Package io.aeron.cluster

Aeron Cluster provides support for fault-tolerant services as replicated state machines based on the Raft consensus algorithm.

See: Description

Interface Summary

Interface	Description
ClusterBackupEventsListener	Listener which can be registered via ClusterBackup.Context for tracking backup progress.
TimerService	Timer service providing API to schedule timers in the Cluster.
TimerService.TimerHandler	Called from the TimerService.poll(long).
TimerService.TimerSnapshotTaker	Called from the TimerService.snapshot(TimerSnapshotTaker).
TimerServiceSupplier	Supplier of TimerService implementations to be used by the ConsensusModule.

Class Summary

Class	Description
AppVersionValidator	Class to be used for determining AppVersion compatibility.
ClusterBackup	Backup component which can run remote from a cluster which polls for snapshots and replicates the log.
ClusterBackup.Configuration	Configuration options for ClusterBackup with defaults and constants for system properties lookup.
ClusterBackup.Context	Context for overriding default configuration for ClusterBackup.
ClusterBackupAgent	Agent which backs up a remote cluster by replicating the log and polling for snapshots.
ClusterBackupMediaDriver	Cluster Backup media driver which is an aggregate of a MediaDriver, Archive, and a ClusterBackup.
ClusterControl	Toggle control ClusterControl.ToggleStates for a cluster node such as ClusterControl.ToggleState.SUSPEND or ClusterControl.ToggleState.RESUME.
ClusteredArchive	Clustered archive which is an aggregate of a ConsensusModule and an Archive.
ClusteredMediaDriver	Clustered media driver which is an aggregate of a MediaDriver, Archive, and a ConsensusModule.
ClusterMember	Represents a member of the cluster that participates in consensus for storing state from the perspective of any single member.
ClusterMembership	Detail for the cluster membership from the perspective of a given member.
ClusterTool	Tool for control and investigating the state of a cluster node.
ConsensusModule	Component which resides on each node and is responsible for coordinating consensus within a cluster in concert with the lifecycle of clustered services.
ConsensusModule.Configuration	Configuration options for cluster.
ConsensusModule.Context	Programmable overrides for configuring the ConsensusModule in a cluster.
ConsensusModuleSnapshotPendingServiceMessagesPatch	A tool to patch the latest consensus module snapshot if it has divergence in the pending service messages state.
MillisecondClusterClock	A ClusterClock implemented by calling System.currentTimeMillis().
NanosecondClusterClock	A ClusterClock implemented by calling HighResolutionClock.epochNanos().
NodeControl	Toggle control NodeControl.ToggleStates for a cluster node such as NodeControl.ToggleState.REPLICATE_STANDBY_SNAPSHOT.
NodeStateFile	An extensible list of information relating to a specific cluster node.
PriorityHeapTimerServiceSupplier	Supplies instances of the PriorityHeapTimerService
RecordingLog	A log of recordings which make up the history of a Raft log across leadership terms.
RecordingLog.Entry	Representation of the entry in the RecordingLog.
RecordingLog.Log	Representation of a log entry in the RecordingLog.
RecordingLog.RecoveryPlan	The snapshots and steps to recover the state of a cluster.
RecordingLog.Snapshot	Representation of a snapshot entry in the RecordingLog.
WheelTimerServiceSupplier	Supplies an instance of a WheelTimerService based on the configuration given to the constructor

Enum Summary

Enum	Description
ClusterBackup.SourceType	Defines the type of node that this will receive log data from
ClusterBackup.State	State of the cluster backup state machine.
ClusterControl.ToggleState	Toggle states for controlling the cluster node once it has entered the active state after initialising.
ConsensusModule.State	Possible states for the ConsensusModule.
ElectionState	Election states for a ConsensusModule which get represented by a ElectionState.code() stored in a Counter of the type ConsensusModule.Configuration.ELECTION_STATE_TYPE_ID.
NodeControl.ToggleState	Toggle states for controlling the cluster node once it has entered the active state after initialising.

Package io.aeron.cluster Description

Aeron Cluster provides support for fault-tolerant services as replicated state machines based on the Raft consensus algorithm.

The purpose of Aeron Cluster is to aggregate and sequence streams from cluster clients into a single log. A number of nodes will replicate and archive the log to achieve fault tolerance. ClusteredServices deterministically process the log and respond to cluster clients.

Aeron Cluster works on the concept of a strong leader using an adaptation of the Raft algorithm. The leader sequences the log and is responsible for replicating it to other cluster members known as followers.

A number of components make up Aeron Cluster. Central is the ConsensusModule which sequences the log and coordinates consensus for the recording of the sequenced log to persistent storage, and the services consuming the log across cluster members. Aeron Archive records the log to durable storage. Services consume the log once a majority of the cluster members have safely recorded the log to durable storage.

To enable fast recovery, the services and consensus module can take a snapshot of their state as of a given log position thus enabling recovery by loading the most recent snapshot and replaying logs from that point forward. Snapshots are recorded as streams in the Archive for local and remote replay so that a distributed file system is not required.

Usage

The cluster can run in various configurations:

• Single Node: For development, debugging, or when a sequenced and archived log on a single node is sufficient.
• Appointed Leader: A leader of the cluster can be appointed via configuration without requiring an election. In the event of a leader failure then a follower will have to be manually appointed the new leader. This is not the recommended way to use Cluster. Automatic elections are more reliable.
• Automatic Elections: Automatic elections (default) can be enabled to have a leader elected at random from the members with the most up to date log.

The majority of cluster members determine consensus. Clusters should typically be 3 or 5 in population size. However, 2 node clusters are supported whereby both members must agree the log and in the event of failure the remaining member must be manually reconfigured as a single node cluster to progress.

Protocol

Messages are specified using SBE in this schema aeron-cluster-codecs.xml