Couchbase Backup Restore

Couchbase Backup & Restore

A skill for planning and executing Couchbase backup and restore operations — both self-managed (cbbackupmgr) and Capella-managed.

Distinct from:

• couchbase-migration-execution — one-time data migration, not recurring backup
• couchbase-mcp — the admin_backup_* tools that interact with the Backup service

When this skill applies

• "How do I back up Couchbase?"
• "What's cbbackupmgr?"
• "How do I set up a backup schedule?"
• "Full vs incremental backup — what's the difference?"
• "How do I restore from backup?"
• "How do I restore just one bucket, or just a few documents?"
• "How do I back up to S3?"
• "How do I verify my backups are actually working?"
• "What's the RPO/RTO I can achieve?"
• "How does Capella handle backups?"

Pick the right reference

| Question | Read | |---|---| | "cbbackupmgr commands — config, backup, restore, list, merge, examine" | references/cbbackupmgr.md | | "Backup strategy — full vs incremental, schedules, retention, S3" | references/strategy.md | | "How to restore — full cluster, single bucket, single document, point-in-time" | references/restore.md | | "Capella managed backups — schedules, retention, restore" | references/capella-backups.md |

Three core principles

Principle 1 — An untested backup is not a backup. Run a restore drill regularly — quarterly at minimum, monthly for production data. cbbackupmgr examine lets you inspect backup contents without a full restore. A restore drill to a staging cluster is the only way to know your backups are complete and your restore procedure works.

Principle 2 — Incremental backups depend on the full backup chain. Couchbase incremental backups are differential (each increment contains only mutations since the prior backup). To restore to a point in time, you need the full backup plus every increment up to that point. A corrupted full backup invalidates the entire chain. Keep multiple full backup generations, not just increments.

Principle 3 — Backup is not DR. Backup protects against data corruption, accidental deletion, and ransomware. It does not protect against site loss with your RPO — that's what XDCR active-passive is for. Most deployments need both: XDCR for RTO (fast failover), backup for RPO (data recovery from logical errors that XDCR would have replicated).

Quick tool map

| Task | Tool | |---|---| | List backup repositories | admin_backup_list_repositories | | Get repository info | admin_backup_get_repository | | Trigger a backup | admin_backup_trigger_backup | | Get backup history | admin_backup_get_backup_info | | Trigger a restore | admin_backup_trigger_restore | | Get restore status | admin_backup_get_restore_status |

These tools interact with the Couchbase Backup Service (available in Couchbase 7.0+ EE). For older clusters or more control, use cbbackupmgr CLI directly.

RPO/RTO guidance

| Approach | RPO | RTO | Notes | |---|---|---|---| | Incremental backup hourly | ~1 hour | Hours (depends on data size) | Standard for most workloads | | Incremental backup every 15 min | ~15 min | Hours | Increases backup storage and I/O | | XDCR active-passive + backup | Near-zero (XDCR) / hours (backup) | Minutes (XDCR failover) | Best of both for production | | Capella managed (daily + auto) | Up to 24 hours | Hours | Acceptable for non-critical data |

Related skills

• couchbase-xdcr — XDCR for near-zero RPO / low RTO (complements backup, not a replacement)
• couchbase-mcp — admin_backup_* tools
• couchbase-migration-execution — one-time migration (uses similar tooling)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: celticht32
• Source: celticht32/Couchbase-Skills-for-Claude.ai
• License: MIT

Install and usage instructions live in the source repository linked above.