Apache Cassandra™ 2.0

Failure detection and recovery

Failure detection is a method for locally determining, from gossip state and history, if another node in the system is up or down. Cassandra uses this information to avoid routing client requests to unreachable nodes whenever possible. (Cassandra can also avoid routing requests to nodes that are alive, but performing poorly, through the dynamic snitch.)

The gossip process tracks state from other nodes both directly (nodes gossiping directly to it) and indirectly (nodes communicated about secondhand, thirdhand, and so on). Rather than have a fixed threshold for marking failing nodes, Cassandra uses an accrual detection mechanism to calculate a per-node threshold that takes into account network performance, workload, or other conditions. During gossip exchanges, every node maintains a sliding window of inter-arrival times of gossip messages from other nodes in the cluster. In Cassandra, configuring the phi_convict_threshold property adjusts the sensitivity of the failure detector. Use default value for most situations, but increase it to 12 for Amazon EC2 (due to the frequently experienced network congestion).

Node failures can result from various causes such as hardware failures and network outages. Node outages are often transient but can last for extended intervals. A node outage rarely signifies a permanent departure from the cluster, and therefore does not automatically result in permanent removal of the node from the ring. Other nodes will periodically try to initiate gossip contact with failed nodes to see if they are back up. To permanently change a node's membership in a cluster, administrators must explicitly add or remove nodes from a Cassandra cluster using the nodetool utility.

When a node comes back online after an outage, it may have missed writes for the replica data it maintains. Once the failure detector marks a node as down, missed writes are stored by other replicas for a period of time providing hinted handoff is enabled. If a node is down for longer than max_hint_window_in_ms (3 hours by default), hints are no longer saved. Because nodes that die may have stored undelivered hints, you should run a repair after recovering a node that has been down for an extended period. Moreover, you should routinely run nodetool repair on all nodes to ensure they have consistent data.

For more explanation about recovery, see Modern hinted handoff.