Visualizing Insights with Datadog APM
Visualizing insights with Datadog APM
This domain tests how you turn trace data into views that reveal patterns: the Service Map, Service Catalog, service pages, flame graphs, latency distributions, dashboards, deployment tracking, SLOs, and monitors. The exam expects you to choose the visualization that answers a specific question, not merely to name widgets.
The APM visualization surfaces
The Service Map shows service-to-service calls, request rates, error rates, and latency, and it makes unhealthy dependencies obvious. The Service Catalog inventories services and their ownership and metadata. A service page combines throughput, errors, latency percentiles, dependencies, endpoints, deployments, and runtime metrics. The flame graph shows the span tree with relative durations, so you can separate time spent in a service from time spent in a downstream call. Latency distribution and percentile views reveal tail behavior that averages hide. Dashboards can use trace metrics and widgets such as timeseries, top list, query value, heatmap, and trace search. Deployment Tracking overlays version changes so you can connect a regression to a release. SLOs and monitors convert APM signals into alerting and reliability targets.
Worked example
The checkout service looks healthy on average but users report slowness. Build a dashboard with a timeseries of p95 and p99 latency for trace.http.request.duration filtered to service:checkout env:prod, a heatmap of latency by resource, a top list of errors by resource_name, and deployment events for the same service. The average stays near 200 ms while p99 rises to 2.5 s. The heatmap points to one resource, and the deployment overlay shows the change began after version 2.3.0. That combination justifies comparing versions in Trace Explorer before changing code.
Traps
Charting only average latency and missing a tail regression. Mixing environments or versions in one widget without filters. Treating one slow trace as representative of the population. Forgetting that trace metrics and indexed spans have different retention and different query behavior. Building a dashboard on indexed spans for a long-term trend when aggregated trace metrics are the better source. Ignoring deployment markers, which often explain a step change faster than any query.
Key visualization choices
Use the Service Map for dependency blast radius, the flame graph for where time went inside one request, percentiles for user-facing latency, top lists for outliers, and deployment overlays for change correlation.