A failure mode where part of the network can't reach another part, the worst-case scenario for distributed-systems correctness.
A network partition (split-brain in some literatures) is a failure mode where some nodes in a distributed system can no longer communicate with the others, even though the nodes themselves are still up. Causes include physical-link failure, cloud-provider issue, misconfigured firewall, BGP route flap. CAP theorem says under partition, you can have either consistency or availability, not both. Most production systems pick availability (return potentially-stale data) and try to reconcile when the partition heals.
Most distributed-systems bugs in production are network-partition bugs that nobody tested for. A test suite that runs everything on one host can't surface them; chaos engineering that injects partitions can. Designing for partitions (idempotent operations, conflict-free data structures, explicit reconciliation) is what makes a system survive when the network does its worst.
See the part of the platform that handles network partition in production.