Interview Preparation

Monitoring tells you that something is wrong using predefined thresholds. Observability tells you why using rich telemetry — including for failure modes never anticipated in advance.

Metrics (numeric time-series), Logs (event records), Traces (end-to-end request paths). Together they provide complete system understanding. Correlation between all three accelerates root cause analysis.

CNCF standard SDK and protocol for vendor-neutral telemetry — collect metrics, logs, and traces once with OTel and export to any compatible backend (Dynatrace supports OTel natively).

Rate, Errors, Duration — a minimal but sufficient set of metrics to characterise any request-based service health. Dynatrace tracks all three automatically per service.

Monitoring systems from the outside (HTTP probes, synthetic tests) without internal instrumentation. Detects user-visible availability issues. Complements white-box (internal APM/traces) monitoring.

A single host agent that automatically instruments all supported processes, collects code-level metrics and traces without config, monitors OS and network metrics, and sends all telemetry to the Dynatrace cluster.

Dynatrace's proxy and extension component — routes OneAgent data to the cluster, polls cloud provider APIs (AWS/Azure/GCP), runs synthetic monitors from private networks, and hosts the extension framework.

Dynatrace's automatically maintained real-time topology map — every host, process, service, and application with every dependency relationship. Updated continuously, no manual configuration.

Monitoring-as-Code CLI tool — store all Dynatrace configuration (dashboards, alerts, SLOs, synthetics) in Git and deploy through CI/CD pipelines with environment-specific variable substitution.

Via the Dynatrace Operator and DynaKube CRD — OneAgent is injected into pods automatically (cloudNativeFullStack mode), and the Operator polls the K8s API for cluster/namespace/workload metadata.

Application Performance Index — 0–1 score measuring user satisfaction: Satisfied (response < T), Tolerating (T–4T), Frustrated (>4T or error). Formula: (Satisfied + Tolerating/2) / Total.

Dynatrace's term for a distributed trace — automatically captured by OneAgent with method-level granularity inside each service, including every DB call and external request, without any SDK instrumentation.

Open a PurePath waterfall for a slow request. If you see many repeated identical database spans (e.g., 50 SELECT calls with the same table), that's N+1. Check the "DB calls per request" service metric for a persistent high value.

Dynatrace always captures errored and slow traces (100%), and samples healthy fast requests. This ensures you never miss a problematic transaction even at 10,000 req/sec production volumes.

A view showing how a specific request type flows through all services — with per-hop throughput, response time, and error rate. Different from Smartscape which shows all dependency types.

Davis uses the Smartscape dependency graph to trace anomalies backwards to their origin. If multiple services degrade simultaneously, Davis identifies the single root entity whose anomaly was first and propagates through dependencies.

Static thresholds fire during planned load increases (weekly spikes, deployments). Dynamic baselines adapt to time-of-day and weekly patterns — only firing when behaviour is genuinely anomalous, dramatically reducing false positives.

CPU throttling is when a container exceeds its CPU limit and is paused by cgroups. It causes latency spikes in APM (long response times) even when node-level CPU looks fine — because the container is waiting for execution time.

Davis traverses Smartscape to link infrastructure anomalies (e.g., disk full on host-X) with the services running on that host. The problem card lists the infrastructure anomaly as the root cause and all impacted application services in the blast radius.

The allowed amount of SLO violation within a period. If SLO is 99.9%, the monthly error budget is ~43 minutes. Burning it fast triggers risk-reduction actions. Dynatrace SLOs track error budget consumption in real time.

Open Davis problem card — it will list the deployment as a contributing event and identify the root cause. Use Smartscape to confirm the new service is a dependency. Open PurePaths for the 5 degraded services and look for calls to the new service with high latency or errors.

Filter PurePaths for duration >5000ms. Analyse the waterfall of slow traces — look for a specific span that is long in slow traces but short in fast ones. Common causes: cache cold starts, DB lock contention, thread pool exhaustion on a specific code path.

Deploy Dynatrace OneAgent via Operator on Kubernetes. Use Davis AI for anomaly detection (no threshold configuration). Set SLOs for core user journeys. Use Monaco for configuration-as-code. Integrate PagerDuty via problem notifications for on-call routing. Add OTel SDK for any custom business metrics.

Two approaches: 1. Push a deployment event to Dynatrace before the deploy — Davis will correlate anomalies with the deployment as context rather than independent problems. 2. Create a short maintenance window matching the deployment window to suppress problem generation during the stabilisation period.

Davis AI eliminates manual correlation time — delivering a root cause identified problem card instead of 100 raw alerts. PurePath provides instant latency attribution. Smartscape shows the blast radius immediately. Together these replace the 30+ minutes of manual correlation with automatic analysis delivered in seconds.