Warm standby is a disaster recovery (DR) strategy where a smaller but fully working copy of your system runs all the time in a second location, and is scaled up to full size if the main site fails. AWS describes it in its Well-Architected guidance as maintaining "a scaled-down but fully functional version of your workload always running in the recovery Region".
It is the middle option between pilot light, where only the data layer runs, and multi-site active/active, where full production runs in two places.
Warm standby vs pilot light
The difference is what can serve traffic straight away. In AWS's words, "pilot light cannot process requests without additional action taken first, while warm standby can handle traffic (at reduced capacity levels) immediately."
| Pilot light | Warm standby | |
|---|---|---|
| Data layer | Replicated and running | Replicated and running |
| Application servers | Not deployed or switched off | Running at reduced capacity |
| Serves traffic immediately? | No, needs action first | Yes, at reduced capacity |
| Recovery step | Deploy or start servers, then scale | Scale up, then switch traffic |
| Standing cost | Low | Medium |
| Recovery speed (relative) | Slower | Faster |
How small can the standby be?
There is no fixed ratio. AWS notes that "the more scaled-up the warm standby is, the lower RTO and control plane reliance will be", and a fully scaled version is known as hot standby. In practice you size it so the whole path (load balancer, application tier, database, monitoring) is proven to work, and so scaling up is a quick change of numbers rather than a build. Service quotas are set per Region, so check that the recovery Region could actually reach production size.
Limits to plan for
- Replication copies mistakes. AWS warns that data replication "will not protect you against data corruption or destruction unless your solution also includes options for point-in-time recovery", so keep backups too.
- Drift. A standby only works if it matches production. Build both from the same infrastructure as code, so a change to one reaches the other.
- Untested scale-up. The step most often skipped is the one that matters: proving the standby can grow to full size in the time your recovery time objective allows. Test it at least twice a year.
For a side-by-side of all four AWS strategies, see AWS Disaster Recovery Strategies. Choosing the right pattern for each system, and proving it in a failover drill, is the core of a Yemberzal Disaster Recovery Assessment.