High availability and load balancing

• How to exclude a host from the master selection?

• Why do I get an error when setting up cascading replication?

• How do I always connect to the master host?

• How do I always connect to the most recent replica?

• Why did the master and the replicas switch places?

• Can I configure multimaster in a cluster or between two clusters?

• How to configure load balancing so that read requests go to replicas?

• Why does the master fail over to a randomly selected host when autofailover is off?

• Why does the master fail over when the cluster recovers, even though autofailover is off?

• Why does the master fails over automatically during maintenance operations, even though autofailover is off?

How to exclude a host from the master selection?How to exclude a host from the master selection?

You can exclude a host from the master selection when replacing the master host automatically or manually. To do this, set up cascade replication: specify a replication source for the host you want to exclude.

Why do I get an error when setting up cascading replication?Why do I get an error when setting up cascading replication?

Error message:

cluster should have at least 2 HA hosts to use cascade host

The error occurs if you specify a replication source for a single non-cascading replica.

To ensure high availability, your cluster must have at least one replica without a replication source. During maintenance or if the master host fails, this replica will take over as the master.

To learn more about replication, see this guide.

How do I always connect to the master host?How do I always connect to the master host?

To connect to the current master host, use a special FQDN. It has this format: c-<cluster_ID>.rw.mdb.yandexcloud.net. When connected to this FQDN, you will be able to perform read and write operations.

psql "host=c-<cluster_ID>.rw.mdb.yandexcloud.net \
      port=6432 \
      sslmode=verify-full \
      dbname=<DB_name> \
      user=<username>"

How do I always connect to the most recent replica?How do I always connect to the most recent replica?

To connect to the most recent replica, use a special FQDN. It has this format: c-<cluster_ID>.ro.mdb.yandexcloud.net. When connected to this FQDN, you can perform only read operations.

psql "host=c-<cluster_ID>.ro.mdb.yandexcloud.net \
      port=6432 \
      sslmode=verify-full \
      dbname=<DB_name> \
      user=<username>"

If there are no active replicas in the cluster, this FQDN will point to the current master host.

Why did the master and the replicas switch places?Why did the master and the replicas switch places?

This means the master has failed over to the replica host. Failover ensures cluster availability during maintenance or if the master fails.

Use a special FQDN to always connect to the current master.

Can I configure multimaster in a cluster or between two clusters?Can I configure multimaster in a cluster or between two clusters?

No, Managed Service for PostgreSQL does not support multimaster configuration.

For more information on how to ensure high availability of a cluster, see High availability managed databases (MDB).

How do I configure load balancing so that read requests only go to replicas?How do I configure load balancing so that read requests only go to replicas?

As Managed Service for PostgreSQL does not perform any load balancing, you need to configure it in your application backend. To send read requests to replicas, the application must independently recognize master and replica hosts, e.g., using libpq. For more information, see this PostgreSQL article.

Alternatively, use a special FQDN that points to the most recent replica.

For more information, see High availability.

Why does the master fail over to a randomly selected host when autofailover is off?Why does the master fail over to a randomly selected host when autofailover is off?

Even with autofailover disabled, the master host may switch to another host during maintenance or cluster updates. This helps keep the cluster working correctly during these operations.

Why does the cluster switch masters after recovery, even though autofailover is disabled?Why does the cluster switch masters after recovery, even though autofailover is disabled?

Even with autofailover disabled, the master host may switch to another host after recovery. This helps keep the cluster working correctly under such conditions. For more information, see High availability of a Managed Service for PostgreSQL cluster.

Why does the master fails over automatically during maintenance operations, even though autofailover is off?Why does the master fails over automatically during maintenance operations, even though autofailover is off?

Even with autofailover disabled, the master host may switch to another host to allow maintenance.