What's changed: Added per-section figures (cert-figure-retrofit). New AI-200 Chapter 2 (containers = ACR/Container Apps/AKS selection, scaling & serverless = KEDA event-driven autoscaling/Azure Functions triggers & bindings/containers vs functions)
2.2Scaling and Serverless Functions
Understand event-driven autoscaling with KEDA, serverless Azure Functions (triggers and bindings), and compute selection—in the context of AI back-ends.
AI workloads vary widely in load (batch jobs, queued tasks, peak requests). KEDA (Kubernetes Event-driven Autoscaling) autoscales to match—used internally by Azure Container Apps and AKS. KEDA scales replicas up/down based on external metrics like queue length or event count, and can scale to zero at idle, enabling cost-efficient event-driven processing.
2.2.1Serverless functions (Azure Functions)
For small single-purpose tasks (generate embeddings on file arrival, summarize on message receipt), Azure Functions fits. Functions use triggers (what starts execution: HTTP, queue, timer, Blob) and bindings (declarative connections to input/output resources) to write event-driven logic concisely, running on consumption (pay-per-use) billing. With no servers to manage, it suits intermittent, event-driven AI processing.
Common: (1) "autoscale (incl. scale-to-zero) by queue length/event count" = KEDA (event-driven autoscaling). (2) "single-purpose task on file arrival/message receipt, pay-per-use" = Azure Functions (triggers + bindings). (3) Functions: "what starts execution" = trigger, "declarative I/O connection" = binding.
Watch out: (1) trigger (what starts execution) vs binding (I/O connection)—do not confuse. (2) Choose Functions (small single-purpose, intermittent) vs containers (always-on APIs, complex workloads) by requirements. (3) KEDA is a scaling mechanism, not the compute itself. (4) Scale-to-zero is cost-efficient but watch cold-start latency.
2.2.2Section summary
- KEDA = event-driven autoscaling by external metrics like queue length/event count (scale-to-zero)
- Azure Functions = event-driven single-purpose logic via triggers + bindings, pay-per-use
- Choose: single-purpose/intermittent = Functions; always-on API/complex = containers
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. Which autoscales (incl. scale-to-zero) based on external metrics like queue length or event count?
Q2. Which Azure service runs single-purpose logic on file arrival/message receipt with pay-per-use billing?
Q3. In Azure Functions, what represents "what starts execution" (HTTP, queue, timer)?
Q4. In Azure Functions, what represents a declarative connection to input/output resources?
Q5. Which is appropriate for choosing between containers (always-on API) and serverless functions?