4.1.3. Design for High Availability
💡 First Principle: High availability is achieved by systematically eliminating single points of failure through redundancy across independent fault domains, ensuring continuous operation despite component or datacenter failures.
What HA protects against, and where DR takes over:
High availability is horizontal, inside one region. Disaster recovery is the arrow to another region. Availability Zones absorb a datacenter failure with no failover event and no data loss, because replication between them is synchronous. Crossing the arrow means asynchronous replication, so it carries a non-zero RPO, and it means declaring a failover, so it carries a real RTO.
⚠️ Exam Trap: when a requirement says "within a single region", every cross-region option is wrong however resilient it sounds — and when it names a regional outage, no amount of zone redundancy is enough.
Targets such as 99.99% for a web tier usually need both layers at once: zone-redundant deployment inside each region, several regions behind a global entry point, and health probes at that entry point that drop an unhealthy region automatically. At 99.99% the downtime budget is only minutes a month (see the SLA arithmetic below), which a manual DNS runbook cannot meet.
Scenario: You are designing a new mission-critical application. It needs to remain operational even if a single server fails, or if an entire datacenter within an Azure Region experiences an outage.
Think of high availability like a power grid — no single power plant failure should black out a city. The grid distributes generation across independent plants, with automatic rerouting when one fails. Azure's HA works the same way: distribute across independent fault domains (Availability Zones), with automatic failover so users never notice a component failure.
| Failure Level | Protection Mechanism | Azure Feature |
|---|---|---|
| Single VM | Multiple instances | VM Scale Sets |
| Server rack | Fault domains | Availability Sets |
| Datacenter | Zone redundancy | Availability Zones |
| Region | Geo-redundancy | Paired regions + ASR |
SLA arithmetic is examined. Tiers that must all be up multiply: 99.95% × 99.99% ≈ 99.94%, so a chain is always weaker than its weakest link. Each extra nine cuts the downtime budget tenfold — 99.9% allows about 43.8 minutes a month (8.76 hours a year), 99.99% about 4.4 minutes. For virtual machines the placement sets the figure: a single VM on Premium SSD 99.9%, two or more in an Availability Set 99.95%, two or more across Availability Zones 99.99%.
Redundancy needs more than one instance behind a health probe. A single Azure Container Instances container group has no replicas or built-in load balancing, so a highly available container API belongs on Azure Container Apps or AKS with several replicas.
High Availability (HA) ensures a system remains operational and accessible, minimizing downtime by deploying redundant components across isolated infrastructure and enabling automatic failover mechanisms.
This task delves into the practical application of HA services. You'll explore how to:
- Design for Azure Availability Zones: Isolate workloads and provide fault tolerance against datacenter-level failures.
- Design for Azure Availability Sets: Distribute VMs across isolated hardware clusters within a single datacenter.
Mastering these concepts is crucial for the AZ-305 exam, as it assesses your ability to analyze, evaluate, and create resilient HA designs.
⚠️ Common Pitfall: Deploying all tiers of an application into a single Availability Zone. While simpler, this creates a single point of failure at the datacenter level.
Key Trade-Offs:
- Availability Zones vs. Availability Sets: Availability Zones provide a higher level of availability (protection from datacenter failure) but are not available in all regions and may have slightly higher inter-zone latency. Availability Sets protect against hardware failure within a datacenter.
Reflection Question: How do Azure's HA options (Availability Zones and Availability Sets) collectively ensure high availability and fault tolerance for your applications by providing distinct levels of redundancy against various failure scenarios?
