Welcome back to our Indispensable Azure Tools series. This time, we’re exploring Azure Compute Fleet, Microsoft’s powerful way to scale compute across regions, sizes, and pricing models.
If you’ve ever battled with capacity limits or worked with overcomplicated provisioning scripts, Compute Fleet is here to simplify your life.
What is Azure Compute Fleet
Capacity in Azure varies across regions, VM types, and over time. Popular regions like West Europe, East US, and Southeast Asia frequently hit capacity limits, especially for in-demand SKUs like the D-series or F-series. When that happens, new deployments can be delayed, quota errors may surface, or spot capacity may be evicted without warning.
Azure Compute Fleet is designed for running compute at scale, efficiently, reliably, and across multiple VM types and regions.
Azure does not guarantee unlimited capacity unless you’ve made a reservation (with reserved instances or committed use). Even then, availability zones or newer VM types might be limited or unavailable. The solution? Spread your workload across multiple VM types and regions, and let Azure decide where and how to deploy.
That’s exactly what Compute Fleet does. Some of the key features include:
- Flexible provisioning: Specify up to 15 different VM sizes, in multiple regions, and with different pricing models (spot, on demand, or mixed).
- Architecture support: Choose x64 or Arm64 based VMs.
- Eviction policies: Decide whether evicted spot VMs are deleted or stopped for reuse.
- Allocation strategies: Prioritize lowest cost or fastest provisioning.
- Price caps: Set maximum hourly rates for spot VMs to control costs.
This flexibility ensures deployments succeed even when specific regions or SKUs are under pressure, making Compute Fleet ideal for dynamic, large scale workloads.
Creating a Compute Fleet is straightforward via the Azure portal, CLI, or Bicep templates. Here’s the process:
- Define the Fleet: Name your fleet and pick a primary region (e.g. West Europe).
(You can allow Azure to span multiple regions or availability zones for resilience). - Choose the Image: Select your OS (such as Ubuntu or Windows Server) and architecture (x64 or Arm64).
- Select VM Types: List up to 15 acceptable VM sizes, for example, D2s v5, F4s v2, and E4s v5. Mix spot and on demand VMs for cost and reliability.
- Set Capacity Preferences: Specify how many VMs you need (up to 10,000 instances per fleet) and whether Azure should maintain that target if spot VMs are evicted.
- Configure Eviction Policies: Choose whether evicted spot VMs are deleted (with disks) or stopped for later use.
- Choose an Allocation Strategy: Opt for lowest price to save money or capacity optimized for provisioning speed and stability.
- Set Guardrails: Cap spot VM prices and configure admin credentials.
Azure uses this configuration to allocate VMs intelligently, balancing your preferences with available capacity. If West Europe lacks D series VMs, Azure might provision F series VMs in North Europe. The result is faster deployments, fewer failures, and optimized costs.
How does Azure Compute Fleet compare?
You may be wondering how Azure Compute Fleet differs from existing options like Virtual Machine Scale Sets (VMSS) or Spot VMs.
Virtual Machine Scale Sets are great when you’re running a known VM configuration in a fixed region or availability zone, and want to scale based on demand. However, VMSS still requires you to define a specific VM size and region. If that SKU is unavailable, your scale-out operation might fail unless you’ve built in custom logic.
Spot VMs offer significant cost savings, but on their own, they don’t handle fallback or provisioning complexity. You still need to choose the right size, region, and deal with evictions manually or through automation.
Azure Compute Fleet actually combines the best of both: it gives you the flexibility of multiple VM sizes, regional redundancy, support for both spot and standard VMs, and automatic allocation based on your priorities and configuration. You don’t have to guess which region or size is available, Compute Fleet figures that out for you.
You can compare VMSS to booking a specific seat on a flight, While Compute Fleet is like saying “Get me to my destination today, at the best available price.” And that flexibility is essential in a cloud landscape where capacity constraints are a daily reality.
Why we like Azure Compute Fleet
Azure Compute Fleet excels in scenarios like: Batch processing; crunch large datasets without capacity bottlenecks. Machine learning training; spin up hundreds of VMs for distributed training, using spot instances to save costs. Load testing; simulate massive traffic with flexible, short lived VM fleets. CI/CD pipelines; power high volume build and test environments with auto scaling compute, and for Bursty workloads; handle unpredictable spikes without over provisioning.
The benefits are compelling. You will experience fewer deployment failures as Compute Fleet will spread workloads across regions and SKUs avoids capacity issues. In many scenario’s it will be cost saving, Spot VMs can cut costs by 40% to 90% compared to on demand, with price caps for control. Compute Fleet will remove the need for complex fallback logic in templates or scripts. And it will enhanced the resilience of your environment by automatically distributing across availability zones and regions reduces downtime risks. For DevOps teams, this means less time fighting Azure’s capacity quirks and more time delivering value. You can find more information and pricing details on the official Azure Compute Fleet product page.
This post is part of the Indispensable Azure Tools series by DevOps Masterminds. Explore the full series here.
Stay tuned as we continue to highlight the tools that make building on Azure smarter, faster, and easier to manage.