How harshavarthan-t/ai-kubernetes-incident-triage works
This is a local Kubernetes incident-triage starter, not yet an automated remediation system. Today’s workflow is an Alertmanager-style webhook creating a JSON incident, followed by retrieval by ID and recording an approval or rejection. The README also describes a sample workload, monitoring setup, runbooks, and planned evidence collection, runbook search, model analysis, and approved recovery. Those planned stages are shown as unimplemented; the README says monitoring is not connected to the API, and approval does not execute cluster actions.
Main entry points: main.py Triage API · README.md Monitoring stack · runbooks Troubleshooting runbooks
Components 11
Triage API
- Triage API
app/triage-api/triage/main.py - Incident JSON files
app/triage-api/triage/main.py
Cluster Monitoring
- Sample workload
k8s/sample-app/deployment.yaml - Monitoring stack
README.md
Diagnosis and Recovery
- Evidence collection
- Troubleshooting runbooks
runbooks - Runbook search
- Model analysis
- Approved recovery
Other components
- Alertmanager
- Local operator
Connections 11
- Alertmanager → Triage API: posts alert
README.md - Triage API → Incident JSON files: writes incident
app/triage-api/triage/main.py - Local operator → Triage API: retrieves incident
app/triage-api/triage/main.py - Local operator → Triage API: records approval
app/triage-api/triage/main.py - Triage API → Incident JSON files: loads incident
app/triage-api/triage/main.py - Monitoring stack → Alertmanager: forwards alerts
README.md - Evidence collection → Runbook search: informs search
- Runbook search → Troubleshooting runbooks: searches
- Runbook search → Model analysis: supplies context
- Model analysis → Approved recovery: suggests action
- Local operator → Approved recovery: approves action