High availability of a Yandex MPP Analytics for PostgreSQL cluster

High availability of a Yandex MPP Analytics for PostgreSQL cluster depends on the client connection settings, available storage, and other parameters.

Connecting to a databaseConnecting to a database

Yandex MPP Analytics for PostgreSQL cluster availability depends on the client you use and database connection settings. To ensure cluster availability:

• Configure security groups correctly.

• When connecting, specify both the primary and standby master hosts or use the special primary master FQDN.

  Warning

  Using a special primary master FQDN makes the cluster temporarily unavailable during a failover to the standby master.

• Create a separate resource group for cluster administration to be able to connect to and administer the cluster even when resource limits in other groups are exhausted.

• Only use recommended clients for connection.

Available storage spaceAvailable storage space

If the database storage is 95% full, the cluster will switch to read-only mode. To keep your cluster writable, regularly check its Disk usage chart or create an alert.

Restoring a cluster from a backupRestoring a cluster from a backup

If a cluster goes down, e.g., if the cluster’s availability zone fails, you can restore it from a backup.

To ensure data integrity and minimize cluster recovery time:

• Make sure automatic backups are created.
• Perform manual backups before making important changes. Manual backups are full copies of all databases, stored indefinitely.
• Conduct scheduled tests of cluster recovery from backups.

Maintaining a cluster and modifying its parametersMaintaining a cluster and modifying its parameters

Running unscheduled maintenance operations may reduce cluster performance or make your cluster unavailable for some time. Set the maintenance start time to a day of week and time interval with no cluster workload.

Actions such as changing the host class, disk type, or increasing the storage size interrupt sessions and make the cluster unavailable for some time. Perform these operations when your cluster has no active workload.