fix: resolve critical alerts — Talos compat, ArgoCD OOM, Longhorn replicas #130

Merged

claude-agent merged 1 commit from fix/critical-alerts-triage-homelab-542 into main

2026-04-02 03:43:13 +00:00

claude-agent commented

2026-04-02 03:43:01 +00:00

Owner

Summary

Disable false-positive control plane alerts — Talos binds kube-controller-manager/scheduler to localhost and uses Cilium (no kube-proxy). Disables ServiceMonitors.
Bump ArgoCD controller memory — 1Gi → 2Gi to prevent OOMKills.
Fix Longhorn replica settings — Default count=2, enable auto-balance, concurrent rebuild=5.

Runtime fixes applied

Patched 18 volumes from 3→2 replicas
Deleted 21 stuck replicas, Longhorn rebuilding
Deleted stale ollama-pull-32b failed job

Part of HOMELAB-542.

## Summary - **Disable false-positive control plane alerts** — Talos binds kube-controller-manager/scheduler to localhost and uses Cilium (no kube-proxy). Disables ServiceMonitors. - **Bump ArgoCD controller memory** — 1Gi → 2Gi to prevent OOMKills. - **Fix Longhorn replica settings** — Default count=2, enable auto-balance, concurrent rebuild=5. ## Runtime fixes applied - Patched 18 volumes from 3→2 replicas - Deleted 21 stuck replicas, Longhorn rebuilding - Deleted stale ollama-pull-32b failed job Part of HOMELAB-542.

claude-agent added 1 commit

2026-04-02 03:43:01 +00:00

fix: resolve critical alerts — Talos compat, ArgoCD OOM, Longhorn replicas

0/0 projects applied successfully.

Release / release (pull_request) Has been cancelled

Details

18a00e9d13

- Disable kubeControllerManager, kubeScheduler, kubeProxy ServiceMonitors
  in kube-prometheus-stack — Talos binds these to localhost (unscrappable)
  and uses Cilium instead of kube-proxy, eliminating false critical alerts
- Bump ArgoCD application-controller memory limit 1Gi → 2Gi to prevent
  OOMKills under load with many managed applications
- Set Longhorn defaultReplicaCount to 2 (matches 2-node storage topology),
  enable replicaAutoBalance and bump concurrent rebuild limit to 5

Part of HOMELAB-542 infrastructure alerts triage.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>