High Availability
😟 THE PROBLEM
The only server crashes at 9:15 during an online exam.
😊 THE SOLUTION
Run copies in several availability zones, so one failure changes nothing.
- One server = one point of failure✓ Run two or more instances
- A whole datacenter loses power✓ Spread across availability zones
- Not sure how much downtime is OK✓ Check the SLA percentage
Quick answer
High availability means designing a system so it stays accessible and running for the maximum possible time, even when individual components fail. In Azure you achieve it with redundancy (multiple instances, availability zones, load balancers). Microsoft expresses the expected uptime of a service as a percentage in its Service Level Agreement (SLA), such as 99.9% or 99.99%.
Exam objective: 1.2.1 Describe the benefits of high availability and scalability in the cloud. (Skills measured as of July 20, 2026)
Start with a question
What happens to an online exam if the only server running it crashes at 9:15?
Ten thousand students see an error page. Some lose their answers. The school has to reschedule, and trust is gone.
Now imagine the same exam running on three servers in three separate buildings. One crashes, and the other two carry on. Students do not even notice. That is high availability.
Explain it like I’m 10
The technical explanation
High availability (HA) is the ability of a system to remain operational and accessible for as much time as possible. Failures will happen: disks die, servers crash, power fails, a datacenter has a problem. HA design makes sure no single failure takes the service down.
The key technique is redundancy, removing every single point of failure:
| Failure | Azure HA technique |
|---|---|
| One VM or instance crashes | Run 2+ instances behind a load balancer (Scale Sets, App Service with multiple instances) |
| A server rack or update domain has a problem | Availability sets spread VMs across fault and update domains |
| An entire datacenter goes down | Availability zones: spread instances across physically separate zones |
| A whole region is unavailable | Multi-region deployment with Azure Front Door or Traffic Manager |
| Storage hardware fails | Redundant storage (LRS, ZRS, GRS, GZRS), covered in Module 2 |
Service Level Agreements
Microsoft publishes an SLA for each service: a formal commitment to a percentage of uptime (or connectivity), with service credits if it falls short. Small differences in percentages mean very different amounts of downtime. Try it:
- Per week
- Per month
- Per year
Maximum downtime still allowed under the SLA.
Combine two services that both must work
Composite SLA: — lower than either service on its own.
Web app (99.95%) calling a database (99.99%): 0.9995 × 0.9999.
Architect level Designing HA without wasting money
- Start with targets: agree on an availability objective with the business (for example 99.95%), plus recovery time and recovery point objectives (RTO, RPO) for disasters.
- Zone-redundant by default in regions that support zones. For many PaaS services (App Service Premium, Azure SQL Database, Storage ZRS) it is a configuration flag, often at little or no extra cost.
- Health probes and graceful degradation: load balancers should remove unhealthy instances quickly; the app should degrade (for example, read-only mode) rather than fail completely.
- Deployments cause outages too: blue-green or slot swaps, feature flags and safe rollout reduce self-inflicted downtime, which is often the biggest source of unavailability.
- Multi-region is expensive in money and complexity (data replication, consistency, failover testing). Use it where the business impact justifies it.
Real-world business scenario
An online payments provider
- The problem
- Every minute of downtime means failed customer payments and penalties from merchants. The single-datacenter setup had two outages last year.
- What they used
- Payment API on App Service Premium with zone redundancy, Azure SQL Database zone-redundant, Azure Front Door for global entry, and a warm standby in a second region.
- Why that choice
- Zones protect against datacenter failures within the region; the second region protects against a full regional outage.
- The result
- Planned maintenance and a datacenter power event in the next year caused no customer-visible downtime.
When to use it and when not to
✓ Use it when…
- Customer-facing or revenue-generating systems.
- Systems with contractual uptime commitments to your own customers.
- Anything where downtime causes safety, legal or reputational harm (healthcare, payments, exams).
✕ Think twice when…
- Spending on multi-region HA for internal tools where a few hours of downtime is acceptable.
- Assuming the provider SLA alone makes your app highly available. Your design matters.
↔ Alternatives to consider
- **Backup and restore** for non-critical systems, which is cheaper but takes longer to recover.
- **Single-region, zone-redundant** design as a cost-effective middle ground.
Interview preparation
Say your answer out loud first, then open the model answer. Interviewers care more about your reasoning than the exact words.
Beginner What is high availability?
Model answer
High availability means a system keeps working with very little downtime even when parts of it fail. For example, running two web servers behind a load balancer so if one crashes, the other keeps serving users.
Interviewer follows up
How does Azure describe availability?
You
With Service Level Agreements (SLAs) that state a guaranteed uptime percentage for each service, like 99.95% or 99.99%. If Microsoft misses it, customers can claim service credits.
Intermediate How much downtime does a 99.9% SLA allow per month, versus 99.99%?
Model answer
About 43.8 minutes per month for 99.9%, and about 4.4 minutes for 99.99%. Every additional nine cuts allowed downtime by a factor of ten.
Experienced developer Your app uses App Service (99.95%) and Azure SQL Database (99.99%). What is the composite SLA and how would you improve it?
Model answer
Both must work, so you multiply, 0.9995 × 0.9999 ≈ 99.94%, which is lower than either part. To raise it, deploy zone-redundant tiers in a region with availability zones, or run active-active in two regions behind Azure Front Door with geo-replicated data. That raises availability because a failure must hit both paths at once. Also remember the SLA is the provider's commitment for its part. Your own code, deployments and dependencies affect real availability too.
Interviewer follows up
Does a higher SLA always justify the cost?
You
No. Start from the business impact of downtime. If an hour of outage costs little, a multi-region active-active design may cost more than it saves. Define availability targets with the business, then design to them.
Scenario An online exam platform cannot be down during exam windows. What would you design for high availability in Azure?
Model answer
At least two instances of each tier spread across availability zones, zone-redundant database and storage, a load balancer or Front Door in front, health probes so unhealthy instances are removed, and deployment slots to avoid downtime during releases. Optionally a secondary region for regional failure.
Test yourself with exercises
4 original practice questions in the AZ-900 style. Pick your answers, then check them. You will see why each option is right or wrong. Score 75% or more and this topic is marked complete.
Quick revision
Key points
- High availability = maximise uptime despite component failures.
- Achieved with redundancy, removing single points of failure.
- Azure tools include multiple instances, availability zones, availability sets, load balancers and zone-redundant services.
- Uptime is promised in SLAs as a percentage.
- Composite SLA: multiply SLAs of components that all must work.
Important terminology
- High availability (HA)
- Keeping a service accessible with minimal downtime, despite failures of individual components.
- Uptime
- The percentage of time a service is available.
- SLA
- Service Level Agreement. Microsoft's formal commitment to a level of uptime or connectivity, often with service credits if missed.
- Redundancy
- Running duplicate components so one failure does not take the service down.
- Single point of failure
- Any component whose failure stops the whole system.
- Availability zone
- A physically separate group of datacenters within an Azure region, with independent power, cooling and networking.
Common mistakes
- Thinking a single VM is highly available. It is a single point of failure.
- Adding SLAs together instead of multiplying them.
Exam tips
- "Ensure the app remains available if a datacenter fails" = availability zones / high availability.
- High availability is about uptime. Scalability is about capacity. Do not mix them up.
Five quick revision questions
-
1.What does high availability aim to maximise?Show
Uptime, meaning the time the service is available.
-
2.What document defines Microsoft's uptime commitment for a service?Show
The Service Level Agreement (SLA).
-
3.About how much downtime per month does 99.9% allow?Show
About 43.8 minutes.
-
4.What is a single point of failure?Show
A component whose failure stops the whole system.
-
5.How do you calculate a composite SLA for two services that both must work?Show
Multiply their SLAs, for example 0.9995 × 0.9999.
Official references
Finished this lesson? Tick it off. Your progress is saved in this browser.
Next up: Scalability