Configure and troubleshoot Harbor pull-through caching for IPv6-only EKS Kubernetes clusters.
npx skills add https://github.com/pytorch/ci-infra --skill osdc-harbor
Please help me install this Agent Skill. Skill: osdc-harbor Source: https://github.com/pytorch/ci-infra/tree/main/osdc/.claude/skills/osdc-harbor Command: npx skills add https://github.com/pytorch/ci-infra --skill osdc-harbor
This Skill helps operators safely configure, deploy, troubleshoot, and maintain Harbor as an always-on pull-through cache for IPv6-only Kubernetes clusters, avoiding image-pull failures, Helm misconfiguration, scheduling issues, and storage corruption.
Use the osdc-harbor skill to review a Harbor deployment change for IPv6 compatibility, Helm value-path errors, scheduling problems, and image-mirroring regressions.
High-intent search queries and answers about installing and using this skill.
To fix Harbor pull-through cache manifest unknown errors causing ImagePullBackOff, diagnose storage inconsistencies, run a dry-run garbage collection, perform the cleanup, and restart the registry components. This workflow resolves cache corruption and restores image pulling.
Configure Harbor Helm values for containerd registry mirroring on IPv6-only EKS by validating chart-specific value paths, setting NodePort exposure, applying node reachability checks, and defining taint tolerations. This ensures correct registry routing and component scheduling.
You must disable S3 lifecycle expiration on registry storage backing Harbor because the Harbor storage model requires manual garbage collection. S3 lifecycle rules can delete unmanaged layers, causing manifest failures and corrupting the pull-through cache.
Yes, you can use Harbor proxy-cache projects to mirror bootstrap images in a Kubernetes cluster. This involves configuring bootstrap image mirroring alongside containerd registry routing and internally built image pushes to maintain reliable caching.
The best way to troubleshoot Harbor scheduling issues and Helm misconfigurations is to review deployment changes for Helm value-path errors, validate taint tolerations, check IAM and secret handling, and apply deployment retries to ensure proper component scheduling.
You need to run Harbor garbage collection when diagnosing cache corruption or manifest failures. The dry-run process identifies storage inconsistencies without deleting data, allowing you to verify the cleanup scope before performing the actual garbage collection and registry restart.
aws-samplesAutomate AWS EKS cluster setup with GPU node pools and observability.
thc1006Deploy Kubernetes clusters on DGX Spark ARM64 with Kubespray.
featbitDeploy FeatBit to Kubernetes using Helm charts with configurable tiers and exposure methods.
Nir-BhayConfigure and manage container registries with Terraform, Docker Compose, and CI/CD pipelines.
