A heartbeat signal is a periodic, lightweight message that each node in a compute cluster sends to its peer nodes (or to a cluster coordinator) at regular, defined intervals, essentially announcing 'I am alive and functioning.' When a node's expected heartbeat fails to arrive within a defined timeout window, the remaining cluster members recognize this absence as an indication that the node has likely failed or become unreachable, triggering appropriate cluster responses such as failover of the affected node's workloads to a healthy peer. This continuous, real-time health-signaling mechanism is precisely how cluster health and failure detection are determined, making it the correct answer. A time-to-live (TTL) value (A) is a mechanism used in networking and data expiration contexts to limit how long a packet or piece of data remains valid or how many hops it can traverse, unrelated to determining node health within a cluster. Notifications (B) are the downstream alerts generated after a health condition, such as a missed heartbeat, has already been detected; they communicate the finding but are not themselves the mechanism that determines health in the first place. Shared storage (D) is often used within clusters to provide a common data repository accessible to multiple nodes, sometimes even as a component of certain failover or quorum mechanisms, but it is not itself the health-determination signaling mechanism; that role belongs specifically to heartbeat signals. Heartbeat signal is correct.

Reference topic: Fault Tolerance Techniques - Cluster Health Monitoring via Heartbeat.