Configuring telemetry collection
Configure collection of telemetry
Each Insights product covers a different layer, and the objective names them explicitly.
| Product | Collects |
|---|---|
| Application Insights | Application telemetry — requests, dependencies, exceptions, traces, custom events |
| VM Insights | Virtual machine performance and dependency mapping |
| Container Insights | AKS and container workload performance |
| Azure Monitor for Storage | Storage account performance and availability |
| Azure Monitor for Networks | Network health and connectivity |
Match the product to the question
A slow checkout is an application question — Application Insights, looking at request duration and dependency calls. A node running out of memory is an infrastructure question — Container or VM Insights.
Instrumenting only infrastructure is the common gap: you learn that CPU is fine while users are failing, because nothing is watching the request path.
Sampling
High-volume applications sample telemetry to control cost. Sampling is a trade: it preserves statistical shape and loses individual events, so a specific failed request may not be there when you look for it. Know the sampling rate before promising you can trace any given transaction.
Instrument early
Telemetry added after an incident does not help with that incident. Instrumentation belongs in the definition of done, not in the post-mortem actions.
Primary sources