| org.jgroups.stack.RpcProtocol org.jgroups.protocols.FLUSH
FLUSH | public class FLUSH extends RpcProtocol (Code) | | The task of the FLUSH protocol is to flush all pending messages out of the system. This is
done before a view change by stopping all senders and then agreeing on what messages
should be delivered in the current view (before switching to the new view). A coordinator
broadcasts a FLUSH message. The message contains an array of the highest sequence number for each member
as seen by the coordinator so far. Each member responds with its highest sequence numbers seen so far (for
each member): if its sequence number for a member P is higher than the one sent by the coordinator, it
will append the messages apparently not received by the coordinator to its reply. The coordinator (when
all replies have been received), computes for each member the lowest and highest sequence number and
re-broadcasts messages accordingly (using ACKs rather then NAKs to ensure reliable delivery). Example:
FLUSH ---> (p=10, q=22, r=7)
<-- (p=10, q=20, r=7) (did not receive 2 messages from q)
<-- (p=12, q=23, r=7) (attached are messages p11, p12, and q23)
<-- (p=10, q=22, r=8) (attached is message r8)
---------------------
min: 11 21 8
max: 12 23 8
The coordinator now computes the range for each member and re-broadcasts messages
p11, p12, q21, q22, q23 and r8.
This is essentially the exclusive min and inclusive max of all replies. Note that messages p11, p12 and q23
were not received by the coordinator itself before. They were only returned as result of the FLUSH replies
and the coordinator now re-broadcasts them.
|
block_timeout | long block_timeout(Code) | | |
digest_timeout | long digest_timeout(Code) | | |
get_msgs_timeout | final long get_msgs_timeout(Code) | | |
highest_delivered_msgs | long[] highest_delivered_msgs(Code) | | |
highest_delivered_mutex | final Object highest_delivered_mutex(Code) | | |
is_server | boolean is_server(Code) | | |
getHighestDeliveredSeqnos | void getHighestDeliveredSeqnos()(Code) | | Returns the highest seqnos (for each member) seen so far (using the NAKACK layer)
|
getMessageDigest | void getMessageDigest(long[] highest_seqnos)(Code) | | Interacts with a lower layer to retrieve unstable messages (e.g. NAKACK)
|
handleDownEvent | public boolean handleDownEvent(Event evt)(Code) | | Callback. Called by superclass when event may be handled.
Do not use PassDown in this method as the event is passed down
by default by the superclass after this method returns !
boolean Defaults to true. If false, event will not be passed down the stack. |
handleFlush | public synchronized Digest handleFlush(Vector flush_dests, long[] highest_seqnos)(Code) | | Called by coordinator running the FLUSH protocol. Argument is an array of the highest seqnos as seen
by the coordinator (for each member). handleFlush() checks for each member its
own highest seqno seen for that member. If it is higher than the one seen by the coordinator,
all higher messages are attached to the return value (a message digest).
Parameters: flush_dests - The members to which this message is sent. Processes not in this list justignore the handleFlush() message. Parameters: highest_seqnos - The highest sequence numbers (order corresponding to membership) as seenby coordinator. Digest An array of the highest seqnos for each member, as seen by this member. If thismember's seqno for a member P is higher than the one in highest_seqnos ,the missing messages are added to the message digest as well. This allows thecoordinator to re-broadcast missing messages. |
handleUpEvent | public boolean handleUpEvent(Event evt)(Code) | | Callback. Called by superclass when event may be handled.
Do not use PassUp in this method as the event is passed up
by default by the superclass after this method returns !
boolean Defaults to true. If false, event will not be passed up the stack. |
receiveDownEvent | public void receiveDownEvent(Event evt)(Code) | | The default handling adds the event to the down-queue where events are handled in order of
addition by a thread. However, there exists a deadlock between the FLUSH and BLOCK_OK down
events: when a FLUSH event is received, a BLOCK is sent up, which triggers a BLOCK_OK event
to be sent down to be handled by the FLUSH layer. However, the FLUSH layer's thread is still
processing the FLUSH down event and is therefore blocked, waiting for a BLOCK_OK event.
Therefore, the BLOCK_OK event has to 'preempt' the FLUSH event processing. This is done by
overriding this method: when a BLOCK_OK event is received, it is processed immediately
(in parallel to the FLUSH event), which causes the FLUSH event processing to return.
|
requiredDownServices | public Vector requiredDownServices()(Code) | | |
Methods inherited from org.jgroups.stack.RpcProtocol | public Object callRemoteMethod(Address dest, String method_name, int mode, long timeout) throws TimeoutException, SuspectedException(Code)(Java Doc) public Object callRemoteMethod(Address dest, String method_name, Object[] args, Class[] types, int mode, long timeout) throws TimeoutException, SuspectedException(Code)(Java Doc) public Object callRemoteMethod(Address dest, String method_name, Object[] args, String[] signature, int mode, long timeout) throws TimeoutException, SuspectedException(Code)(Java Doc) public Object callRemoteMethod(Address dest, MethodCall method_call, int mode, long timeout) throws TimeoutException, SuspectedException(Code)(Java Doc) public RspList callRemoteMethods(Vector dests, String method_name, Object[] args, Class[] types, int mode, long timeout)(Code)(Java Doc) public RspList callRemoteMethods(Vector dests, String method_name, Object[] args, String[] signature, int mode, long timeout)(Code)(Java Doc) public RspList callRemoteMethods(Vector dests, MethodCall method_call, int mode, long timeout)(Code)(Java Doc) public String getName()(Code)(Java Doc) public Object handle(Message req)(Code)(Java Doc) public boolean handleDownEvent(Event evt)(Code)(Java Doc) public boolean handleUpEvent(Event evt)(Code)(Java Doc)
|
|
|