Build dashboards for app telemetry

What this covers

Dashboards give at-a-glance health for AI services: request rate, failure rate, P95 latency and dependency health. In Azure you pin charts from Azure Monitor metrics, Log Analytics / Application Insights queries, Markdown and resource tiles onto a shared Azure dashboard in the portal. This page covers pinning metrics and KQL results, how tiles refresh, sharing with RBAC, and design guidance.

Exam weight: Secure, monitor, and troubleshoot Azure solutions (20–25%)

Skills measured

• "Write KQL queries to analyze logs and metrics"

Key concepts

Azure dashboard

An Azure dashboard is a customizable portal page of tiles: metrics charts, log query results, Markdown text and items from the Tile Gallery. One dashboard can combine resources from multiple subscriptions and resource groups, for example App Insights next to AKS node health and storage throughput. Use it as the "front door" for service health on a team screen or at stand-up.

Dashboards vs workbooks

Metrics tiles vs log tiles

Common Application Insights metrics to pin: Server response time, Failed requests, Dependency failures, Server request rate, Availability. Pick the aggregation deliberately: Avg for typical latency, Sum for totals, Max for worst-case spikes. Use filters (for example one cloud role) and splitting (by operation or result code) before pinning.

Typical AI app tiles

Refresh and sharing

Tiles are not real-time; they show results as of the last scheduled refresh. During an incident use Live Metrics or a workbook. To share, select Share → Publish; the dashboard becomes an Azure resource governed by Azure RBAC. Viewers need read access to the dashboard and to each underlying data source, or the tile shows an access error.

How it works

The chart type you pick when pinning (or the render in the query) is preserved on the tile.

Hands-on

Azure portal

Pin a metric:

1. Application Insights → Metrics.
2. Choose Server response time, aggregation Avg; add a filter on cloud role name; optionally Apply splitting by operation name.
3. Select Pin to dashboard → choose an existing or new dashboard.

Pin a log query:

1. Open Application Insights → Logs.
2. Run a query ending with | render timechart (or barchart).
3. Select Pin to → Azure dashboard.
4. Open the dashboard → Edit: resize, arrange and rename tiles; add a Markdown tile; save, then Share → Publish.

Azure CLI

# Export a dashboard you built in the portal, then recreate it elsewhere
az portal dashboard show --name <dashboard-name> --resource-group <resource-group> > dashboard.json
az portal dashboard create --name <dashboard-name> --resource-group <resource-group> \
  --location eastus --input-path dashboard.json

Python

from datetime import timedelta
from azure.identity import DefaultAzureCredential
from azure.monitor.query import MetricsQueryClient

client = MetricsQueryClient(DefaultAzureCredential())
resp = client.query_resource(
    "<app-insights-resource-id>",
    metric_names=["requests/failed"],
    timespan=timedelta(hours=24),
    granularity=timedelta(hours=1),
    aggregations=["Count"],
)
for ts in resp.metrics[0].timeseries:
    print([p.count for p in ts.data])

KQL tiles

Failure rate over time (workspace names):

AppRequests
| where TimeGenerated > ago(24h)
| summarize
    Requests = count(),
    Failures = countif(Success == false)
    by bin(TimeGenerated, 1h)
| extend FailureRate = round(100.0 * Failures / Requests, 2)
| render timechart

Failed requests by service (classic names):

requests
| where timestamp > ago(24h)
| where success == false
| summarize failedCount = count() by cloud_RoleName
| render barchart

Latency percentiles; a widening gap between p50 and p95 means a subset of requests is degrading:

requests
| where timestamp > ago(24h)
| summarize p50 = percentile(duration, 50),
    p95 = percentile(duration, 95),
    p99 = percentile(duration, 99)
    by bin(timestamp, 1h)
| render timechart

Exam traps

Quick review

• Pin from Metrics, from Logs (with render), or from Application Insights blades.
• A dashboard can mix resources from multiple subscriptions.
• Metrics tiles = simple and fast; log tiles = full KQL.
• Tiles refresh periodically; use Live Metrics for real time.
• Share → Publish; RBAC applies to the dashboard and the data sources.
• Dashboards answer "is it healthy?"; workbooks answer "why?".

Practice

Practice: KQL questions

Official references

• Build dashboards for app telemetry (Learn unit)
• Create Azure dashboards
• Pin Log Analytics charts
• AI-200 study guide