7 Real-World Kubernetes Failures on AWS EKS (Argo CD + Helm) — Diagnose & Fix Like a DevOps Engineer
The video demonstrates diagnosing and fixing seven common real-world Kubernetes failures across nine microservices running on AWS EKS and deployed via Argo CD with Helm charts managed through a GitOps GitHub repo. The presenter uses kubectl and Argo CD events/logs to troubleshoot issues such as ImagePullBackOff (wrong image tag and incorrect ECR repository name), CrashLoopBackOff due to OOMKilled (exit code 137 from low memory limits), Pending pods from insufficient cluster resources (overstated memory requests and high replica count), and services stuck at 0/1 because of failing readiness or liveness probes (wrong ports/paths and overly aggressive probe timings). A Helm template/indentation error causing Argo CD sync failure prevents pod creation for one service until corrected. The episode emphasizes starting with pod describe/events before logs and fixing problems by updating Helm values in GitOps.
00:00 Kubernetes Failures Overview
00:47 Architecture and Argo CD Setup
01:42 Common Error Cheat Sheet
03:28 Fix ImagePullBackOff Auth Service
06:44 GitOps Sync and Autosync Settings
09:16 Fix Catalog Image and Memory
14:49 Inventory Pod Pending Diagnosis
15:29 Fix Memory Limits
16:37 Adjust Replica Count
17:06 Sync Changes ArgoCD
17:43 Verify Inventory Pod
18:30 Manufacturing Probe Port
22:05 QC Liveness Path
24:03 Tune QC Probe Timings
25:47 Notification Readiness Fix
27:28 Supplier Helm Syntax
29:49 Wrap Up Troubleshooting