When troubleshooting performance issues in an enterprise Azure OpenAI deployment using Application Insights, which TWO Kusto Query Language (KQL) queries or techniques help identify slow model responses?
Trap 1: Querying the 'performanceCounters' table for CPU usage.
Performance counters track node-level resource usage, not specific model latency.
- A
Querying the 'performanceCounters' table for CPU usage.
Why wrong: Performance counters track node-level resource usage, not specific model latency.
- B
Querying the 'availabilityResults' table for ping test response codes.
Availability results check synthetic pings, not detailed model response latency.
- C
Using 'summarize avg(duration) by cloud_RoleName' on the requests table to check application-level latency.
This summarizes the average request duration across service roles.
- D
Querying the 'dependencies' table where type == 'Http' and duration > 5000 to find slow upstream calls.
Upstream HTTP calls to Azure OpenAI are logged in the dependencies table.
- E
Checking the 'browserTimings' table for client-side rendering durations.
Browser timings relate to frontend web UI performance, not backend AI generation speed.