Dynatrace insights

AI model token usage surge forecasting cost overruns

Token consumption for LLM requests increases rapidly, forecasted to exceed budget. Cost metrics lag behind usage metrics, delaying detection until overspend occurs. Root cause may be inefficient prompt engineering or lack of caching.

Provider-level guardrails (e.g., PII detection, content filtering, denied topics) activate at elevated rates, indicating prompt quality issues or user behavior changes. Without monitoring, these violations go unnoticed until user complaints surface.

Out-of-the-box AI Observability dashboards consume DPS trace query capacity via span queries, even when dashboards show no data or AI Observability is not fully configured. This leads to unexpected query costs.

Large unoptimized images or assets (e.g., background images on login portals) inflate Largest Contentful Paint (LCP), degrading user experience during traffic spikes. Infrastructure metrics show no anomalies, masking the frontend root cause.

Majority of OneAgents lose connection to Dynatrace server, causing visibility loss. Dynatrace suppresses host unavailability alerts and flags the monitoring interruption, but root cause may be ActiveGate operational issues or firewall misconfigurations.

Authentication failure spikes indicating security or integration issues

Sudden increases in authentication failures (login, MFA, password reset) may indicate security attacks, integration misconfigurations, or provider outages. Without correlation to application performance, these appear as isolated auth service issues.

Background threads consume excessive CPU (approaching 100%) while individual request latency remains within normal bounds, masking resource exhaustion that will degrade future performance.