CPU and Memory Scaling

Question

Hey Team,

I have a query regarding CPU and Memory Scaling for our application on Azure container apps. I have already configured the setup to handle concurrent requests, but I now need to implement scaling based on resource utilization.

Could you please provide the steps or configuration details required to set up CPU and Memory-based scaling?
and please let me know the costing. if there is any cheaper method please let me know that too.

Answer 1

Refer to https://learn.microsoft.com/en-us/azure/container-apps/tutorial-scaling?tabs=bash#cpu-and-memory-scaling

Regarding cost, Azure Container Apps pricing is consumption-based, depending on:

Active usage — vCPU and memory allocated while the app is running.

Idle usage — when your app has no active requests but still retains minimum resources.

Requests and executions — small per-request charge (if using event-driven scaling).

Resource Type	Price (approx.)	Notes
vCPU (active)	~$0.000024 per vCPU-second (~$0.086/hour)	Billed per second of use
--------	--------	--------
vCPU (active)	~$0.000024 per vCPU-second (~$0.086/hour)	Billed per second of use
Memory (active)	~$0.000003 per GiB-second (~$0.011/hour)	Billed per second of use
Idle vCPU	~1/10th of active cost	Only if min replicas > 0
Idle Memory	~1/10th of active cost	Same as above
Environment surcharge	~$0.004 per vCPU per hour	Applies to total environment usage

If you're looking to minimize the cost, consider the following:

1. Use 0 minimum replicas — scale to zero when idle → pay only when traffic arrives.
2. Use smaller container sizes — e.g., 0.25 vCPU / 0.5 GiB instead of 1 vCPU / 2 GiB.
3. Use Azure Functions (Consumption Plan) — for sporadic workloads, it’s cheaper since it bills only per execution time, not for reserved CPU/memory.
4. Use Azure Kubernetes Service (AKS) with Spot Nodes — cheaper for batch or non-critical workloads but less predictable availability.
5. Use Pre-configured autoscaling rules carefully — avoid over-provisioning; use metrics like 80% CPU instead of 50%.

---

If the above response helps answer your question, remember to "Accept Answer" so that others in the community facing similar issues can easily find the solution. Your contribution is highly appreciated.

hth

Marcin