// the find
senthilrch/kube-fledged
A kubernetes operator for creating and managing a cache of container images directly on the cluster worker nodes, so application pods start almost instantly
kube-fledged is a Kubernetes operator that pre-pulls container images onto worker nodes so pods start without waiting for registry pulls. It introduces a custom `ImageCache` CRD and manages image lifecycle via Kubernetes jobs. Most useful for latency-sensitive workloads, edge deployments with unreliable connectivity to registries, or clusters where cold-start time matters.
The CRI-agnostic approach (docker, containerd, cri-o) means it works across most cluster setups without modification. Using Kubernetes jobs for the actual pull/delete operations is the right call — it offloads scheduling, retries, and node targeting to the control plane rather than rolling a custom worker. The refresh mechanism with configurable frequency and on-demand annotation is genuinely useful for keeping cached images current. Helm chart with GPG signature verification shows someone thought about supply chain.
Last commit was February 2024 and the project shows signs of stagnation — Travis CI badges instead of GitHub Actions (the workflows exist but the README still references Travis), and the CRD version is v1alpha2, suggesting the API never stabilized. The purge-before-delete workflow is awkward UX: you must annotate to purge, wait, check status, then delete — a finalizer would handle this automatically. Cache consistency on node failure is not addressed in the docs; if a node goes down and comes back, it's unclear whether the image manager reconciles correctly or leaves the ImageCache status in a confused state. No mention of OCI artifact support or image signature verification, which is now table stakes for anything security-conscious.