgoutput plugin to detect any new or changed data.

Learn more in our logical replication documentation.

XMIN

The XMIN method is based on the hidden xmin system column that is present in all PostgreSQL tables. With XMIN, Fivetran must scan every table in full to detect updated data. We do not recommend XMIN for near real-time data needs because XMIN replication is slower than logical replication and doesn’t allow Fivetran to detect deleted rows.

Learn more in our XMIN documentation.

Connect directly

Fivetran connects directly to your database instance. This is the simplest and most secure method.

To connect directly, do the following to configure your firewall and/or other access control systems:

• Allow incoming connections to your PostgreSQL port (usually 5432) from Fivetran’s IPs for your database’s region.
• Allow outgoing connections from all ports (1024 to 65535) to Fivetran’s IPs for your database’s region.

How you do this will vary based on how your PostgreSQL database is hosted (cloud platform, on-premises, etc.).

Connect using SSH

Fivetran connects to a separate server in your network that provides an SSH tunnel to your database. You must connect through SSH if your database is in an inaccessible subnet.

To connect using SSH, follow our SSH connection instructions to give Fivetran access to your SSH tunnel.

Connect using AWS PrivateLink BETA

IMPORTANT: You must have a Business Critical plan to use AWS PrivateLink.

AWS PrivateLink allows VPCs and AWS-hosted or on-premises services to communicate with one another without exposing traffic to the public internet. Learn more in AWS’ PrivateLink documentation.

Follow our AWS PrivateLink setup guide to configure PrivateLink for your database.

Restrict access to tables (optional)

If you want to limit Fivetran’s access to your tables, grant the Fivetran user access to only the tables that you would like to sync. You must individually grant access for each table that you want to sync. You cannot grant access to all tables and then revoke access for a subset of tables.

1. Ensure that the Fivetran user has access to the schema that contains your table(s).

   GRANT USAGE ON SCHEMA "some_schema" TO <username>;
   

2. Revoke any previously granted permission to all tables in that schema.

   ALTER DEFAULT PRIVILEGES IN SCHEMA "some_schema" REVOKE SELECT ON TABLES FROM <username>;
   REVOKE SELECT ON ALL TABLES IN SCHEMA "some_schema" FROM <username>;
   

3. Repeat the following command for each table you want Fivetran to sync.

   GRANT SELECT ON "some_schema"."some_table" TO <username>;
   

4. By default, any tables that you create in the future will be excluded from the Fivetran user’s access. To grant access to new tables, run the following command.

   ALTER DEFAULT PRIVILEGES IN SCHEMA "some_schema" GRANT SELECT ON TABLES TO <username>;
   

Restrict access to columns (optional)

If you want to limit Fivetran’s access to the columns in your tables, grant the Fivetran user access to only certain columns. You must individually grant access for each column that you want to sync.

1. Revoke any previously granted permission to read all columns in the table.

   REVOKE SELECT ON "some_schema"."some_table" FROM <username>;
   

2. Grant permission to the specific columns you want to sync (for example, some_column and other_column).

   NOTE: If you chose XMIN as your incremental update mechanism, you must grant us access to the hidden system column xmin.

   GRANT SELECT (xmin, "some_column", "other_column") ON "some_schema"."some_table" TO <username>;
   

Once you restrict access to columns within a table, the Fivetran user will not have access to any new columns added to that table in the future. To grant access to new columns, you must rerun the command above.

Logical replication with the test_decoding plugin

To enable logical replication with the test_decoding plugin, follow these steps:

1. Go to your PostgreSQL database.

2. Ensure that your server has ample free space for the logs. As soon as Fivetran processes a log, we delete it. However, we don’t delete logs if the sync is interrupted (for example, if we lose access to your database). In this case, logs may accumulate on your server and consume additional storage. The amount of additional disk space that these logs consume is proportional to the number of changes committed on the server. If we can’t resume a lost connection quickly enough and you need more disk space, you can drop the replication slot, which deletes its unconsumed logs.

3. Ensure that the statement_timeout setting on your server is either 0 (the default value to disable the timeout) or greater than 5 minute.

4. Set the wal_level parameter in your database configuration to logical. For a standard PostgreSQL database, do this by adding a wal_level = logical line to the postgresql.conf1 file. Restart the server for this change to take effect.

5. Ensure that your max_replication_slots value is equal to or higher than the number of PostgreSQL connectors that use WAL plus the number of other replication slots your database uses.

6. Set the wal_sender_timeout parameter in your database configuration to 0 to disable the timeout.

7. Add a record to your pg_hba.conf file that allows your database to authenticate Fivetran’s connection to the WAL.

8. Go to the postgresql.conf file and ensure that the max_wal_senders parameter, which specifies the maximum number of concurrent connections to the WAL, is at least twice the total number of logical replication slots. For example, if your database uses 11 replication slots in total, then the max_wal_senders value must be 22 or greater.

9. Log in to a PostgreSQL console (such as a SQL workbench or psql) as a superuser.

10. Create a logical replication slot for the database you want to sync by running the following command. You must use the output plugin test_decoding supplied in the postgresql-contrib subpackage.

    IMPORTANT: The replication slot name fivetran_replication_slot quoted throughout this guide is used purely as an example. The actual replication slot name should be unique for every connector using the same PostgreSQL cluster. Replication slot names cannot start with a number.

    SELECT pg_create_logical_replication_slot('fivetran_replication_slot', 'test_decoding');
    

11. Grant permission to the Fivetran user for reading the replication slot.

    ALTER ROLE <username> WITH REPLICATION;
    

12. Log in as the Fivetran user.

13. Verify that the Fivetran user can read the replication slot by running the following command:

    SELECT count(*) FROM pg_logical_slot_peek_changes('fivetran_replication_slot', null, null);
    

    If the query succeeds, then permissions are sufficient.

IMPORTANT: You must periodically tune the checkpoint_timeout and max_wal_size parameters based on your PostgreSQL database operations. If you do not, you may experience replication failures. To learn how to tune, read this tuning checkpoints documentation.

Logical replication with the pgoutput plugin Beta

To enable logical replication with the pgoutput plugin, follow these steps:

1. Go to your PostgreSQL database.

2. Ensure that your server has ample free space for the logs. As soon as Fivetran processes a log, we delete it. However, we don’t delete logs if the sync is interrupted (for example, if we lose access to your database). In this case, logs may accumulate on your server and consume additional storage. The amount of additional disk space that these logs consume is proportional to the number of changes committed on the server. If we can’t resume a lost connection quickly enough and you need more disk space, you can drop the replication slot, which deletes its unconsumed logs.

3. Ensure that the statement_timeout setting on your server is either 0 (the default value to disable the timeout) or greater than 5 minute.

4. Set the wal_level parameter in your database configuration to logical. For a standard PostgreSQL database, do this by adding a wal_level = logical line to the postgresql.conf1 file. Restart the server for this change to take effect.

5. Ensure that your max_replication_slots value is equal to or higher than the number of PostgreSQL connectors that use WAL plus the number of other replication slots your database uses.

6. Set the wal_sender_timeout parameter in your database configuration to 0 to disable the timeout.

7. Add a record to your pg_hba.conf file that allows your database to authenticate Fivetran’s connection to the WAL.

8. Go to the postgresql.conf file and ensure that the max_wal_senders parameter, which specifies the maximum number of concurrent connections to the WAL, is at least twice the total number of logical replication slots. For example, if your database uses 11 replication slots in total, then the max_wal_senders value must be 22 or greater.

9. Log in to a PostgreSQL console (such as a SQL workbench or psql) as a superuser.

10. Create a logical replication slot for the database you want to sync by running the following command. You must use the standard output plugin pgoutput.

    IMPORTANT: The replication slot name fivetran_pgoutput_slot quoted throughout this guide is used purely as an example. The actual replication slot name should be unique for every connector using the same PostgreSQL cluster. Replication slot names cannot start with a number.

    SELECT pg_create_logical_replication_slot('fivetran_pgoutput_slot', 'pgoutput');
    

11. Create a publication for your tables. If you want, you can create a publication for only certain tables so that you add or remove tables from the publication later on. Only changes from tables in the publication are replicated to Fivetran. Each database can have multiple distinct publications. You must have CREATE privileges or above to run this command.

    IMPORTANT: The publication name fivetran_pub quoted throughout this guide is used purely as an example. The actual publication name should be unique for every database and cannot start with a number.

    CREATE PUBLICATION fivetran_pub FOR TABLE table2, table4, table8;
    

    To add or remove a table from a publication, run the following command. You must have ownership rights over the table(s).

    ALTER PUBLICATION fivetran_pub ADD/DROP TABLE table_name;
    

    Alternatively, you can create a publication for all of your tables. However, you cannot remove any table from this publication later on. You must have superuser privileges to run this command.

     CREATE PUBLICATION fivetran_pub FOR ALL TABLES;
    

    (Optional) You can choose which operations to include in the publication. For example, the following publication includes only INSERT and UPDATE operations.

    CREATE PUBLICATION insert_only_pub FOR TABLE table1 WITH (publish = 'INSERT, UPDATE');
    

12. Verify that your chosen tables are in the publication.

    SELECT * FROM pg_publication_tables.
    

13. Grant the Fivetran user permission to read the replication slot.

    ALTER ROLE <username> WITH REPLICATION;
    

14. Log in as the Fivetran user.

15. Verify that the Fivetran user can read the replication slot by running the following command. Replace fivetran_pgoutput_slot with your replication slot name and fivetran_pub with the publication name.

    SELECT count(*) FROM pg_logical_slot_peek_binary_changes('fivetran_pgoutput_slot', null, null, 'proto_version', '1', 'publication_names', 'fivetran_pub');
    

    If the query succeeds, then permissions are sufficient.

IMPORTANT: You must periodically tune the checkpoint_timeout and max_wal_size parameters based on your PostgreSQL database operations. If you do not, you may experience replication failures. To learn how to tune, read this tuning checkpoints documentation.

XMIN

The XMIN method is based on the hidden xmin system column that is present in all PostgreSQL tables. To enable XMIN, ensure that the statement_timeout setting on your server is either 0 (the default value to disable the timeout) or greater than 5 minute.