Services and cluster networking

Pods come and go, services do not

Pod ip addresses change on every restart, so nothing should ever be configured with one. A service provides a stable name and address, and load balances across whichever pods currently match its selector.

The types stack: a cluster-internal service is the default, a node port exposes it on every node, and a load balancer asks the cloud provider for an external one. Most application services should stay internal, with an ingress in front.

The selector is the whole binding mechanism. A label typo produces a service with no endpoints, which fails as a connection timeout rather than an obvious error.