> For the complete documentation index, see [llms.txt](https://wiki.datadike.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://wiki.datadike.com/pam/deploy-operate/backup-and-restore.md).

# Back up and restore

Backing up the installation — database, configuration and recordings — where each of those actually lives, and the test restore that turns a backup into something you can rely on.

## What this is and when to use it

This page is about backing up **the installation**: the database, the configuration, and the session recordings. It is what you need to bring the service back after losing a server.

It is not the credential backup. That is a separate feature — an encrypted export of the vault, delivered by email, SFTP or webhook, with the decryption key split between two custodians — and it exists for a different scenario: recovering strategic credentials without restoring the environment at all. It is documented under [The credential vault](https://wiki.datadike.com/pam/administration/accounts/vault.md). Configure both; they cover different disasters.

{% hint style="danger" %}
**A backup that has never been restored is a hypothesis.** The backup job reporting success proves that a file was written, not that anything can be recovered from it. Everything under [How to verify it worked](#how-to-verify-it-worked) exists because that distinction is where recoveries fail.
{% endhint %}

{% hint style="warning" %}
**The backup routine for an installation is defined with DataDike implementation support.** The commands, the schedule and the destinations depend on the deployment. What follows is what the platform is specified to do and what the routine has to cover, rather than a sequence to run as it stands.
{% endhint %}

## What has to be in the backup

Four things, in three different places. A backup covering only the first is the most common way a restore fails.

| What                                          | Where it lives                                                                                                                         | Why it needs separate thought                                                                                                                                                                                        |
| --------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **The database**                              | Internal to the appliance                                                                                                              | Holds users, assets, accounts, authorizations, and the audit and session records. The platform supports backup and recovery of the database and of the software configuration                                        |
| **Configuration that is not in the database** | The appliance configuration file, `/opt/datadike/config/config.txt` — the file the syslog settings and the cleanup interval are set in | Changes to it take effect only when the services restart, which also means a restored file does nothing until the services come back                                                                                 |
| **Session recordings**                        | The object storage backend configured under **System Settings › Storage › Object Storage**                                             | By default this is the `default` backend, type Server — the local filesystem of the server. If you moved it to an external bucket, that bucket is backed up on its own terms and a database backup does not touch it |
| **The broker PKI**, if you use a broker       | The broker VM                                                                                                                          | The certificates are the trust anchor for every gateway and connector enrolled through that broker                                                                                                                   |

{% hint style="warning" %}
**Back up the broker VM, and address the broker by DNS name.** Losing the broker VM without a backup means reinstalling it and re-enrolling every remote gateway and connector. And the address is written into the certificates and the enrolment tokens: if the broker was addressed by DNS name you can point that name at a restored VM and the certificates stay valid, but if it was addressed by a literal IP the address cannot be changed afterwards without reinstalling the broker.
{% endhint %}

## What the platform is specified to do

These are the capabilities stated for the solution. Treat them as the shape of what to ask for, and confirm how each is configured in your installation.

| Capability          | As stated                                                                                                                                                                                     |
| ------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Destinations        | Backup and restore of all data and configuration, exportable to a remote server via rsync or over CIFS/NFS network shares                                                                     |
| Off-site copies     | System backups, logs, videos and credentials can be extracted to a server in a remote data centre, so that all configuration and data can be restored from there                              |
| Contents protection | No account or password appears in clear text in the backup file                                                                                                                               |
| Backup retention    | Backups without a retention restriction, encrypted and protected against tampering; reports and logs retained in backup for at least ten years, and session recordings for at least ten years |
| Live retention      | Reports and historical files remain available without a restore for at least 360 days                                                                                                         |

## Backup is not the same as high availability

The architecture already protects against a node failing: active/active operation, database replication with automatic primary election, cache and queue replication with failover, storage shared between nodes, and a single virtual address in front. The stated continuity objectives are **RTO of one hour or less and RPO of fifteen minutes or less**.

None of that is a backup. Replication propagates a deletion exactly as fast as it propagates a write. High availability covers a node dying; the backup covers something being deleted, corrupted, or changed wrongly — including by you, during a change window.

## Before you start

* **Decide the destination** — remote server over rsync, or a CIFS/NFS share — and confirm the appliance can reach it. An off-site copy is what survives the event that takes the primary site.
* **Find out where recordings actually are.** Open **System Settings › Storage › Object Storage** and note which backend is marked **Default**. Exactly one is. If it is an external bucket, its backup is a separate arrangement.
* **Confirm the credential backup separately**, including whether two custodians are configured — with only one, the key is not split at all. See [The credential vault](https://wiki.datadike.com/pam/administration/accounts/vault.md).
* **Have somewhere to restore into.** A test instance is not optional equipment; it is the only place the verification below can happen.
* **Take a full backup before any structural change**, and specifically before enabling and synchronising an external account storage vault — the source calls this out, along with testing that integration in a staging environment first.

## Restoring

{% hint style="warning" %}
**No restore procedure could be ported: the source does not document one.** Engage DataDike implementation support for the restore of an installation. What follows is what the source does establish about recovery paths, which is worth knowing *before* you need them.
{% endhint %}

**Credentials can be recovered without restoring the environment.** The credential backup can be opened with its encryption key, so that in a disaster strategic credentials are recoverable even while the platform itself is unavailable. This is the fastest recovery path for the case "we need the passwords right now", and it is entirely separate from restoring the installation.

**Recovering the backup encryption key needs several people.** The key recovery process assigns administrators as custodians of parts of the key, and a predefined number of them must be present to reconstruct it during disaster recovery.

{% hint style="danger" %}
**The vault key must not be held by a single person.** That is a design requirement of the solution, not a preference. It also means recovery is a scheduled, multi-person activity: find out now who the custodians are and whether they are all still with the organisation, rather than during the incident.
{% endhint %}

**A restored broker keeps its certificates only if it was addressed by DNS name.** Otherwise the broker is reinstalled and every remote gateway and connector is re-enrolled.

## How to verify it worked

This is the section that matters. Restoring is the only test of a backup, and the checks below are ordered so that each one covers a gap the previous one cannot see.

1. **Restore into a test instance** — never over production. If you cannot name the date of your last successful test restore, you do not have a verified backup, whatever the job history says.
2. **Check `/api/health/` on the restored instance**: `db_status` and `redis_status` both true. This confirms the restored database is not just present but reachable and answering.
3. **Check System Settings › Components**: every component **Normal**. A restore that brings back the database but not the components produces a console that loads and a platform that cannot open a session.
4. **Open a real session** on the restored instance, on a protocol you use, and confirm it completes and is recorded.
5. **Play back a recording made before the backup was taken.** This is the check that catches the most common gap in the whole procedure: the recordings live in a storage backend, and a database-only backup restores every session *record* while every replay is missing. If playback fails here, your backup does not cover recordings.
6. **Confirm the server date and time on the restored instance.** A wrong clock makes OTP codes invalid, so MFA logins fail on an otherwise perfectly restored system — see [Multi-factor authentication](https://wiki.datadike.com/pam/administration/identities/mfa.md).
7. **Verify the credential backup on its own terms**: the job succeeded, and each custodian received their part — two parts, if you configured two custodians.
8. **Write down the date, the version restored, and what failed.** The value of a test restore is mostly in the list of things that did not work.

## Common errors

| Symptom                                                                | Cause                                                                                                                                       | Fix                                                                                                                    |
| ---------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- |
| Restored instance works, but no session replay will play               | The recordings live in the object storage backend, which the database backup never touched                                                  | Back up the storage backend too; check which backend is **Default** under **Storage › Object Storage**                 |
| Recording files are gone while the session records still list them     | Console retention governs database records; the files' lifecycle is the bucket's own lifecycle policy, configured separately                | Align the bucket lifecycle policy with your retention requirement                                                      |
| An old audit event cannot be found at all                              | Its retention window elapsed and the nightly cleanup deleted the record. No error is shown — an empty result looks like "it never happened" | Set retention values against your audit obligations before you need the evidence                                       |
| Credential backup arrives as one file that a single recipient can open | Only one custodian was configured, so the key was never split                                                                               | Configure both custodians — see [The credential vault](https://wiki.datadike.com/pam/administration/accounts/vault.md) |
| Backup encryption key cannot be reconstructed                          | Fewer custodians available than the number required to recover the key                                                                      | The threshold is a design requirement; keep the custodian list current and documented                                  |
| Broker restored, but gateways and connectors never reconnect           | The broker was addressed by a literal IP, so its certificates cannot be re-pointed at a restored VM                                         | Address the broker by DNS name. Otherwise reinstall the broker and re-enrol each remote point                          |
| A restored configuration file appears to have no effect                | Configuration in the appliance's `config.txt` is read when the services start                                                               | Restart the services after restoring the file                                                                          |
| MFA login fails on a freshly restored instance                         | Server date and time incorrect, so OTP codes are rejected                                                                                   | Correct the server clock                                                                                               |

## Related pages

* [The credential vault](https://wiki.datadike.com/pam/administration/accounts/vault.md)
* [Session recordings](https://wiki.datadike.com/pam/administration/audit/recordings.md)
* [Audit trail](https://wiki.datadike.com/pam/administration/audit/audit-trail.md)
* [Upgrade](https://wiki.datadike.com/pam/deploy-operate/upgrade.md)
* [Monitoring and health](https://wiki.datadike.com/pam/deploy-operate/monitoring.md)
