Skip to main content

Backup, Restore, and Disaster Recovery

Three different capabilities that solve three different problems. They are routinely confused, and the confusion is expensive — it is usually discovered during an incident that none of them covers.

The three, compared

High AvailabilityBackup and restoreDisaster recovery
Protects againstA server or process failingDamage to the data itselfLosing the whole site
ScopeOne site, two serversPoint-in-time copies, held elsewhereA second location
Recovery timeSecondsHoursHours to days
Data lossNoneBack to the last good copyBack to the last replicated copy
AutomaticYesThe copies are; the restore is notNo — a declared decision
What Linkiir providesThe active/standby pairInstance and project backups you configureSupport for building it; the sites are yours

HA gives you uptime. Backups give you recoverability. DR gives you a second place to run. A production deployment needs all three, and each covers a category the others cannot.

Why HA is not a backup

The two servers share one copy of your data

An HA pair has one working directory and one log database. That sharing is exactly what makes failover fast — and it means anything that damages the data damages it for both servers at the same instant. Delete a project on the active server and it is gone on the standby too, immediately.

EventDoes HA help?What actually recovers it
Active server diesYes — that is what HA is forHA
Someone deletes a projectNoProject backup
A bad configuration change is savedNoInstance backup
Database corruptionNoDatabase backup
Ransomware reaches the shared storageNoOff-site, versioned backups
Shared storage array failsNoStorage redundancy, then backups
The site is lostNoDisaster recovery

What a complete Linkiir backup covers

Three separate things, and a backup is only complete when all three are configured. Configuring the first one alone is the most common gap.

WhatHow it is protectedCadence
Instance configuration — settings, users, rolesInstance backup to a remote you provideAutomatic, hourly
Projects and workflowsEach project pushed to its own remoteOn change
Log database — message history and log recordsYour database platform's own backupsPer your retention policy
The instance backup does not include your integrations

Projects are separate repositories with their own remotes. An instance backup protects settings, users, and roles — not your workflows. Configure both.

See Backup and Restore for how each is set up, and make the backup destinations reachable from both servers so a failover does not interrupt them.

Restoring, and what changes

A restore rebuilds an installation from those copies. Two consequences to plan for rather than discover:

A restore onto replacement servers needs the license re-applied. Put it in the runbook as an explicit step, and agree it with Linkiir support in advance rather than discovering it mid-incident. See HA Licensing.

Restoring a pair means restoring one shared working directory, not two copies. Both servers must end up opening the same restored directory, at the same path. Restoring into two directories and pointing one server at each produces two separate installations rather than a pair.

Test a restore on a schedule. An untested backup is an assumption.

Disaster recovery

DR is about a second location, and it is a different project from HA.

Why an HA pair is not DR

Both servers in a pair sit in one site, share one storage system, and depend on one broker cluster. A site-level event takes all of it.

Why a pair cannot be stretched across two sites

The broker cluster needs a majority of its nodes reachable to accept writes. Split three nodes across two sites and one site holds two nodes and the other holds one: losing the first site halts the cluster, while losing the second does not. The result survives one specific site failure and not the other, which is not a DR posture anyone should sign off.

Two approaches that do work:

ApproachHow it worksTrade-off
Witness siteTwo main sites plus a small third location holding one broker node, so a majority survives losing either main siteNeeds a third location with reliable, low-latency connectivity to both
Independent deployments with replicationA separate Linkiir installation at the DR site, kept current from backups and replicated project remotes, brought up on a declared decisionRecovery is manual and slower, but each site stands alone

For most organisations the second approach is the practical one, because it needs no third location and no low-latency link between sites.

Planning a DR site

Settle these before you build it:

  • Your RPO and RTO for a site loss. They are hours, not the seconds an HA failover gives you. Write them down, because they drive everything else.
  • How data reaches the DR site — database replication, restored backups, replicated project remotes, or a combination.
  • How integration partners reach the DR site. Feeds arrive on listener ports at a specific address. Moving them is usually the slowest part of a real DR event.
  • The licensing position for the DR installation. Agree it with us in advance. See HA Licensing.
  • Who declares a disaster, and what the runbook says. DR is not automatic, and it should not be.
  • How you test it. Untested DR is a plan, not a capability.

Putting it together

LayerCoversRecoveryWho builds it
HA pairServer and process failureSeconds, automaticLinkiir with you
Instance and project backupsData damage and mistakesHours, manualYou, in Linkiir
Database backupsLog and history lossHours, manualYour database team
Storage redundancyShared-storage failureDepends on the platformYour infrastructure team
DR siteSite lossHours to days, declaredYou, with Linkiir support

Next