Issue Detection & AI Analysis
Understand how Optral detects issues and provides AI-powered root cause analysis.
How It Works
Optral continuously monitors your Kubernetes clusters for issues, anomalies, and potential problems. When an issue is detected, our AI analyzes the context, related events, and historical patterns to provide actionable insights and recommended fixes.
Detected Issue Types
Optral automatically detects these common Kubernetes issues
OOMKilled
Container killed due to memory limit exceeded
CrashLoopBackOff
Container repeatedly crashing and restarting
HighMemory
Memory usage approaching or exceeding thresholds
HighCPU
CPU usage consistently above threshold
PodPending
Pod unable to be scheduled for extended time
ImagePullBackOff
Failed to pull container image
NodeNotReady
Node reporting unhealthy status
PVCPending
Persistent volume claim waiting for binding
Severity Levels
Immediate attention required
Service is down, data at risk, or security issue. Examples: OOMKilled in production, node failures.
Degraded performance or reliability risk
Service impacted but functional. Examples: High memory usage, frequent restarts.
Potential issue developing
Not urgent but should be addressed. Examples: Pending pods, configuration drift.
Informational or best practice
Improvements recommended. Examples: Missing resource requests, outdated images.
AI-Powered Analysis
How our AI analyzes and explains issues
When you click "Analyze" on an issue, Optral's AI examines:
Resource Context
Pod specs, container logs, events, and related resources
Timeline
Recent changes, deployments, and correlated events
Metrics
CPU, memory, network patterns leading up to the issue
Best Practices
Kubernetes best practices and known issue patterns
Analysis Output:
- Summary: Plain-language explanation of what happened
- Root Cause: Most likely cause based on evidence
- Impact: What services or users are affected
- Recommendations: Specific steps to resolve the issue
- Prevention: How to prevent recurrence
Issue Lifecycle
Issues are automatically marked as resolved when the underlying condition clears. You can also manually resolve or acknowledge issues.