antiaffinity

Antiaffinity is a scheduling constraint in distributed systems that prevents certain entities from sharing the same resource, such as a node, host, or zone. It is used to improve fault tolerance, isolation, and resource distribution by avoiding co-location of related workloads.

In container orchestration platforms like Kubernetes, pod anti-affinity enables label-based rules that restrict where pods may be scheduled relative to other pods. These rules can be hard or soft: strict constraints use components like RequiredDuringSchedulingIgnoredDuringExecution, while softer preferences use PreferredDuringSchedulingIgnoredDuringExecution. A topologyKey defines the dimension over which to spread pods (for example, kubernetes.io/hostname for node-level distribution or topology.kubernetes.io/zone for zone-level distribution). Labels on the target pods determine which pods to avoid co-locating with.

Common use cases include spreading replicas of a service across nodes or zones to reduce the risk of a single point of failure, and avoiding contention for shared resources on a node. Anti-affinity can also help satisfy policy or licensing requirements that limit co-location of certain workloads.

Limitations include increased scheduling complexity and the possibility of unschedulable pods if capacity is insufficient. Rigid hard anti-affinity rules can reduce utilization; soft preferences guide the scheduler while allowing co-location if necessary. Anti-affinity should be considered alongside affinity rules and resource requests to avoid conflicts.

Across platforms, anti-affinity is a general concept with platform-specific implementations, notably in Kubernetes through PodAntiAffinity.