Skip to content

Slowest Requests by Path

Scenario: Users report slowness, but only for some endpoints. Data Source: AppServiceHTTPLogs Purpose: Ranks request paths by tail latency to identify endpoint-level hotspots.

graph TD
    A[AppServiceHTTPLogs] --> B[Group by CsUriStem]
    B --> C[Avg + P95 Latency per Path]
    C --> D[Top 20 Slowest Endpoints]

Run It in the Portal

The query terminates in top 20 by P95 desc, so its output is a short ranked table rather than a timechart. Tighten the blade-level Time range to Last hour so it matches the ago(1h) filter inside the query, otherwise P95 is computed over a wider window than the interpretation notes assume.

Query

AppServiceHTTPLogs
| where TimeGenerated > ago(1h)
| summarize AvgTime=avg(TimeTaken), P95=percentile(TimeTaken, 95), Count=count() by CsUriStem
| top 20 by P95 desc

Interpretation Notes

  • Normal: top paths have expected P95 relative to their workload type and stable request counts.
  • Abnormal: one/few paths show disproportionately high P95 with meaningful request volume.
  • Reading tip: prioritize paths with both high P95 and non-trivial Count to avoid chasing one-off outliers.

Limitations

  • Freshness depends on logging pipeline delay.
  • Paths with very low count can produce unstable percentile values.
  • This query cannot reveal internal function-level bottlenecks inside a path.

See Also

Sources