Active-Active vs Active-Passive Availability
What you'll learn: How to choose between two fundamental redundancy patterns that directly impact your system's availability characteristics.
The Core Difference
When you deploy redundant systems to improve availability, you face a fundamental choice: should all instances serve traffic simultaneously (active-active), or should some stay idle until needed (active-passive)?
Active-Active Configuration
All redundant instances actively handle requests in parallel. If one fails, the remaining instances simply continue serving traffic—users experience no interruption, though possibly reduced capacity.
Availability impact: Typically higher. There's no failover delay since traffic already flows to healthy instances. Your system can often tolerate multiple failures.
Tradeoffs:
- Requires sophisticated load balancing and state synchronization
- Higher infrastructure cost (all resources run continuously)
- Complex consistency challenges with shared data
- Better resource utilization—you're using what you're paying for
Active-Passive Configuration
One instance (primary) handles all traffic while backups (standby) wait idle. When the primary fails, a failover mechanism promotes a standby to active.
Availability impact: Slightly lower due to failover time. Your MTTR includes detection lag + failover duration (seconds to minutes).
Tradeoffs:
- Simpler to reason about—one writer, clear ownership
- Wasted capacity during normal operation
- Risk that standby has drifted or won't start cleanly
- Lower infrastructure efficiency but easier state management
Calculating the Difference
If failover takes 30 seconds and you average one failure per month, active-passive costs you ~30 seconds of downtime per incident. With 99.9% availability budgets (43 minutes/month), that's significant. Active-active eliminates this failover window entirely.
Key Takeaway: Active-active offers higher availability and better resource utilization but requires sophisticated state management; active-passive is simpler but burns availability budget during every failover event. Choose based on your SLO requirements and operational complexity tolerance.