Redundancy and Failover
Redundancy provides alternative components or capacity, while failover transfers service to those alternatives when a failure occurs.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/redundancy-and-failover/
01What it means
Redundancy means maintaining additional resources so a single component’s loss need not stop service. Failover is the detection and transition process that makes the alternative serve the workload. Active-active and active-passive arrangements differ in how spare capacity is used. Merely having replicas does not establish successful failover or protect against every shared failure.[1][2]
02Why it matters
Failure handling depends on routing, state, capacity, and dependencies working together. Planning the transition helps teams avoid discovering during an incident that the spare system cannot accept real traffic.
03When to use it
Use redundancy for workloads whose availability requirements justify its cost. Match separation to the failure being addressed, such as an instance, zone, or region. Keep backups and recovery plans for destructive data changes that replicated infrastructure can reproduce everywhere.
04How it works
- Identify relevant failure domains and shared dependencies.
- Choose alternative capacity and a state-replication arrangement.
- Define health signals, routing changes, and safeguards against conflicting active owners.
- Verify the surviving capacity can support the required workload.
- Exercise failover and return-to-normal behavior, including data reconciliation.
05A practical example
A service runs across two zones, but both instances depend on one stateful database. A failover exercise exposes that shared point of failure. The team adds a tested database recovery path and checks that either zone can handle the required traffic alone.
06Where it breaks
Redundancy increases cost and maintenance and may create correlated failures through shared configuration or credentials. Failover can interrupt sessions or lose recent updates. Multiple replicas are not independent backups when corruption propagates between them.
Common mistakes
- Counting replicas without examining their shared failure domain.
- Provisioning a standby too small for the recovery workload.
- Testing the switch away from failure but never testing safe return.
07Sources
- Architecture strategies for designing for redundancyMicrosoft · Official documentationArchitecture strategies for designing for redundancy
- REL10-BP01 Deploy the workload to multiple locationsAmazon Web Services · Official documentationREL10-BP01 Deploy the workload to multiple locations