Pilot light is a disaster recovery (DR) strategy where the data layer of a system runs all the time in a second location, but the application servers stay switched off until they are needed. The name comes from the small flame in a gas boiler: always lit, so the full burner can start quickly.

It is one of the four DR strategies AWS describes in its Disaster Recovery of Workloads on AWS whitepaper. In AWS's words, the resources that support "data replication and backup, such as databases and object storage, are always on", while application servers are "switched off" and started only during failover.

What runs, and what doesn't

Always running in the recovery RegionReady but not running
Database replica (for example an Aurora or RDS cross-Region replica)Application servers (EC2, containers)
Replicated S3 buckets and backupsLoad balancers and auto scaling capacity
Machine images and infrastructure-as-code templatesDNS cut-over, ready to switch

Where it sits among the four strategies

StrategyWhat runs in the recovery RegionRecovery speed (relative)Running cost (relative)
Backup and restoreBackups onlySlowest (hours)Lowest
Pilot lightData layer onlyFast, but servers must startLow
Warm standbyA small, working copy of productionFaster, only needs scaling upMedium
Multi-site active/activeFull productionNear-zeroHighest

Because data is replicated continuously, the recovery point is usually close to the moment of failure. Recovery time depends mostly on automation: with infrastructure as code and a rehearsed runbook, pilot light typically recovers in tens of minutes; done by hand, it can take hours. AWS gives no fixed figures, so measure your own in a test.

Two limits worth knowing

  • Replication copies mistakes. If data is corrupted or encrypted by ransomware, the replica receives the damage too. AWS notes that replication "may not protect against disaster events such as data corruption or malicious attack", so point-in-time backups are still needed.
  • Untested pilot lights fail. The servers have never run in the recovery Region until the day you need them. A failover test at least twice a year is what turns the pattern into an actual recovery capability.

For a side-by-side of all four patterns with costs, see AWS Disaster Recovery Strategies. Choosing and testing the right pattern for each system is the core of a Yemberzal Disaster Recovery Assessment.