16. Monitoring
Tracing
Application Insights receives a trace for every activity call and every worker update, with the run id and task reference. Model calls carry OpenTelemetry gen_ai attributes: model, tokens in and out, latency. Log Analytics keeps the audit trail: who started, changed, or decided what.
Custom metrics
| Metric | Dimensions |
|---|---|
queue_depth | task type |
poll_latency | task type |
task_duration | task type, status |
llm_tokens | model, workflow |
human_gate_wait | workflow |
run_duration | workflow, status |
Operator views
- The Durable Task Scheduler dashboard: raw orchestration state, for debugging.
- Azure Monitor workbooks: runs by status, queue depth, worker liveness, model spend.
- A Static Web App over the engine API: search runs, open one, see each task's input and output, apply run control.