DaemonSets

Some workloads don't scale by replica count. They scale with your cluster: one Pod per node, no more, no less. A log collector needs to read /var/log on every node. A metrics exporter needs access to every node's /sys and /proc. A CNI plugin must be running before any other Pod can communicate. These are node-level services, and a Deployment with a fixed replica count is the wrong abstraction for them.

That's the problem DaemonSets solve. The DaemonSet controller ensures exactly one copy of a Pod runs on every eligible node. Nodes join the cluster, they get the Pod. Nodes leave, the Pod gets garbage collected. No manual scaling, no anti-affinity hacks.