By default every Pod can reach every other Pod in the cluster. NetworkPolicies, enforced by the network plugin, restrict that. A policy selects the Pods it protects with podSelector; once any policy selects a Pod for Ingress, only traffic some policy allows gets in. A policy with podSelector: {} and no rules is the usual default deny for a namespace.
Allow rules list peers in from and ports. A podSelector peer matches Pods in the policy's own namespace; a namespaceSelector matches whole namespaces, best selected by their automatic kubernetes.io/metadata.name label. In one from entry both must hold; in separate entries either is enough.
Two details catch people. Ports are the Pods' ports, because the policy is checked after the Service has translated 80 to the container's port. And peers are matched by the labels on Pods, so a client is admitted by labeling its Pod template, not its Deployment.
Only the frontend, only on the app's port.