Data resiliency

This page covers the disaster recovery recommendations for ClickHouse Cloud, and guidance for customers to recover from an outage. ClickHouse Cloud doesn’t currently support automatic failover, or automatic syncing across multiple geographical regions.

​

Definitions

It is helpful to cover some definitions first. RPO (Recovery Point Objective): The maximum acceptable data loss measured in time following a disruptive event. Example: An RPO of 30 mins means that in the event of a failure the DB should be restorable to data no older than 30 mins. This, of course, depends on how frequently backups are taken. RTO (Recovery Time Objective): The maximum allowable downtime before normal operations must resume following an outage. Example: An RTO of 30 mins means that in the event of a failure, the team is able to restore data and applications and get normal operations going within 30 mins. Database Backups and Snapshots: Backups provide durable long-term storage with a separate copy of the data. Snapshots don’t create an additional copy of the data, are usually faster, and provide better RPOs.

​

Database backups

Having a backup of your primary service is an effective way to utilize the backup and restore from it in the event of primary service downtime. ClickHouse Cloud supports the following capabilities for backups.

1. Default backups

By default, ClickHouse Cloud takes a backup of your service every 24 hours. These backups are in the same region as the service, and happen in the ClickHouse CSP (cloud service provider) storage bucket. In the event that the data in the primary service gets corrupted, the backup can be used to restore to a new service.

2. External backups (in the customer’s own storage bucket)

You can export backups to your own object storage in your account, in the same region, or in another region. Cross-cloud backup export support is coming soon. Applicable data transfer charges will apply for cross-region, and cross-cloud backups.

3. Configurable backups

Customers can configure backups to happen at a higher frequency, up to every 6 hours, to improve the RPO. Customers can also configure longer retention. The backups currently available for the service are listed on the “backups” page on the ClickHouse Cloud console. This section also provides the success / failure status for each backup.

​

Restoring from a Backup

1. Default backups, in the ClickHouse Cloud bucket, can be restored to a new service in the same region.
2. External backups (in customer object storage) can be restored to a new service in the same or different region.

​

Backup and restore duration guidance

Backup and restore durations vary significantly with data size, schema complexity, and the number of tables. The figures below are observations from internal testing and are not performance commitments. In our testing, a backup of roughly 1 TB has taken about 10 to 15 minutes, and 20 TB about an hour. Larger datasets take proportionally longer, and backups of several hundred TB or more can take many hours, although there are economies of scale that help at larger sizes. Restores are typically slower than the equivalent backup, particularly when the restore includes a long chain of incremental backups, since the full backup and all subsequent incremental backups in the chain must be applied. Restoring a very large service can take longer than a day. Because durations depend heavily on your data, schema, and backup configuration, we strongly recommend testing a restore against your own data before relying on any target recovery time.

​

Recovery process

This section explains the various recovery options and the process that can be followed in each case.

​

Primary service data corruption

In this case, the data can be restored from the backup to another service in the same region. The backup could be up to 24 hours old if using the default backup policy, or up to 6 hours old (if using configurable backups with 6 hours frequency).

​

Restoration steps

To restore from an existing backup

1. Go to the “Backups” section of the ClickHouse Cloud console.
2. Click on the three dots under “Actions” for the specific backup that you want to restore from.
3. Give the new service a specific name and restore from this backup

​

Primary region downtime

You can export backups to your own cloud provider bucket. If you’re concerned about regional failures, we recommend exporting backups to a different region. Keep in mind that cross-region data transfer charges will apply. If the primary region goes down, the backup in another region can be restored to a new service in a different region. Once the backup has been restored to another service, you will need to ensure that any DNS, load balancer, or connection string configurations are updated to point to the new service. This may involve:

• Updating environment variables or secrets
• Restarting application services to establish new connections

​

Additional options

There are some additional options to consider.

1. Dual-writing to separate clusters

In this option, you can set up 2 separate clusters in different regions and dual-write to both. This option inherently comes with a higher cost as it involves running multiple services but provides higher availability in case of one region being unavailable.

2. Utilize CSP replication

With this option, you would utilize the cloud service provider’s native object storage replication to copy data over. For instance, with BYOB you can export the backup to a bucket that you own in the primary region, and have that replicated over to another region using AWS cross region replication.