Armada KVArmada KV
/vmain
Documentation
Edit on GitHub

Backups

Armada supports creating backups and restoring from them through the Maintenance gRPC API and the arctl backup and restore commands.

Important: Backing up and restoring can be done only in a leader cluster.

To interact with Armada's Maintenance API, use the arctl binary, which can be downloaded from the Releases GitHub page or use the Docker Image.

Create backup

To create backups, Maintenance API must be enabled during Armada startup. See the Helm Chart or the CLI documentation for reference. The chart path still uses the legacy regatta name.

Additionally, a token must be provided during Armada startup which is then provided to the arctl backup command:

arctl backup \
      --address=https://127.0.0.1:8443 \
      --token=$(BACKUP_TOKEN) \
      --ca=ca.crt \
      --dir=/backup \
      --json=true

The command then creates binary file for each table and a human-readable JSON manifest from Armada leader cluster running on 127.0.0.1:8443.

Periodically backing up to S3 Bucket

Armada Helm Chart also offers a CronJob to periodically create backup and push it to an S3 Bucket.

Shared-Store Snapshot Exporter

Armada includes a shared-store snapshot exporter that can continuously or periodically push snapshots to an object store (e.g., S3). This capability, along with HTTP live fallback, provides a resilient and resumable path for snapshot queries and recovery without overloading the leader cluster's local disks or RPC network during large restores.

Incremental Snapshots

In addition to full cluster backups, Armada supports Incremental Snapshots. This allows taking backups of only the differences (deltas) since the last full snapshot, significantly reducing backup sizes and the network footprint required to transfer them.

Restore from backup

Warning: Restoring from backups is a destructive operation and should be used only as a part of a break-glass procedure.

To restore from backups, Maintenance API must be enabled during Armada startup and a token and a directory containing binary backups and the JSON manifest must be provided to the arctl restore command. All tables present in the manifest are then restored.

arctl restore \
      --address=https://127.0.0.1:8443 \
      --token=$(BACKUP_TOKEN) \
      --ca=ca.crt \
      --dir=./backup \
      --json=true

This command overwrites all the tables specified in the backup directory in an Armada leader cluster running on 127.0.0.1:8443.

Resetting a follower cluster

Data in the follower cluster can also be wiped completely, forcing the follower to reload all the data directly from the leader. See the Reset method in the Maintenance gRPC API documentation for more information.

← PreviousIntroductionNext →arctl