Réplication en cascade

Welcome to Cascading Replication! In typical PostgreSQL replication, all standby servers connect directly to the primary server.

However, for larger setups or when you need many read replicas, this can put a lot of load on the primary. Cascading replication offers a solution.

It means a standby server can itself act as a primary for other standbys, forming a chain.

Imagine a chain: The primary server sends its Write-Ahead Log (WAL) to a first standby (let's call it Standby A).

Then, Standby A, instead of just consuming the WAL, also acts as a source, sending that same WAL data to a second standby (Standby B).

This creates a flow: Primary → Standby A → Standby B. Standby A is an 'intermediate' standby.

Why go through the trouble of chaining standbys? Cascading replication offers several advantages:

• Reduced Load on Primary: The primary only needs to send WAL to Standby A, not to all standbys directly.
• Scalability for Reads: You can create many read replicas without overwhelming the primary's network or I/O.
• Geographical Distribution: Standbys can be spread across different data centers, with intermediate standbys bridging the distance.

The first step is to set up Standby A as a regular streaming replication standby. This involves taking a base backup from the primary and configuring Standby A to connect to the primary.

The key here is that Standby A must also be configured to send WAL files to other standbys, not just receive them.

For Standby A to act as a source for other standbys, you need to adjust its PostgreSQL configuration (`postgresql.conf`).

The important parameters are wal_level and max_wal_senders. hot_standby should also be on to allow queries.

Now, for Standby B, instead of pointing its primary_conninfo to the actual primary server, you point it to Standby A.

This tells Standby B to receive its WAL stream from Standby A.

Just like any new standby, Standby B needs an initial copy of the data. You can take this base backup directly from the primary, or, more commonly in a cascading setup, you can take it from Standby A.

Using Standby A for the base backup further reduces load on the primary.

It's crucial to monitor the health of your cascading setup. The pg_stat_replication view is your best friend.

On the primary, you'll see Standby A connected. On Standby A, you'll see Standby B connected.

Cascading replication shines in scenarios where:

• You have a high number of read queries and need many read replicas.
• You want to offload backup operations to a standby, but also need other standbys for reads.
• You are distributing your database across multiple regions and want to minimize cross-region traffic from the primary.
• You're building complex disaster recovery tiers.

Let's test your understanding of cascading replication.

In this lesson, we explored cascading replication, a powerful architecture for scaling PostgreSQL read replicas and distributing replication load.

You learned how standbys can form a chain, with one standby replicating from another, and the key configurations and benefits, such as reduced primary load and enhanced read scalability.

This technique is essential for robust, distributed PostgreSQL environments.