A network partition, or netsplit, occurs when nodes that previously communicated can no longer do so. The cause may be a lost or slow connection, a crashed node, or an overloaded node; from the remaining node’s perspective, these failures look the same. section 12.2.5
A partition can create split-brain: disconnected clusters continue serving clients independently and accept conflicting updates. Reliable distributed systems therefore need an explicit partition-handling strategy rather than assuming the network will remain available. section 12.2.1 section 12.2.5
Practical consequences include:
- Treat an unresponsive remote node as potentially partitioned, without assuming the precise underlying failure. section 12.2.5
- Prepare distributed services for conflicting cluster state and choose how replication or clustering behaves during disconnection. section 12.2.1 section 12.4
$ iex --sname node1@localhost iex(node1@localhost)1> :net_kernel.monitor_nodes(true)
Monitor node connectivity in IEx with :net_kernel.monitor_nodes(true), then observe nodeup and nodedown events as nodes connect or disappear. section 12.2.5
