High Availability for Compute
VM SLAs · Zones · Scale sets · Load balancing · PaaS HA
Practice questions for this topic →
VM SLAs
| Deployment | SLA |
|---|---|
| 2+ VMs across availability zones | 99.99% |
| 2+ VMs in an availability set or dedicated host group | 99.95% |
| Single VM, Premium SSD or Ultra Disk on all disks | 99.9% |
| Single VM, Standard SSD | 99.5% |
| Single VM, Standard HDD | 95% |
Evidence: the ranking above is confirmed on Microsoft Learn (the disks high-availability article and the availability sets article, which states 99.95%). The percentages match the wording of Microsoft's China-cloud SLA page exactly (99.99 / 99.95 / 99.9 / 99.5 / 95%), and a Microsoft Q&A moderator quotes 99.9% for Premium SSD from the global SLA; I couldn't open the global SLA document itself.
Composite SLA for services in series: multiply them (for example 99.99% × 99.95% ≈ 99.94%); dependencies in series lower the result, redundant paths raise it.
HA patterns
| Workload | HA approach |
|---|---|
| VMs | VM Scale Set Flexible across zones + a zone-redundant Standard Load Balancer |
| AKS | Node pools across zones; Standard or Premium tier uptime SLA: API server 99.95% with zones, 99.9% without; Free tier has no financially backed SLA |
| Azure SQL Database | Local redundancy by default (failover within the same datacenter); enable zone redundancy to survive a zone outage. Availability models: remote storage (General Purpose), local storage (Business Critical), Hyperscale |
| Containers, App Service, Functions | Zone-redundant configuration where the service offers it; check each service's reliability guide |
| Global | Active-active or active-passive regions behind Front Door or Traffic Manager |
Facts to know
- Microsoft's HA ranking (Learn): a single VM using only Premium SSD OS disks and Ultra, Premium SSD v2 or Premium SSD data disks has the highest single-VM uptime SLA. Multiple VMs and disks across three availability zones have the highest multi-VM SLA; across fault domains (regional Flexible scale sets or availability sets) the second-highest. Use ZRS disks so data stays accessible through a zone outage and when sharing disks between VMs.
- Managed disks are designed for 99.999% availability, replicate data three times, and provide at least 11 nines of durability (LRS) or 12 nines (ZRS). Durability and a VM's uptime SLA are different measures.
- Availability zones: 3 per supported region, each with its own power, cooling and networking; separated by several kilometers, usually within 100 km of each other.
- Without zone redundancy, an Azure SQL failover stays inside one datacenter and a datacenter-wide outage is recoverable only through DR (geo-failover or geo-restore).
Exam Tips
Tick each tip once you can answer it without looking.
“Highest VM SLA in one region” → 2+ VMs across availability zones (99.99%).
“SLA for a single VM with Premium SSD” → 99.9%.
“Region has no availability zones; need VM HA” → Availability set (99.95%).
“Web app must survive a regional outage with automatic failover” → Deploy to two regions + Front Door (or Traffic Manager).
“Composite SLA: two services in series” → Multiply SLAs (e.g. 99.99% × 99.95%); adding redundant paths raises it.
“Stateless tier must scale and survive zone failure” → VMSS Flexible spanning zones behind a zone-redundant Standard Load Balancer.
“Availability zones per region” → 3 in each supported region, each with independent power, cooling and networking. Trap: Typically within 100 km of each other.
“AKS API server SLA” → Standard and Premium tiers: 99.95% with zones, 99.9% without. Trap: Free tier has no financially backed SLA.
“Azure SQL Database default availability” → Local redundancy: failover stays inside one datacenter; enable zone redundancy for a zone outage. Trap: Without it, a datacenter-wide outage needs DR.
“Highest SLA for an application on a single VM” → Only Premium SSD OS disks and Ultra, Premium SSD v2 or Premium SSD data disks. Trap: Mixing in Standard SSD or Standard HDD lowers the single-VM tier.
“VM SLA ranking for multiple VMs” → VMs and disks across three availability zones: highest. Across fault domains (regional Flexible scale sets or availability sets): second-highest. Trap: A zone-spanning Flexible scale set is the recommended way to spread across zones.