org.apache.brooklyn.api.mgmt.ha

Interface HighAvailabilityManager

All Known Implementing Classes:

HighAvailabilityManagerImpl

public interface HighAvailabilityManager

Monitors other management nodes (via the ManagementPlaneSyncRecordPersister) to detect if the current master has failed or stopped. If so, then deterministically chooses a new master. If that master is self, then promotes. Users are not expected to implement this class, or to call methods on it directly. Expected lifecycle of methods calls on this is:

1. setPersister(ManagementPlaneSyncRecordPersister)
2. Exactly one of #disabled() or start(HighAvailabilityMode)
3. stop()

Since:

0.7.0

Method Summary

Modifier and Type	Method and Description
void	changeMode(HighAvailabilityMode mode) changes the mode that this HA server is running in
void	disabled(boolean persistenceEnabled) Indicates that HA is disabled: this node will act as the only management node in this management plane, and will not persist HA meta-information (meaning other nodes cannot join).
ManagementPlaneSyncRecord	getLastManagementPlaneSyncRecord() Returns a snapshot of the management-plane's current / most-recently-known status, as last read from loadManagementPlaneSyncRecord(boolean), or null if none read.
long	getLastStateChange() The time in milliseconds when the state was last changed.
java.util.Map<java.lang.String,java.lang.Object>	getMetrics() Returns a collection of metrics
ManagementNodeState	getNodeState() Returns the current node state, including ManagementNodeState.INITIALIZING on some transitions, but may return ManagementNodeState.MASTER before server is persisting.
ManagementPlaneSyncRecordPersister	getPersister()
long	getPriority()
boolean	isRunning() Whether HA mode is operational
ManagementPlaneSyncRecord	loadManagementPlaneSyncRecord(boolean useLocalKnowledgeForThisNode)
void	publishClearNonMaster() deletes non-master node records; active nodes (including this) will republish, so this provides a simple way to clean out the cache of dead brooklyn nodes
HighAvailabilityManager	setPersister(ManagementPlaneSyncRecordPersister persister)
void	setPriority(long priority) sets the priority, and publishes it synchronously so it is canonical
void	start(HighAvailabilityMode startMode) Starts the monitoring of other nodes (and thus potential promotion of this node from standby to master).
void	stop() Stops this node, then publishes own status (via ManagementPlaneSyncRecordPersister of ManagementNodeState.TERMINATED.

Method Detail

getNodeState

ManagementNodeState getNodeState()

Returns the current node state, including ManagementNodeState.INITIALIZING on some transitions, but may return ManagementNodeState.MASTER before server is persisting.

See ManagementContext.getNodeState() to ensure ManagementNodeState.MASTER is definitive and ready for use.

getLastStateChange

long getLastStateChange()

The time in milliseconds when the state was last changed. -1 if no state transition has occurred yet.

setPersister

HighAvailabilityManager setPersister(ManagementPlaneSyncRecordPersister persister)

Parameters:

persister -

Returns:

self

disabled

void disabled(boolean persistenceEnabled)

Indicates that HA is disabled: this node will act as the only management node in this management plane, and will not persist HA meta-information (meaning other nodes cannot join).

Subsequently can expect getNodeState() to be ManagementNodeState.MASTER and loadManagementPlaneSyncRecord(boolean) to show just this one node -- as if it were running HA with just one node -- but isRunning() will return false.

Currently this method is intended to be called early in the lifecycle, instead of start(HighAvailabilityMode). It may be an error if this is called after this HA Manager is started.

isRunning

boolean isRunning()

Whether HA mode is operational

start

void start(HighAvailabilityMode startMode)

Starts the monitoring of other nodes (and thus potential promotion of this node from standby to master).

When this method returns, the status of this node will be set, either ManagementNodeState.MASTER if appropriate or ManagementNodeState.STANDBY / ManagementNodeState.HOT_STANDBY / ManagementNodeState.HOT_BACKUP.

Parameters:

startMode - mode to start with

Throws:

java.lang.IllegalStateException - if current state of the management-plane doesn't match that desired by startMode

stop

void stop()

Stops this node, then publishes own status (via ManagementPlaneSyncRecordPersister of ManagementNodeState.TERMINATED.

changeMode

void changeMode(HighAvailabilityMode mode)

changes the mode that this HA server is running in

note it will be an error to changeMode(HighAvailabilityMode) to ManagementNodeState.MASTER when there is already a master; to promote a node explicitly set its priority higher than the others and invoke changeMode(HighAvailabilityMode) to a standby mode on the existing master

setPriority

void setPriority(long priority)

sets the priority, and publishes it synchronously so it is canonical

getPriority

long getPriority()

publishClearNonMaster

void publishClearNonMaster()

deletes non-master node records; active nodes (including this) will republish, so this provides a simple way to clean out the cache of dead brooklyn nodes

getLastManagementPlaneSyncRecord

ManagementPlaneSyncRecord getLastManagementPlaneSyncRecord()

Returns a snapshot of the management-plane's current / most-recently-known status, as last read from loadManagementPlaneSyncRecord(boolean), or null if none read.

loadManagementPlaneSyncRecord

ManagementPlaneSyncRecord loadManagementPlaneSyncRecord(boolean useLocalKnowledgeForThisNode)

Parameters:

useLocalKnowledgeForThisNode - - if true, the record for this mgmt node will be replaced with the actual current status known in this JVM (may be more recent than what is persisted); for most purposes there is little difference but in some cases the local node being updated may be explicitly wanted or not wanted

getPersister

ManagementPlaneSyncRecordPersister getPersister()

getMetrics

java.util.Map<java.lang.String,java.lang.Object> getMetrics()

Returns a collection of metrics