package org.jgroups.protocols.pbcast;
import org.jgroups.Address;
import org.jgroups.Event;
import org.jgroups.Message;
import org.jgroups.View;
import org.jgroups.annotations.*;
import org.jgroups.protocols.TP;
import org.jgroups.stack.DiagnosticsHandler;
import org.jgroups.stack.Protocol;
import org.jgroups.util.*;
import java.util.*;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.Executor;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicLong;
import java.util.concurrent.locks.Condition;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock;
/**
* Negative AcKnowledgement layer (NAKs). Messages are assigned a monotonically
* increasing sequence number (seqno). Receivers deliver messages ordered
* according to seqno and request retransmission of missing messages.<br/>
* Retransmit requests are usually sent to the original sender of a message, but
* this can be changed by xmit_from_random_member (send to random member) or
* use_mcast_xmit_req (send to everyone). Responses can also be sent to everyone
* instead of the requester by setting use_mcast_xmit to true.
*
* @author Bela Ban
*/
@MBean(description="Reliable transmission multipoint FIFO protocol")
public class NAKACK2 extends Protocol implements DiagnosticsHandler.ProbeHandler {
protected static final int NUM_REBROADCAST_MSGS=3;
/* ----------------------------------------------------- Properties --------------------- ------------------------------------ */
@Property(description="Max number of messages to be removed from a RingBuffer. This property might " +
"get removed anytime, so don't use it !")
protected int max_msg_batch_size=100;
/**
* Retransmit messages using multicast rather than unicast. This has the advantage that, if many receivers
* lost a message, the sender only retransmits once
*/
@Property(description="Retransmit retransmit responses (messages) using multicast rather than unicast")
protected boolean use_mcast_xmit=true;
/**
* Use a multicast to request retransmission of missing messages. This may
* be costly as every member in the cluster will send a response
*/
@Property(description="Use a multicast to request retransmission of missing messages")
protected boolean use_mcast_xmit_req=false;
/**
* Ask a random member for retransmission of a missing message. If set to
* true, discard_delivered_msgs will be set to false
*/
@Property(description="Ask a random member for retransmission of a missing message. Default is false")
protected boolean xmit_from_random_member=false;
/**
* Messages that have been received in order are sent up the stack (= delivered to the application).
* Delivered messages are removed from the retransmission buffer, so they can get GC'ed by the JVM. When this
* property is true, everyone (except the sender of a message) removes the message from their retransission
* buffers as soon as it has been delivered to the application
*/
@Property(description="Should messages delivered to application be discarded")
protected boolean discard_delivered_msgs=true;
@Property(description="Timeout to rebroadcast messages. Default is 2000 msec")
protected long max_rebroadcast_timeout=2000;
/**
* When not finding a message on an XMIT request, include the last N
* stability messages in the error message
*/
@Property(description="Should stability history be printed if we fail in retransmission. Default is false")
protected boolean print_stability_history_on_failed_xmit=false;
/** If true, logs messages discarded because received from other members */
@Property(description="discards warnings about promiscuous traffic")
protected boolean log_discard_msgs=true;
@Property(description="If true, trashes warnings about retransmission messages not found in the xmit_table (used for testing)")
protected boolean log_not_found_msgs=true;
@Property(description="Interval (in milliseconds) at which missing messages (from all retransmit buffers) " +
"are retransmitted")
protected long xmit_interval=1000;
@Property(description="Number of rows of the matrix in the retransmission table (only for experts)",writable=false)
protected int xmit_table_num_rows=50;
@Property(description="Number of elements of a row of the matrix in the retransmission table (only for experts). " +
"The capacity of the matrix is xmit_table_num_rows * xmit_table_msgs_per_row",writable=false)
protected int xmit_table_msgs_per_row=10000;
@Property(description="Resize factor of the matrix in the retransmission table (only for experts)",writable=false)
protected double xmit_table_resize_factor=1.2;
@Property(description="Number of milliseconds after which the matrix in the retransmission table " +
"is compacted (only for experts)",writable=false)
protected long xmit_table_max_compaction_time=10000;
@Property(description="Size of the queue to hold messages received after creating the channel, but before being " +
"connected (is_server=false). After becoming the server, the messages in the queue are fed into up() and the " +
"queue is cleared. The motivation is to avoid retransmissions (see https://issues.jboss.org/browse/JGRP-1509 " +
"for details). 0 disables the queue.")
protected int become_server_queue_size=50;
@Property(description="Time during which identical warnings about messages from a non member will be suppressed. " +
"0 disables this (every warning will be logged). Setting the log level to ERROR also disables this.")
protected long suppress_time_non_member_warnings=60000;
/* -------------------------------------------------- JMX ---------------------------------------------------------- */
@ManagedAttribute(description="Number of retransmit requests received")
protected final AtomicLong xmit_reqs_received=new AtomicLong(0);
@ManagedAttribute(description="Number of retransmit requests sent")
protected final AtomicLong xmit_reqs_sent=new AtomicLong(0);
@ManagedAttribute(description="Number of retransmit responses received")
protected final AtomicLong xmit_rsps_received=new AtomicLong(0);
@ManagedAttribute(description="Number of retransmit responses sent")
protected final AtomicLong xmit_rsps_sent=new AtomicLong(0);
@ManagedAttribute(description="Is the retransmit task running")
public boolean isXmitTaskRunning() {return xmit_task != null && !xmit_task.isDone();}
@ManagedAttribute(description="Number of messages sent")
protected int num_messages_sent=0;
@ManagedAttribute(description="Number of messages received")
protected int num_messages_received=0;
@ManagedAttribute(description="Number of messages from non-members")
public int getNonMemberMessages() {
return suppress_log_non_member != null? suppress_log_non_member.getCache().size() : 0;
}
@ManagedOperation(description="Clears the cache for messages from non-members")
public void clearNonMemberCache() {
if(suppress_log_non_member != null)
suppress_log_non_member.getCache().clear();
}
/* ------------------------------------------------- Fields ------------------------------------------------------------------------- */
protected volatile boolean is_server=false;
protected Address local_addr=null;
protected volatile List<Address> members=new ArrayList<Address>();
protected View view;
private final AtomicLong seqno=new AtomicLong(0); // current message sequence number (starts with 1)
/** Map to store sent and received messages (keyed by sender) */
protected final ConcurrentMap<Address,Table<Message>> xmit_table=Util.createConcurrentMap();
/** RetransmitTask running every xmit_interval ms */
protected Future<?> xmit_task;
/** Used by the retransmit task to keep the last retransmitted seqno per sender (https://issues.jboss.org/browse/JGRP-1539) */
protected final Map<Address,Long> xmit_task_map=new HashMap<Address,Long>();
protected volatile boolean leaving=false;
protected volatile boolean running=false;
protected TimeScheduler timer=null;
protected final Lock rebroadcast_lock=new ReentrantLock();
protected final Condition rebroadcast_done=rebroadcast_lock.newCondition();
// set during processing of a rebroadcast event
protected volatile boolean rebroadcasting=false;
protected final Lock rebroadcast_digest_lock=new ReentrantLock();
@GuardedBy("rebroadcast_digest_lock")
protected Digest rebroadcast_digest=null;
/** BoundedList<Digest>, keeps the last 10 stability messages */
protected final BoundedList<Digest> stability_msgs=new BoundedList<Digest>(10);
/** Keeps a bounded list of the last N digest sets */
protected final BoundedList<String> digest_history=new BoundedList<String>(10);
protected BoundedList<Message> become_server_queue;
/** Log to suppress identical warnings for messages from non-members */
protected SuppressLog<Address> suppress_log_non_member;
public long getXmitRequestsReceived() {return xmit_reqs_received.get();}
public long getXmitRequestsSent() {return xmit_reqs_sent.get();}
public long getXmitResponsesReceived() {return xmit_rsps_received.get();}
public long getXmitResponsesSent() {return xmit_rsps_sent.get();}
public boolean isUseMcastXmit() {return use_mcast_xmit;}
public boolean isXmitFromRandomMember() {return xmit_from_random_member;}
public boolean isDiscardDeliveredMsgs() {return discard_delivered_msgs;}
public boolean getLogDiscardMessages() {return log_discard_msgs;}
public void setUseMcastXmit(boolean use_mcast_xmit) {this.use_mcast_xmit=use_mcast_xmit;}
public void setUseMcastXmitReq(boolean flag) {this.use_mcast_xmit_req=flag;}
public void setLogDiscardMessages(boolean flag) {log_discard_msgs=flag;}
public void setXmitFromRandomMember(boolean xmit_from_random_member) {
this.xmit_from_random_member=xmit_from_random_member;
}
public void setDiscardDeliveredMsgs(boolean discard_delivered_msgs) {
this.discard_delivered_msgs=discard_delivered_msgs;
}
@ManagedAttribute(description="Actual size of the become_server_queue")
public int getBecomeServerQueueSizeActual() {
return become_server_queue != null? become_server_queue.size() : -1;
}
/** Returns the receive window for sender; only used for testing. Do not use ! */
public Table<Message> getWindow(Address sender) {
return xmit_table.get(sender);
}
/** Only used for unit tests, don't use ! */
public void setTimer(TimeScheduler timer) {this.timer=timer;}
@ManagedAttribute(description="Total number of undelivered messages in all retransmit buffers")
public int getXmitTableUndeliveredMsgs() {
int num=0;
for(Table<Message> buf: xmit_table.values())
num+=buf.size();
return num;
}
@ManagedAttribute(description="Total number of missing (= not received) messages in all retransmit buffers")
public int getXmitTableMissingMessages() {
int num=0;
for(Table<Message> buf: xmit_table.values())
num+=buf.getNumMissing();
return num;
}
@ManagedAttribute(description="Capacity of the retransmit buffer. Computed as xmit_table_num_rows * xmit_table_msgs_per_row")
public long getXmitTableCapacity() {
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
return table != null? table.capacity() : 0;
}
@ManagedAttribute(description="Prints the number of rows currently allocated in the matrix. This value will not " +
"be lower than xmit_table_now_rows")
public int getXmitTableNumCurrentRows() {
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
return table != null? table.getNumRows() : 0;
}
@ManagedAttribute(description="Returns the number of bytes of all messages in all retransmit buffers. " +
"To compute the size, Message.getLength() is used")
public long getSizeOfAllMessages() {
long retval=0;
for(Table<Message> buf: xmit_table.values())
retval+=sizeOfAllMessages(buf,false);
return retval;
}
@ManagedAttribute(description="Returns the number of bytes of all messages in all retransmit buffers. " +
"To compute the size, Message.size() is used")
public long getSizeOfAllMessagesInclHeaders() {
long retval=0;
for(Table<Message> buf: xmit_table.values())
retval+=sizeOfAllMessages(buf, true);
return retval;
}
@ManagedAttribute(description="Number of retransmit table compactions")
public int getXmitTableNumCompactions() {
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
return table != null? table.getNumCompactions() : 0;
}
@ManagedAttribute(description="Number of retransmit table moves")
public int getXmitTableNumMoves() {
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
return table != null? table.getNumMoves() : 0;
}
@ManagedAttribute(description="Number of retransmit table resizes")
public int getXmitTableNumResizes() {
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
return table != null? table.getNumResizes(): 0;
}
@ManagedAttribute(description="Number of retransmit table purges")
public int getXmitTableNumPurges() {
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
return table != null? table.getNumPurges(): 0;
}
@ManagedOperation(description="Prints the contents of the receiver windows for all members")
public String printMessages() {
StringBuilder ret=new StringBuilder(local_addr + ":\n");
for(Map.Entry<Address,Table<Message>> entry: xmit_table.entrySet()) {
Address addr=entry.getKey();
Table<Message> buf=entry.getValue();
ret.append(addr).append(": ").append(buf.toString()).append('\n');
}
return ret.toString();
}
@ManagedAttribute public long getCurrentSeqno() {return seqno.get();}
@ManagedOperation(description="Prints the stability messages received")
public String printStabilityMessages() {
StringBuilder sb=new StringBuilder();
sb.append(Util.printListWithDelimiter(stability_msgs, "\n"));
return sb.toString();
}
@ManagedOperation(description="Keeps information about the last N times a digest was set or merged")
public String printDigestHistory() {
StringBuilder sb=new StringBuilder(local_addr + ":\n");
for(String tmp: digest_history)
sb.append(tmp).append("\n");
return sb.toString();
}
@ManagedOperation(description="Compacts the retransmit buffer")
public void compact() {
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
if(table != null)
table.compact();
}
@ManagedOperation(description="Prints the number of rows currently allocated in the matrix for all members. " +
"This value will not be lower than xmit_table_now_rows")
public String dumpXmitTablesNumCurrentRows() {
StringBuilder sb=new StringBuilder();
for(Map.Entry<Address,Table<Message>> entry: xmit_table.entrySet()) {
sb.append(entry.getKey()).append(": ").append(entry.getValue().getNumRows()).append("\n");
}
return sb.toString();
}
@ManagedOperation(description="Resets all statistics")
public void resetStats() {
num_messages_sent=num_messages_received=0;
xmit_reqs_received.set(0);
xmit_reqs_sent.set(0);
xmit_rsps_received.set(0);
xmit_rsps_sent.set(0);
stability_msgs.clear();
digest_history.clear();
Table<Message> table=local_addr != null? xmit_table.get(local_addr) : null;
if(table != null)
table.resetStats();
}
public void init() throws Exception {
if(xmit_from_random_member) {
if(discard_delivered_msgs) {
discard_delivered_msgs=false;
log.debug("xmit_from_random_member set to true: changed discard_delivered_msgs to false");
}
}
TP transport=getTransport();
if(transport != null) {
transport.registerProbeHandler(this);
if(!transport.supportsMulticasting()) {
if(use_mcast_xmit) {
log.warn("use_mcast_xmit should not be used because the transport (" + transport.getName() +
") does not support IP multicasting; setting use_mcast_xmit to false");
use_mcast_xmit=false;
}
if(use_mcast_xmit_req) {
log.warn("use_mcast_xmit_req should not be used because the transport (" + transport.getName() +
") does not support IP multicasting; setting use_mcast_xmit_req to false");
use_mcast_xmit_req=false;
}
}
}
if(become_server_queue_size > 0)
become_server_queue=new BoundedList<Message>(become_server_queue_size);
if(suppress_time_non_member_warnings > 0)
suppress_log_non_member=new SuppressLog<Address>(log, "MsgDroppedNak", "SuppressMsg");
}
public Map<String,Object> dumpStats() {
Map<String,Object> retval=super.dumpStats();
retval.put("msgs", printMessages());
return retval;
}
public String printStats() {
StringBuilder sb=new StringBuilder();
sb.append("\nStability messages received\n");
sb.append(printStabilityMessages()).append("\n");
return sb.toString();
}
public String printStabilityHistory() {
StringBuilder sb=new StringBuilder();
int i=1;
for(Digest digest: stability_msgs) {
sb.append(i++).append(": ").append(digest).append("\n");
}
return sb.toString();
}
public List<Integer> providedUpServices() {
return Arrays.asList(Event.GET_DIGEST,Event.SET_DIGEST,Event.OVERWRITE_DIGEST,Event.MERGE_DIGEST);
}
public void start() throws Exception {
timer=getTransport().getTimer();
if(timer == null)
throw new Exception("timer is null");
running=true;
leaving=false;
startRetransmitTask();
}
public void stop() {
running=false;
is_server=false;
if(become_server_queue != null)
become_server_queue.clear();
stopRetransmitTask();
xmit_task_map.clear();
reset();
}
/**
* <b>Callback</b>. Called by superclass when event may be handled.<p> <b>Do not use <code>down_prot.down()</code> in this
* method as the event is passed down by default by the superclass after this method returns !</b>
*/
public Object down(Event evt) {
switch(evt.getType()) {
case Event.MSG:
Message msg=(Message)evt.getArg();
Address dest=msg.getDest();
if(dest != null || msg.isFlagSet(Message.NO_RELIABILITY))
break; // unicast address: not null and not mcast, pass down unchanged
send(evt, msg);
return null; // don't pass down the stack
case Event.STABLE: // generated by STABLE layer. Delete stable messages passed in arg
stable((Digest)evt.getArg());
return null; // do not pass down further (Bela Aug 7 2001)
case Event.GET_DIGEST:
return getDigest((Address)evt.getArg());
case Event.SET_DIGEST:
setDigest((Digest)evt.getArg());
return null;
case Event.OVERWRITE_DIGEST:
overwriteDigest((Digest)evt.getArg());
return null;
case Event.MERGE_DIGEST:
mergeDigest((Digest)evt.getArg());
return null;
case Event.TMP_VIEW:
View tmp_view=(View)evt.getArg();
List<Address> mbrs=tmp_view.getMembers();
members=new ArrayList<Address>(mbrs);
break;
case Event.VIEW_CHANGE:
tmp_view=(View)evt.getArg();
mbrs=tmp_view.getMembers();
members=new ArrayList<Address>(mbrs);
view=tmp_view;
adjustReceivers(mbrs);
is_server=true; // check vids from now on
if(suppress_log_non_member != null)
suppress_log_non_member.removeExpired(suppress_time_non_member_warnings);
xmit_task_map.keySet().retainAll(tmp_view.getMembers());
break;
case Event.BECOME_SERVER:
is_server=true;
flushBecomeServerQueue();
break;
case Event.SET_LOCAL_ADDRESS:
local_addr=(Address)evt.getArg();
break;
case Event.DISCONNECT:
leaving=true;
reset();
break;
case Event.REBROADCAST:
rebroadcasting=true;
rebroadcast_digest=(Digest)evt.getArg();
try {
rebroadcastMessages();
}
finally {
rebroadcasting=false;
rebroadcast_digest_lock.lock();
try {
rebroadcast_digest=null;
}
finally {
rebroadcast_digest_lock.unlock();
}
}
return null;
}
return down_prot.down(evt);
}
/**
* <b>Callback</b>. Called by superclass when event may be handled.<p> <b>Do not use <code>PassUp</code> in this
* method as the event is passed up by default by the superclass after this method returns !</b>
*/
public Object up(Event evt) {
switch(evt.getType()) {
case Event.MSG:
Message msg=(Message)evt.getArg();
if(msg.isFlagSet(Message.NO_RELIABILITY))
break;
NakAckHeader2 hdr=(NakAckHeader2)msg.getHeader(this.id);
if(hdr == null)
break; // pass up (e.g. unicast msg)
if(!is_server) { // discard messages while not yet server (i.e., until JOIN has returned)
if(become_server_queue != null) {
become_server_queue.add(msg);
if(log.isTraceEnabled())
log.trace(local_addr + ": message " + msg.getSrc() + "::" + hdr.seqno + " was added to queue (not yet server)");
}
else {
if(log.isTraceEnabled())
log.trace(local_addr + ": message " + msg.getSrc() + "::" + hdr.seqno + " was discarded (not yet server)");
}
return null;
}
// Changed by bela Jan 29 2003: we must not remove the header, otherwise further xmit requests will fail !
//hdr=(NakAckHeader2)msg.removeHeader(getName());
switch(hdr.type) {
case NakAckHeader2.MSG:
handleMessage(msg, hdr);
return null; // transmitter passes message up for us !
case NakAckHeader2.XMIT_REQ:
SeqnoList missing=(SeqnoList)msg.getObject();
if(missing == null)
return null;
handleXmitReq(msg.getSrc(), missing, hdr.sender);
return null;
case NakAckHeader2.XMIT_RSP:
handleXmitRsp(msg, hdr);
return null;
default:
if(log.isErrorEnabled())
log.error("NakAck header type " + hdr.type + " not known !");
return null;
}
case Event.STABLE: // generated by STABLE layer. Delete stable messages passed in arg
stable((Digest)evt.getArg());
return null; // do not pass up further (Bela Aug 7 2001)
case Event.SUSPECT:
// release the promise if rebroadcasting is in progress... otherwise we wait forever. there will be a new
// flush round anyway
if(rebroadcasting) {
cancelRebroadcasting();
}
break;
}
return up_prot.up(evt);
}
// ProbeHandler interface
public Map<String, String> handleProbe(String... keys) {
Map<String,String> retval=new HashMap<String,String>();
for(String key: keys) {
if(key.equals("digest-history"))
retval.put(key, printDigestHistory());
if(key.equals("dump-digest"))
retval.put(key, "\n" + printMessages());
}
return retval;
}
// ProbeHandler interface
public String[] supportedKeys() {
return new String[]{"digest-history", "dump-digest"};
}
/* --------------------------------- Private Methods --------------------------------------- */
/**
* Adds the message to the sent_msgs table and then passes it down the stack. Change Bela Ban May 26 2002: we don't
* store a copy of the message, but a reference ! This saves us a lot of memory. However, this also means that a
* message should not be changed after storing it in the sent-table ! See protocols/DESIGN for details.
* Made seqno increment and adding to sent_msgs atomic, e.g. seqno won't get incremented if adding to
* sent_msgs fails e.g. due to an OOM (see http://jira.jboss.com/jira/browse/JGRP-179). bela Jan 13 2006
*/
protected void send(Event evt, Message msg) {
if(msg == null)
throw new NullPointerException("msg is null; event is " + evt);
if(!running) {
if(log.isTraceEnabled())
log.trace(local_addr + ": discarded message as we're not in the 'running' state, message: " + msg);
return;
}
long msg_id;
Table<Message> buf=xmit_table.get(local_addr);
if(buf == null) // discard message if there is no entry for local_addr
return;
if(msg.getSrc() == null)
msg.setSrc(local_addr); // this needs to be done so we can check whether the message sender is the local_addr
msg_id=seqno.incrementAndGet();
long sleep=10;
while(running) {
try {
msg.putHeader(this.id, NakAckHeader2.createMessageHeader(msg_id));
buf.add(msg_id, msg);
break;
}
catch(Throwable t) {
if(!running)
break;
Util.sleep(sleep);
sleep=Math.min(5000, sleep*2);
}
}
// moved down_prot.down() out of synchronized clause (bela Sept 7 2006) http://jira.jboss.com/jira/browse/JGRP-300
if(log.isTraceEnabled())
log.trace(local_addr + " sending " + local_addr + "#" + msg_id);
down_prot.down(evt); // if this fails, since msg is in sent_msgs, it can be retransmitted
num_messages_sent++;
}
/**
* Finds the corresponding retransmit buffer and adds the message to it (according to seqno). Then removes as many
* messages as possible and passes them up the stack. Discards messages from non-members.
*/
protected void handleMessage(Message msg, NakAckHeader2 hdr) {
Address sender=msg.getSrc();
Table<Message> buf=xmit_table.get(sender);
if(buf == null) { // discard message if there is no entry for sender
if(leaving)
return;
if(log.isWarnEnabled() && log_discard_msgs) {
if(suppress_log_non_member != null) {
suppress_log_non_member.log(SuppressLog.Level.warn, sender, suppress_time_non_member_warnings,
local_addr, hdr.seqno, sender, view);
}
else
log.warn(Util.getMessage("MsgDroppedNak", local_addr, hdr.seqno, sender, view));
}
return;
}
num_messages_received++;
boolean loopback=local_addr.equals(sender);
boolean added=loopback || buf.add(hdr.seqno, msg);
if(added && log.isTraceEnabled())
log.trace(new StringBuilder().append(local_addr).append(": received ").append(sender).append('#').append(hdr.seqno));
// OOB msg is passed up. When removed, we discard it. Affects ordering: http://jira.jboss.com/jira/browse/JGRP-379
if(added && msg.isFlagSet(Message.OOB)) {
if(loopback)
msg=buf.get(hdr.seqno); // we *have* to get a message, because loopback means we didn't add it to win !
if(msg != null && msg.isFlagSet(Message.OOB)) {
if(msg.setTransientFlagIfAbsent(Message.OOB_DELIVERED)) {
if(log.isTraceEnabled())
log.trace(new StringBuilder().append(local_addr).append(": delivering ").append(sender).append('#').append(hdr.seqno));
try {
up_prot.up(new Event(Event.MSG, msg));
}
catch(Throwable t) {
log.error("failed to deliver OOB message " + msg, t);
}
}
}
}
// Efficient way of checking whether another thread is already processing messages from 'sender'.
// If that's the case, we return immediately and let the existing thread process our message
// (https://jira.jboss.org/jira/browse/JGRP-829). Benefit: fewer threads blocked on the same lock, these threads
// can be returned to the thread pool
final AtomicBoolean processing=buf.getProcessing();
if(!processing.compareAndSet(false, true)) {
return;
}
boolean remove_msgs=discard_delivered_msgs && !loopback;
boolean released_processing=false;
try {
while(true) {
// we're removing a msg and set processing to false (if null) *atomically* (wrt to add())
List<Message> msgs=buf.removeMany(processing, remove_msgs, max_msg_batch_size);
if(msgs == null || msgs.isEmpty()) {
released_processing=true;
if(rebroadcasting)
checkForRebroadcasts();
return;
}
for(final Message msg_to_deliver: msgs) {
// discard OOB msg if it has already been delivered (http://jira.jboss.com/jira/browse/JGRP-379)
if(msg_to_deliver.isFlagSet(Message.OOB) && !msg_to_deliver.setTransientFlagIfAbsent(Message.OOB_DELIVERED))
continue;
//msg_to_deliver.removeHeader(getName()); // Changed by bela Jan 29 2003: not needed (see above)
try {
if(log.isTraceEnabled()) {
NakAckHeader2 header=(NakAckHeader2)msg_to_deliver.getHeader(this.id);
log.trace(new StringBuilder().append(local_addr).append(": delivering ").append(sender).append('#').append(header.seqno));
}
up_prot.up(new Event(Event.MSG, msg_to_deliver));
}
catch(Throwable t) {
log.error("failed to deliver message " + msg_to_deliver, t);
}
}
}
}
finally {
// processing is always set in win.remove(processing) above and never here ! This code is just a
// 2nd line of defense should there be an exception before win.remove(processing) sets processing
if(!released_processing)
processing.set(false);
}
}
/**
* Retransmits messsages first_seqno to last_seqno from original_sender from xmit_table to xmit_requester,
* called when XMIT_REQ is received.
* @param xmit_requester The sender of the XMIT_REQ, we have to send the requested copy of the message to this address
* @param missing_msgs A list of seqnos that have to be retransmitted
* @param original_sender The member who originally sent the messsage. Guaranteed to be non-null
*/
protected void handleXmitReq(Address xmit_requester, SeqnoList missing_msgs, Address original_sender) {
if(log.isTraceEnabled()) {
StringBuilder sb=new StringBuilder();
sb.append(local_addr).append(": received xmit request from ").append(xmit_requester).append(" for ");
sb.append(original_sender).append(missing_msgs);
log.trace(sb);
}
if(stats)
xmit_reqs_received.addAndGet(missing_msgs.size());
Table<Message> buf=xmit_table.get(original_sender);
if(buf == null) {
if(log.isErrorEnabled()) {
StringBuilder sb=new StringBuilder();
sb.append("(requester=").append(xmit_requester).append(", local_addr=").append(this.local_addr);
sb.append(") ").append(original_sender).append(" not found in retransmission table");
// don't print the table unless we are in trace mode because it can be LARGE
if(log.isTraceEnabled()) {
sb.append(":\n").append(printMessages());
}
if(print_stability_history_on_failed_xmit) {
sb.append(" (stability history:\n").append(printStabilityHistory());
}
log.error(sb.toString());
}
return;
}
for(long i: missing_msgs) {
Message msg=buf.get(i);
if(msg == null) {
if(log.isWarnEnabled() && log_not_found_msgs && !local_addr.equals(xmit_requester) && i > buf.getLow()) {
StringBuilder sb=new StringBuilder();
sb.append("(requester=").append(xmit_requester).append(", local_addr=").append(this.local_addr);
sb.append(") message ").append(original_sender).append("::").append(i);
sb.append(" not found in retransmission table of ").append(original_sender).append(":\n").append(buf);
if(print_stability_history_on_failed_xmit) {
sb.append(" (stability history:\n").append(printStabilityHistory());
}
log.warn(sb.toString());
}
continue;
}
if(log.isTraceEnabled())
log.trace(local_addr + ": resending " + original_sender + "::" + i);
sendXmitRsp(xmit_requester, msg);
}
}
/**
* Flushes the queue. Done in a separate thread as we don't want to block the
* {@link ClientGmsImpl#installView(org.jgroups.View,org.jgroups.util.Digest)} method (called when a view is installed).
*/
protected void flushBecomeServerQueue() {
if(become_server_queue != null && !become_server_queue.isEmpty()) {
if(log.isTraceEnabled())
log.trace(local_addr + ": flushing become_server_queue (" + become_server_queue.size() + " elements)");
TP transport=getTransport();
Executor thread_pool=transport.getDefaultThreadPool(), oob_thread_pool=transport.getOOBThreadPool();
for(final Message msg: become_server_queue) {
Executor pool=msg.isFlagSet(Message.Flag.OOB)? oob_thread_pool : thread_pool;
pool.execute(new Runnable() {
public void run() {
try {
up(new Event(Event.MSG, msg));
}
finally {
become_server_queue.remove(msg);
}
}
});
}
}
}
protected void cancelRebroadcasting() {
rebroadcast_lock.lock();
try {
rebroadcasting=false;
rebroadcast_done.signalAll();
}
finally {
rebroadcast_lock.unlock();
}
}
/**
* Sends a message msg to the requester. We have to wrap the original message into a retransmit message, as we need
* to preserve the original message's properties, such as src, headers etc.
* @param dest
* @param msg
*/
protected void sendXmitRsp(Address dest, Message msg) {
if(msg == null)
return;
if(stats)
xmit_rsps_sent.incrementAndGet();
if(msg.getSrc() == null)
msg.setSrc(local_addr);
if(use_mcast_xmit) { // we simply send the original multicast message
down_prot.down(new Event(Event.MSG, msg));
return;
}
Message xmit_msg=msg.copy(true, true); // copy payload and headers
xmit_msg.setDest(dest);
NakAckHeader2 hdr=(NakAckHeader2)xmit_msg.getHeader(id);
NakAckHeader2 newhdr=hdr.copy();
newhdr.type=NakAckHeader2.XMIT_RSP; // change the type in the copy from MSG --> XMIT_RSP
xmit_msg.putHeader(id, newhdr);
down_prot.down(new Event(Event.MSG,xmit_msg));
}
protected void handleXmitRsp(Message msg, NakAckHeader2 hdr) {
if(msg == null)
return;
try {
if(stats)
xmit_rsps_received.incrementAndGet();
msg.setDest(null);
NakAckHeader2 newhdr=hdr.copy();
newhdr.type=NakAckHeader2.MSG; // change the type back from XMIT_RSP --> MSG
msg.putHeader(id, newhdr);
up(new Event(Event.MSG, msg));
if(rebroadcasting)
checkForRebroadcasts();
}
catch(Exception ex) {
if(log.isErrorEnabled()) {
log.error("failed handling retransmitted message",ex);
}
}
}
/**
* Takes the argument highest_seqnos and compares it to the current digest. If the current digest has fewer messages,
* then send retransmit messages for the missing messages. Return when all missing messages have been received. If
* we're waiting for a missing message from P, and P crashes while waiting, we need to exclude P from the wait set.
*/
protected void rebroadcastMessages() {
Digest their_digest;
long sleep=max_rebroadcast_timeout / NUM_REBROADCAST_MSGS;
long wait_time=max_rebroadcast_timeout, start=System.currentTimeMillis();
while(wait_time > 0) {
rebroadcast_digest_lock.lock();
try {
if(rebroadcast_digest == null)
break;
their_digest=rebroadcast_digest.copy();
}
finally {
rebroadcast_digest_lock.unlock();
}
Digest my_digest=getDigest();
boolean xmitted=false;
for(Digest.DigestEntry entry: their_digest) {
Address member=entry.getMember();
long[] my_entry=my_digest.get(member);
if(my_entry == null)
continue;
long their_high=entry.getHighest();
// Cannot ask for 0 to be retransmitted because the first seqno in NAKACK2 and UNICAST(2) is always 1 !
// Also, we need to ask for retransmission of my_high+1, because we already *have* my_high, and don't
// need it, so the retransmission range is [my_high+1 .. their_high]: *exclude* my_high, but *include*
// their_high
long my_high=Math.max(my_entry[0], my_entry[1]);
if(their_high > my_high) {
if(log.isTraceEnabled())
log.trace(local_addr + " fetching " + my_high + "-" + their_high + " from " + member);
retransmit(my_high+1, their_high, member, true); // use multicast to send retransmit request
xmitted=true;
}
}
if(!xmitted)
return; // we're done; no retransmissions are needed anymore. our digest is >= rebroadcast_digest
rebroadcast_lock.lock();
try {
try {
my_digest=getDigest();
rebroadcast_digest_lock.lock();
try {
if(!rebroadcasting || my_digest.isGreaterThanOrEqual(rebroadcast_digest))
return;
}
finally {
rebroadcast_digest_lock.unlock();
}
rebroadcast_done.await(sleep, TimeUnit.MILLISECONDS);
wait_time-=(System.currentTimeMillis() - start);
}
catch(InterruptedException e) {
}
}
finally {
rebroadcast_lock.unlock();
}
}
}
protected void checkForRebroadcasts() {
Digest tmp=getDigest();
boolean cancel_rebroadcasting;
rebroadcast_digest_lock.lock();
try {
cancel_rebroadcasting=tmp.isGreaterThanOrEqual(rebroadcast_digest);
}
finally {
rebroadcast_digest_lock.unlock();
}
if(cancel_rebroadcasting) {
cancelRebroadcasting();
}
}
/**
* Remove old members from the retransmission buffers. Essentially removes all members that are not
* in <code>members</code>. This method is not called concurrently multiple times
*/
protected void adjustReceivers(List<Address> new_members) {
for(Address member: xmit_table.keySet()) {
if(!new_members.contains(member)) {
if(local_addr != null && local_addr.equals(member))
continue;
Table<Message> buf=xmit_table.remove(member);
if(buf != null) {
if(log.isDebugEnabled())
log.debug("removed " + member + " from xmit_table (not member anymore)");
}
}
}
}
/**
* Returns a message digest: for each member P the highest delivered and received seqno is added
*/
public Digest getDigest() {
final Map<Address,long[]> map=new HashMap<Address,long[]>();
for(Map.Entry<Address,Table<Message>> entry: xmit_table.entrySet()) {
Address sender=entry.getKey(); // guaranteed to be non-null (CCHM)
Table<Message> buf=entry.getValue(); // guaranteed to be non-null (CCHM)
long[] seqnos=buf.getDigest();
map.put(sender, seqnos);
}
return new Digest(map);
}
public Digest getDigest(Address mbr) {
if(mbr == null)
return getDigest();
Table<Message> buf=xmit_table.get(mbr);
if(buf == null)
return null;
long[] seqnos=buf.getDigest();
return new Digest(mbr, seqnos[0], seqnos[1]);
}
/**
* Creates a retransmit buffer for each sender in the digest according to the sender's seqno.
* If a buffer already exists, it resets it.
*/
protected void setDigest(Digest digest) {
setDigest(digest,false);
}
/**
* For all members of the digest, adjust the retransmit buffers in xmit_table. If no entry
* exists, create one with the initial seqno set to the seqno of the member in the digest. If the member already
* exists, and is not the local address, replace it with the new entry (http://jira.jboss.com/jira/browse/JGRP-699)
* if the digest's seqno is greater than the seqno in the window.
*/
protected void mergeDigest(Digest digest) {
setDigest(digest,true);
}
/**
* Overwrites existing entries, but does NOT remove entries not found in the digest
* @param digest
*/
protected void overwriteDigest(Digest digest) {
if(digest == null)
return;
StringBuilder sb=new StringBuilder("\n[overwriteDigest()]\n");
sb.append("existing digest: " + getDigest()).append("\nnew digest: " + digest);
for(Digest.DigestEntry entry: digest) {
Address member=entry.getMember();
if(member == null)
continue;
long highest_delivered_seqno=entry.getHighestDeliveredSeqno();
Table<Message> buf=xmit_table.get(member);
if(buf != null) {
if(local_addr.equals(member)) {
// Adjust the highest_delivered seqno (to send msgs again): https://jira.jboss.org/browse/JGRP-1251
buf.setHighestDelivered(highest_delivered_seqno);
continue; // don't destroy my own window
}
xmit_table.remove(member);
}
buf=createTable(highest_delivered_seqno);
xmit_table.put(member, buf);
}
sb.append("\n").append("resulting digest: " + getDigest());
digest_history.add(sb.toString());
if(log.isDebugEnabled())
log.debug(sb.toString());
}
/**
* Sets or merges the digest. If there is no entry for a given member in xmit_table, create a new buffer.
* Else skip the existing entry, unless it is a merge. In this case, skip the existing entry if its seqno is
* greater than or equal to the one in the digest, or reset the window and create a new one if not.
* @param digest The digest
* @param merge Whether to merge the new digest with our own, or not
*/
protected void setDigest(Digest digest, boolean merge) {
if(digest == null)
return;
StringBuilder sb=new StringBuilder(merge? "\n[" + local_addr + " mergeDigest()]\n" : "\n["+local_addr + " setDigest()]\n");
sb.append("existing digest: " + getDigest()).append("\nnew digest: " + digest);
boolean set_own_seqno=false;
for(Digest.DigestEntry entry: digest) {
Address member=entry.getMember();
if(member == null)
continue;
long highest_delivered_seqno=entry.getHighestDeliveredSeqno();
Table<Message> buf=xmit_table.get(member);
if(buf != null) {
// We only reset the window if its seqno is lower than the seqno shipped with the digest. Also, we
// don't reset our own window (https://jira.jboss.org/jira/browse/JGRP-948, comment 20/Apr/09 03:39 AM)
if(!merge
|| (local_addr != null && local_addr.equals(member)) // never overwrite our own entry
|| buf.getHighestDelivered() >= highest_delivered_seqno) // my seqno is >= digest's seqno for sender
continue;
xmit_table.remove(member);
// to get here, merge must be false !
if(member.equals(local_addr)) { // Adjust the seqno: https://jira.jboss.org/browse/JGRP-1251
seqno.set(highest_delivered_seqno);
set_own_seqno=true;
}
}
buf=createTable(highest_delivered_seqno);
xmit_table.put(member, buf);
}
sb.append("\n").append("resulting digest: " + getDigest());
if(set_own_seqno)
sb.append("\nnew seqno for " + local_addr + ": " + seqno);
digest_history.add(sb.toString());
if(log.isDebugEnabled())
log.debug(sb.toString());
}
protected Table<Message> createTable(long initial_seqno) {
return new Table<Message>(xmit_table_num_rows, xmit_table_msgs_per_row,
initial_seqno, xmit_table_resize_factor, xmit_table_max_compaction_time);
}
/**
* Garbage collect messages that have been seen by all members. Update sent_msgs: for the sender P in the digest
* which is equal to the local address, garbage collect all messages <= seqno at digest[P]. Update xmit_table:
* for each sender P in the digest and its highest seqno seen SEQ, garbage collect all delivered_msgs in the
* retransmit buffer corresponding to P which are <= seqno at digest[P].
*/
protected void stable(Digest digest) {
long my_highest_rcvd; // highest seqno received in my digest for a sender P
long stability_highest_rcvd; // highest seqno received in the stability vector for a sender P
if(members == null || local_addr == null || digest == null)
return;
if(log.isTraceEnabled())
log.trace(local_addr + ": received stable digest " + digest);
stability_msgs.add(digest);
for(Digest.DigestEntry entry: digest) {
Address member=entry.getMember();
if(member == null)
continue;
long hd=entry.getHighestDeliveredSeqno();
long hr=entry.getHighestReceivedSeqno();
// check whether the last seqno received for a sender P in the stability vector is > last seqno
// received for P in my digest. if yes, request retransmission (see "Last Message Dropped" topic
// in DESIGN)
Table<Message> buf=xmit_table.get(member);
if(buf != null) {
my_highest_rcvd=buf.getHighestReceived();
stability_highest_rcvd=hr;
if(stability_highest_rcvd >= 0 && stability_highest_rcvd > my_highest_rcvd) {
if(log.isTraceEnabled()) {
log.trace(local_addr + ": my_highest_rcvd (" + my_highest_rcvd + ") < stability_highest_rcvd (" +
stability_highest_rcvd + "): requesting retransmission of " +
member + '#' + stability_highest_rcvd);
}
retransmit(stability_highest_rcvd, stability_highest_rcvd, member);
}
}
if(hd < 0)
continue;
if(log.isTraceEnabled())
log.trace(local_addr + ": deleting msgs <= " + hd + " from " + member);
// delete *delivered* msgs that are stable
if(buf != null)
buf.purge(hd); // delete all messages with seqnos <= seqno
}
}
protected void retransmit(long first_seqno, long last_seqno, Address sender) {
if(first_seqno <= last_seqno)
retransmit(first_seqno,last_seqno,sender,false);
}
protected void retransmit(long first_seqno, long last_seqno, final Address sender, boolean multicast_xmit_request) {
retransmit(new SeqnoList(first_seqno,last_seqno),sender,multicast_xmit_request);
}
protected void retransmit(SeqnoList missing_msgs, final Address sender, boolean multicast_xmit_request) {
Address dest=sender; // to whom do we send the XMIT request ?
if(multicast_xmit_request || this.use_mcast_xmit_req) {
dest=null;
}
else {
if(xmit_from_random_member && !local_addr.equals(sender)) {
Address random_member=(Address)Util.pickRandomElement(members);
if(random_member != null && !local_addr.equals(random_member)) {
dest=random_member;
if(log.isTraceEnabled())
log.trace(local_addr + ": picked random member " + dest + " to send XMIT request to");
}
}
}
NakAckHeader2 hdr=NakAckHeader2.createXmitRequestHeader(sender);
Message retransmit_msg=new Message(dest, null, missing_msgs);
retransmit_msg.setFlag(Message.OOB);
if(log.isTraceEnabled())
log.trace(local_addr + ": sending XMIT_REQ (" + missing_msgs + ") to " + dest);
retransmit_msg.putHeader(this.id, hdr);
down_prot.down(new Event(Event.MSG,retransmit_msg));
if(stats)
xmit_reqs_sent.addAndGet(missing_msgs.size());
}
protected void reset() {
seqno.set(0);
xmit_table.clear();
}
protected static long sizeOfAllMessages(Table<Message> buf, boolean include_headers) {
Counter counter=new Counter(include_headers);
buf.forEach(buf.getHighestDelivered() + 1, buf.getHighestReceived(), counter);
return counter.getResult();
}
protected void startRetransmitTask() {
if(xmit_task == null || xmit_task.isDone())
xmit_task=timer.scheduleWithFixedDelay(new RetransmitTask(), 0, xmit_interval, TimeUnit.MILLISECONDS);
}
protected void stopRetransmitTask() {
if(xmit_task != null) {
xmit_task.cancel(true);
xmit_task=null;
}
}
/**
* Retransmitter task which periodically (every xmit_interval ms) looks at all the retransmit tables and
* sends retransmit request to all members from which we have missing messages
*/
protected class RetransmitTask implements Runnable {
public void run() {
triggerXmit();
}
}
@ManagedOperation(description="Triggers the retransmission task, asking all senders for missing messages")
public void triggerXmit() {
SeqnoList missing;
for(Map.Entry<Address,Table<Message>> entry: xmit_table.entrySet()) {
Address target=entry.getKey(); // target to send retransmit requests to
Table<Message> buf=entry.getValue();
if(buf != null && buf.getNumMissing() > 0 && (missing=buf.getMissing()) != null) { // getNumMissing() is fast
long highest=missing.getLast();
Long prev_seqno=xmit_task_map.get(target);
if(prev_seqno == null) {
xmit_task_map.put(target, highest); // no retransmission
}
else {
missing.removeHigherThan(prev_seqno); // we only retransmit the 'previous batch'
if(highest > prev_seqno)
xmit_task_map.put(target, highest);
if(missing.size() > 0)
retransmit(missing, target, false);
}
}
else if(!xmit_task_map.isEmpty())
xmit_task_map.remove(target); // no current gaps for target
}
}
protected static class Counter implements Table.Visitor<Message> {
protected final boolean count_size; // use size() or length()
protected long result=0;
public Counter(boolean count_size) {
this.count_size=count_size;
}
public long getResult() {return result;}
public boolean visit(long seqno, Message element, int row, int column) {
if(element != null)
result+=count_size? element.size() : element.getLength();
return true;
}
}
}