DATABASES & BACKUPS

Backups you have actually tested.

Scheduled backups, point-in-time recovery, and automated restore verification across Postgres, MySQL and MariaDB. A backup nobody has restored is a hypothesis, not a backup.

Install itSee a verification run
PostgreSQL·MySQL·MariaDB·S3 / R2 / MinIO

THE MANUAL WAY

The failure mode nobody plans for.

03:40 · SATURDAY

A migration went wrong four hours ago. You have nightly dumps, so you are about to lose everything since midnight — and you are not certain the dump restores cleanly, because nobody has tried since March.

THE QUARTERLY AUDIT

Someone asks when the disaster recovery plan was last tested. The honest answer is that it has never been tested, only written down.

SIX MONTHS OF GREEN

The backup job has reported success every night. It has also been backing up an empty schema since the database URL changed, and nothing checked the contents.

WHAT TALOS DOES

Backups that prove themselves.

Scheduled backups

Per-database schedules with retention tiers. Expiry runs as its own audited job rather than as a side effect.

Point-in-time recovery

Continuous archiving lets you restore to a moment inside the retention window, not just to last night.

Automated restore verification

Every backup is restored into a throwaway instance and checked. Unverified backups are flagged, not counted.

Encrypted off-site storage

Stream to S3, R2 or MinIO with credentials held encrypted by Talos rather than in a cron environment.

Size and drift alerts

Track size over time. A backup that suddenly shrinks by 90% raises an alert before you need it.

Gated restores

Restoring over a live database is a two-person operation. The gate names the target and the timestamp explicitly.

talos.internal/databases/backups
Servers
Kubernetes
Frappe
Docker
Databases
Pipelines
Monitoring
Backups18 databases · 4 verified today
acme-erp · pg16VERIFIED
42.1 GB · PITR 14d · verified 04:12
billing · pg15VERIFIED
8.4 GB · PITR 30d · verified 04:31
analytics · mysql8TESTING
210 GB · nightly · restoring now
legacy-crm · mariaDRIFT
3.1 GB · size dropped 92%
$ db verify acme-erp --latest
→ restoring into ephemeral pg16 instance …
→ row counts: 1,284,551 across 212 tables
→ integrity checks passed · schema matches
✓ backup verified restorable · destroyed

HOW IT RUNS

Every action is a tracked job.

Requestyou, API or schedule
Queuegated if required
the Sentinelbackground worker
AdapterSSH · Ansible · Terraform · API
Your infrastructurelogs stream back live

WHAT MAKES THIS DIFFERENT

A backup that was never restored is not a backup.

Talos restores your backups on a schedule into throwaway instances, checks row counts and schema integrity, then destroys them. A backup only earns the word "verified" after it has actually come back. If verification fails, you find out on a Tuesday afternoon — not at 03:40 on a Saturday.

Point-in-time recovery means the blast radius is seconds, not the hours since your last nightly dump.
Verification runs against real restore tooling, not a checksum of the archive.
A failed verification raises an alert the same way a production incident does.
verification timeline · acme-erp
04:12 · base backup taken
42.1 GB · encrypted · uploaded
DONE
04:19 · restored to ephemeral
pg16 · isolated · no prod access
DONE
04:25 · integrity verified
212 tables · 1,284,551 rows
PASS
04:26 · instance destroyed
verified flag written to ledger
CLEAN

RELATED CAPABILITIES

Frappe

Site-level backup and restore for every ERPNext tenant.

explore →

Pipelines & Gates

Require two-person approval before any restore over live data.

explore →

Monitoring

Alert on failed verification, size drift and missed windows.

explore →

Find out if your backups restore.

Point Talos at one database and run a verification. It never writes to the source.

Install in one commandSee every capability →