LazyReplicatedMappublic class LazyReplicatedMap extends AbstractReplicatedMap implements org.apache.catalina.tribes.ChannelListener, org.apache.catalina.tribes.group.RpcCallback, org.apache.catalina.tribes.MembershipListenerA smart implementation of a stateful replicated map. uses primary/secondary backup strategy.
One node is always the primary and one node is always the backup.
This map is synchronized across a cluster, and only has one backup member.
A perfect usage for this map would be a session map for a session manager in a clustered environment.
The only way to modify this list is to use the put, putAll, remove methods.
entrySet, entrySetFull, keySet, keySetFull, returns all non modifiable sets.
If objects (values) in the map change without invoking put() or remove()
the data can be distributed using two different methods:
replicate(boolean) and replicate(Object, boolean)
These two methods are very important two understand. The map can work with two set of value objects:
1. Serializable - the entire object gets serialized each time it is replicated
2. ReplicatedMapEntry - this interface allows for a isDirty() flag and to replicate diffs if desired.
Implementing the ReplicatedMapEntry interface allows you to decide what objects
get replicated and how much data gets replicated each time.
If you implement a smart AOP mechanism to detect changes in underlying objects, you can replicate
only those changes by implementing the ReplicatedMapEntry interface, and return true when isDiffable()
is invoked.
This map implementation doesn't have a background thread running to replicate changes.
If you do have changes without invoking put/remove then you need to invoke one of the following methods:
replicate(Object,boolean) - replicates only the object that belongs to the key
replicate(boolean) - Scans the entire map for changes and replicates data
the boolean value in the replicate method used to decide
whether to only replicate objects that implement the ReplicatedMapEntry interface
or to replicate all objects. If an object doesn't implement the ReplicatedMapEntry interface
each time the object gets replicated the entire object gets serialized, hence a call to replicate(true)
will replicate all objects in this map that are using this node as primary.
REMBER TO CALL breakdown() or finalize() when you are done with the map to
avoid memory leaks.
|
Fields Summary |
---|
protected static org.apache.juli.logging.Log | log |
Constructors Summary |
---|
public LazyReplicatedMap(Object owner, org.apache.catalina.tribes.Channel channel, long timeout, String mapContextName, int initialCapacity, float loadFactor, ClassLoader[] cls)Creates a new map
//------------------------------------------------------------------------------
// CONSTRUCTORS / DESTRUCTORS
//------------------------------------------------------------------------------
super(owner,channel,timeout,mapContextName,initialCapacity,loadFactor, Channel.SEND_OPTIONS_DEFAULT,cls);
| public LazyReplicatedMap(Object owner, org.apache.catalina.tribes.Channel channel, long timeout, String mapContextName, int initialCapacity, ClassLoader[] cls)Creates a new map
super(owner, channel,timeout,mapContextName,initialCapacity, LazyReplicatedMap.DEFAULT_LOAD_FACTOR, Channel.SEND_OPTIONS_DEFAULT, cls);
| public LazyReplicatedMap(Object owner, org.apache.catalina.tribes.Channel channel, long timeout, String mapContextName, ClassLoader[] cls)Creates a new map
super(owner, channel,timeout,mapContextName, LazyReplicatedMap.DEFAULT_INITIAL_CAPACITY,LazyReplicatedMap.DEFAULT_LOAD_FACTOR,Channel.SEND_OPTIONS_DEFAULT, cls);
|
Methods Summary |
---|
protected int | getStateMessageType()
return AbstractReplicatedMap.MapMessage.MSG_STATE;
| protected org.apache.catalina.tribes.Member[] | publishEntryInfo(java.lang.Object key, java.lang.Object value)publish info about a map pair (key/value) to other nodes in the cluster
if (! (key instanceof Serializable && value instanceof Serializable) ) return new Member[0];
Member[] members = getMapMembers();
int firstIdx = getNextBackupIndex();
int nextIdx = firstIdx;
Member[] backup = new Member[0];
//there are no backups
if ( members.length == 0 || firstIdx == -1 ) return backup;
boolean success = false;
do {
//select a backup node
Member next = members[nextIdx];
//increment for the next round of back up selection
nextIdx = nextIdx + 1;
if ( nextIdx >= members.length ) nextIdx = 0;
if (next == null) {
continue;
}
MapMessage msg = null;
try {
backup = wrap(next);
//publish the backup data to one node
msg = new MapMessage(getMapContextName(), MapMessage.MSG_BACKUP, false,
(Serializable) key, (Serializable) value, null, backup);
if ( log.isTraceEnabled() )
log.trace("Publishing backup data:"+msg+" to: "+next.getName());
UniqueId id = getChannel().send(backup, msg, getChannelSendOptions());
if ( log.isTraceEnabled() )
log.trace("Data published:"+msg+" msg Id:"+id);
//we published out to a backup, mark the test success
success = true;
}catch ( ChannelException x ) {
log.error("Unable to replicate backup key:"+key+" to backup:"+next+". Reason:"+x.getMessage(),x);
}
try {
//publish the data out to all nodes
Member[] proxies = excludeFromSet(backup, getMapMembers());
if (success && proxies.length > 0 ) {
msg = new MapMessage(getMapContextName(), MapMessage.MSG_PROXY, false,
(Serializable) key, null, null, backup);
if ( log.isTraceEnabled() )
log.trace("Publishing proxy data:"+msg+" to: "+Arrays.toNameString(proxies));
getChannel().send(proxies, msg, getChannelSendOptions());
}
}catch ( ChannelException x ) {
//log the error, but proceed, this should only happen if a node went down,
//and if the node went down, then it can't receive the message, the others
//should still get it.
log.error("Unable to replicate proxy key:"+key+" to backup:"+next+". Reason:"+x.getMessage(),x);
}
} while ( !success && (firstIdx!=nextIdx));
return backup;
|
|