/**
* @param views Guaranteed to be non-null and to have >= 2 members, or else this thread would
* not be started
*/
public synchronized void start(Map<Address, View> views) {
if (thread == null || thread.isAlive()) {
this.coords.clear();
// now remove all members which don't have us in their view, so RPCs won't block (e.g.
// FLUSH)
// https://jira.jboss.org/browse/JGRP-1061
sanitizeViews(views);
// Add all different coordinators of the views into the hashmap and sets their members:
Collection<Address> coordinators = Util.determineMergeCoords(views);
for (Address coord : coordinators) {
View view = views.get(coord);
if (view != null) this.coords.put(coord, new ArrayList<Address>(view.getMembers()));
}
// For the merge participants which are not coordinator, we simply add them, and the
// associated
// membership list consists only of themselves
Collection<Address> merge_participants = Util.determineMergeParticipants(views);
merge_participants.removeAll(coordinators);
for (Address merge_participant : merge_participants) {
Collection<Address> tmp = new ArrayList<Address>();
tmp.add(merge_participant);
coords.putIfAbsent(merge_participant, tmp);
}
thread = gms.getThreadFactory().newThread(this, "MergeTask");
thread.setDaemon(true);
thread.start();
}
}
/**
* An event is to be sent down the stack. The layer may want to examine its type and perform some
* action on it, depending on the event's type. If the event is a message MSG, then the layer may
* need to add a header to it (or do nothing at all) before sending it down the stack using <code>
* PassDown</code>. In case of a GET_ADDRESS event (which tries to retrieve the stack's address
* from one of the bottom layers), the layer may need to send a new response event back up the
* stack using <code>up_prot.up()</code>. The PING protocol is interested in several different
* down events, Event.FIND_INITIAL_MBRS - sent by the GMS layer and expecting a GET_MBRS_OK
* Event.TMP_VIEW and Event.VIEW_CHANGE - a view change event Event.BECOME_SERVER - called after
* client has joined and is fully working group member Event.CONNECT, Event.DISCONNECT.
*/
@SuppressWarnings("unchecked")
public Object down(Event evt) {
switch (evt.getType()) {
case Event.FIND_INITIAL_MBRS: // sent by GMS layer
case Event.FIND_ALL_VIEWS:
// sends the GET_MBRS_REQ to all members, waits 'timeout' ms or until 'num_initial_members'
// have been retrieved
long start = System.currentTimeMillis();
boolean find_all_views = evt.getType() == Event.FIND_ALL_VIEWS;
Promise<JoinRsp> promise = (Promise<JoinRsp>) evt.getArg();
List<PingData> rsps = find_all_views ? findAllViews(promise) : findInitialMembers(promise);
long diff = System.currentTimeMillis() - start;
if (log.isTraceEnabled())
log.trace("discovery took " + diff + " ms: responses: " + Util.printPingData(rsps));
return rsps;
case Event.TMP_VIEW:
case Event.VIEW_CHANGE:
List<Address> tmp;
view = (View) evt.getArg();
if ((tmp = view.getMembers()) != null) {
synchronized (members) {
members.clear();
members.addAll(tmp);
}
}
current_coord = !members.isEmpty() ? members.get(0) : null;
is_coord = current_coord != null && local_addr != null && current_coord.equals(local_addr);
return down_prot.down(evt);
case Event.BECOME_SERVER: // called after client has joined and is fully working group member
down_prot.down(evt);
is_server = true;
return null;
case Event.SET_LOCAL_ADDRESS:
local_addr = (Address) evt.getArg();
return down_prot.down(evt);
case Event.CONNECT:
case Event.CONNECT_WITH_STATE_TRANSFER:
case Event.CONNECT_USE_FLUSH:
case Event.CONNECT_WITH_STATE_TRANSFER_USE_FLUSH:
is_leaving = false;
group_addr = (String) evt.getArg();
Object ret = down_prot.down(evt);
handleConnect();
return ret;
case Event.DISCONNECT:
is_leaving = true;
handleDisconnect();
return down_prot.down(evt);
default:
return down_prot.down(evt); // Pass on to the layer below us
}
}
/** Execute when new member join or leave Group */
public void viewAccepted(View v) {
memberSize = v.size();
if (mainFrame != null) setTitle();
members.clear();
members.addAll(v.getMembers());
if (v instanceof MergeView) {
System.out.println("** " + v);
// This is a simple merge function, which fetches the state from the coordinator
// on a merge and overwrites all of its own state
if (useState && !members.isEmpty()) {
Address coord = members.get(0);
Address local_addr = channel.getAddress();
if (local_addr != null && !local_addr.equals(coord)) {
try {
// make a copy of our state first
Map<Point, Color> copy = null;
if (send_own_state_on_merge) {
synchronized (drawPanel.state) {
copy = new LinkedHashMap<Point, Color>(drawPanel.state);
}
}
System.out.println("fetching state from " + coord);
channel.getState(coord, 5000);
if (copy != null)
sendOwnState(copy); // multicast my own state so everybody else has it too
} catch (Exception e) {
e.printStackTrace();
}
}
}
} else System.out.println("** View=" + v);
}
protected void handleViewChange(View v) {
Address old_coord;
List<Address> new_members = v.getMembers();
boolean send_up_exception = false;
this.view = v;
synchronized (members) {
old_coord = (!members.isEmpty() ? members.get(0) : null);
members.clear();
members.addAll(new_members);
// this handles the case where a coord dies during a state transfer; prevents clients from
// hanging forever
// Note this only takes a coordinator crash into account, a getState(target, timeout), where
// target is not
// null is not handled ! (Usually we get the state from the coordinator)
// http://jira.jboss.com/jira/browse/JGRP-148
if (waiting_for_state_response && old_coord != null && !members.contains(old_coord))
send_up_exception = true;
}
if (send_up_exception) {
log.warn("%s: discovered that the state provider (%s) left", local_addr, old_coord);
waiting_for_state_response = false;
Exception ex = new EOFException("state provider " + old_coord + " left");
up_prot.up(new Event(Event.GET_STATE_OK, new StateTransferResult(ex)));
openBarrierAndResumeStable();
}
// remove non members from list of members requesting state
state_requesters.retainAll(new_members);
}
private synchronized void handleViewChange(View view, boolean makeServer) {
if (makeServer) initializeNewSymmetricKey(view instanceof MergeView);
// if view is a bit broken set me as keyserver
List<Address> members = view.getMembers();
if (members == null || members.isEmpty() || members.get(0) == null) {
becomeKeyServer(local_addr, false);
return;
}
// otherwise get keyserver from view controller
Address tmpKeyServer = view.getMembers().get(0);
// I am keyserver - either first member of group or old key server is no more and
// I have been voted new controller
if (makeServer || (tmpKeyServer.equals(local_addr))) becomeKeyServer(tmpKeyServer, makeServer);
else handleNewKeyServer(tmpKeyServer, view instanceof MergeView);
}
/**
* Merge all MergeData. All MergeData elements should be disjunct (both views and digests).
* However, this method is prepared to resolve duplicate entries (for the same member).
* Resolution strategy for views is to merge only 1 of the duplicate members. Resolution
* strategy for digests is to take the higher seqnos for duplicate digests.
*
* <p>After merging all members into a Membership and subsequent sorting, the first member of
* the sorted membership will be the new coordinator. This method has a lock on merge_rsps.
*
* @param merge_rsps A list of MergeData items. Elements with merge_rejected=true were removed
* before. Is guaranteed not to be null and to contain at least 1 member.
*/
private MergeData consolidateMergeData(Vector<MergeData> merge_rsps) {
long logical_time = 0; // for new_vid
Membership new_mbrs = new Membership();
Vector<View> subgroups =
new Vector<View>(11); // contains a list of Views, each View is a subgroup
for (MergeData tmp_data : merge_rsps) {
View tmp_view = tmp_data.getView();
if (tmp_view != null) {
ViewId tmp_vid = tmp_view.getVid();
if (tmp_vid != null) {
// compute the new view id (max of all vids +1)
logical_time = Math.max(logical_time, tmp_vid.getId());
}
// merge all membership lists into one (prevent duplicates)
new_mbrs.add(tmp_view.getMembers());
subgroups.addElement((View) tmp_view.clone());
}
}
// the new coordinator is the first member of the consolidated & sorted membership list
new_mbrs.sort();
Address new_coord = new_mbrs.size() > 0 ? new_mbrs.elementAt(0) : null;
if (new_coord == null) {
if (log.isErrorEnabled()) log.error("new_coord == null");
return null;
}
// should be the highest view ID seen up to now plus 1
ViewId new_vid = new ViewId(new_coord, logical_time + 1);
// determine the new view
MergeView new_view = new MergeView(new_vid, new_mbrs.getMembers(), subgroups);
// determine the new digest
Digest new_digest = consolidateDigests(merge_rsps, new_mbrs.size());
if (new_digest == null) {
if (log.isErrorEnabled())
log.error("Merge leader " + gms.local_addr + ": could not consolidate digest for merge");
return null;
}
if (log.isDebugEnabled())
log.debug(
"Merge leader "
+ gms.local_addr
+ ": consolidated view="
+ new_view
+ "\nconsolidated digest="
+ new_digest);
return new MergeData(gms.local_addr, new_view, new_digest);
}
/**
* Merge all MergeData. All MergeData elements should be disjunct (both views and digests).
* However, this method is prepared to resolve duplicate entries (for the same member).
* Resolution strategy for views is to merge only 1 of the duplicate members. Resolution
* strategy for digests is to take the higher seqnos for duplicate digests.
*
* <p>After merging all members into a Membership and subsequent sorting, the first member of
* the sorted membership will be the new coordinator. This method has a lock on merge_rsps.
*
* @param merge_rsps A list of MergeData items. Elements with merge_rejected=true were removed
* before. Is guaranteed not to be null and to contain at least 1 member.
*/
private MergeData consolidateMergeData(List<MergeData> merge_rsps) {
long logical_time = 0; // for new_vid
List<View> subgroups =
new ArrayList<View>(11); // contains a list of Views, each View is a subgroup
Collection<Collection<Address>> sub_mbrships = new ArrayList<Collection<Address>>();
for (MergeData tmp_data : merge_rsps) {
View tmp_view = tmp_data.getView();
if (tmp_view != null) {
ViewId tmp_vid = tmp_view.getVid();
if (tmp_vid != null) {
// compute the new view id (max of all vids +1)
logical_time = Math.max(logical_time, tmp_vid.getId());
}
// merge all membership lists into one (prevent duplicates)
sub_mbrships.add(new ArrayList<Address>(tmp_view.getMembers()));
subgroups.add(tmp_view.copy());
}
}
// determine the new digest
Digest new_digest = consolidateDigests(merge_rsps, merge_rsps.size());
if (new_digest == null) return null;
// remove all members from the new member list that are not in the digest
Collection<Address> digest_mbrs = new_digest.getMembers();
for (Collection<Address> coll : sub_mbrships) coll.retainAll(digest_mbrs);
List<Address> merged_mbrs = gms.computeNewMembership(sub_mbrships);
// the new coordinator is the first member of the consolidated & sorted membership list
Address new_coord = merged_mbrs.isEmpty() ? null : merged_mbrs.get(0);
if (new_coord == null) return null;
// should be the highest view ID seen up to now plus 1
ViewId new_vid = new ViewId(new_coord, logical_time + 1);
// determine the new view
MergeView new_view = new MergeView(new_vid, merged_mbrs, subgroups);
if (log.isTraceEnabled())
log.trace(
gms.local_addr
+ ": consolidated view="
+ new_view
+ "\nconsolidated digest="
+ new_digest);
return new MergeData(gms.local_addr, new_view, new_digest);
}
protected void handleView(View view) {
this.view = view;
if (log.isDebugEnabled()) log.debug("view=" + view);
List<Address> members = view.getMembers();
_consumerLock.lock();
try {
// This removes the consumers that were registered that are now gone
Iterator<Owner> iterator = _consumersAvailable.iterator();
while (iterator.hasNext()) {
Owner owner = iterator.next();
if (!members.contains(owner.getAddress())) {
iterator.remove();
sendRemoveConsumerRequest(owner);
}
}
// This removes the tasks that those requestors are gone
iterator = _runRequests.iterator();
while (iterator.hasNext()) {
Owner owner = iterator.next();
if (!members.contains(owner.getAddress())) {
iterator.remove();
sendRemoveRunRequest(owner);
}
}
synchronized (_awaitingReturn) {
for (Entry<Owner, Runnable> entry : _awaitingReturn.entrySet()) {
// The person currently servicing our request has gone down
// without completing so we have to keep our request alive by
// sending ours back to the coordinator
Owner owner = entry.getKey();
if (!members.contains(owner.getAddress())) {
Runnable runnable = entry.getValue();
// We need to register the request id before sending the request back to the coordinator
// in case if our task gets picked up since another was removed
_requestId.put(runnable, owner.getRequestId());
_awaitingConsumer.add(runnable);
sendToCoordinator(RUN_REQUEST, owner.getRequestId(), local_addr);
}
}
}
} finally {
_consumerLock.unlock();
}
}
public void startFlush(List<Address> flushParticipants, boolean automatic_resume)
throws Exception {
if (!flushSupported())
throw new IllegalStateException(
"Flush is not supported, add pbcast.FLUSH protocol to your configuration");
View v = getView();
boolean validParticipants = v != null && v.getMembers().containsAll(flushParticipants);
if (!validParticipants)
throw new IllegalArgumentException(
"Current view " + v + " does not contain all flush participants " + flushParticipants);
try {
down(new Event(Event.SUSPEND, flushParticipants));
} catch (Exception e) {
throw new Exception("Flush failed", e.getCause());
} finally {
if (automatic_resume) stopFlush(flushParticipants);
}
}
protected void handleView(View view) {
this.view = view;
if (log.isDebugEnabled()) log.debug("view=" + view);
List<Address> members = view.getMembers();
for (Map.Entry<String, ServerLock> entry : server_locks.entrySet()) {
String lock_name = entry.getKey();
ServerLock server_lock = entry.getValue();
Lock lock = _getLock(lock_name);
lock.lock();
try {
server_lock.handleView(members);
if (server_lock.isEmpty()
&& server_lock.current_owner == null
&& server_lock.condition.queue.isEmpty()) server_locks.remove(lock_name);
} finally {
lock.unlock();
}
}
}
protected boolean _startFlush(
final View new_view, int maxAttempts, long randomFloor, long randomCeiling) {
if (!flushProtocolInStack) return true;
if (flushInvokerClass != null) {
try {
Callable<Boolean> invoker =
flushInvokerClass.getDeclaredConstructor(View.class).newInstance(new_view);
return invoker.call();
} catch (Throwable e) {
return false;
}
}
try {
boolean successfulFlush = false;
boolean validView = new_view != null && new_view.size() > 0;
if (validView && flushProtocolInStack) {
int attemptCount = 0;
while (attemptCount < maxAttempts) {
try {
up_prot.up(new Event(Event.SUSPEND, new ArrayList<Address>(new_view.getMembers())));
successfulFlush = true;
break;
} catch (Exception e) {
Util.sleepRandom(randomFloor, randomCeiling);
attemptCount++;
}
}
if (successfulFlush) {
if (log.isTraceEnabled()) log.trace(local_addr + ": successful GMS flush by coordinator");
} else {
if (log.isWarnEnabled()) log.warn(local_addr + ": GMS flush by coordinator failed");
}
}
return successfulFlush;
} catch (Exception e) {
return false;
}
}
/**
* Removes all members from a given view which don't have us in their view
* (https://jira.jboss.org/browse/JGRP-1061). Example:
*
* <pre>
* A: AB
* B: AB
* C: ABC
* </pre>
*
* becomes
*
* <pre>
* A: AB
* B: AB
* C: C // A and B don't have C in their views
* </pre>
*
* @param map A map of members and their associated views
*/
public static void sanitizeViews(Map<Address, View> map) {
if (map == null) return;
for (Map.Entry<Address, View> entry : map.entrySet()) {
Address key = entry.getKey();
Collection<Address> members = new ArrayList<Address>(entry.getValue().getMembers());
boolean modified = false;
for (Iterator<Address> it = members.iterator(); it.hasNext(); ) {
Address val = it.next();
if (val.equals(key)) // we can always talk to ourself !
continue;
View view = map.get(val);
final Collection<Address> tmp_mbrs = view != null ? view.getMembers() : null;
if (tmp_mbrs != null && !tmp_mbrs.contains(key)) {
it.remove();
modified = true;
}
}
if (modified) {
View old_view = entry.getValue();
entry.setValue(new View(old_view.getVid(), members));
}
}
}
protected static boolean contains(View view, Address... members) {
List<Address> mbrs = view.getMembers();
for (Address member : members) if (!mbrs.contains(member)) return false;
return true;
}
public Object down(Event evt) {
switch (evt.getType()) {
case Event.MSG: // Add UnicastHeader, add to AckSenderWindow and pass down
Message msg = (Message) evt.getArg();
Address dst = msg.getDest();
/* only handle unicast messages */
if (dst == null || msg.isFlagSet(Message.NO_RELIABILITY)) break;
if (!started) {
if (log.isTraceEnabled())
log.trace("discarded message as start() has not yet been called, message: " + msg);
return null;
}
SenderEntry entry = send_table.get(dst);
if (entry == null) {
entry = new SenderEntry(getNewConnectionId());
SenderEntry existing = send_table.putIfAbsent(dst, entry);
if (existing != null) entry = existing;
else {
if (log.isTraceEnabled())
log.trace(
local_addr
+ ": created connection to "
+ dst
+ " (conn_id="
+ entry.send_conn_id
+ ")");
if (cache != null && !members.contains(dst)) cache.add(dst);
}
}
long seqno = -2;
short send_conn_id = -1;
Unicast2Header hdr;
entry.lock(); // threads will only sync if they access the same entry
try {
seqno = entry.sent_msgs_seqno;
send_conn_id = entry.send_conn_id;
hdr = Unicast2Header.createDataHeader(seqno, send_conn_id, seqno == DEFAULT_FIRST_SEQNO);
msg.putHeader(this.id, hdr);
entry.sent_msgs.addToMessages(
seqno, msg); // add *including* UnicastHeader, adds to retransmitter
entry.sent_msgs_seqno++;
entry.update();
} finally {
entry.unlock();
}
if (log.isTraceEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append(local_addr)
.append(" --> DATA(")
.append(dst)
.append(": #")
.append(seqno)
.append(", conn_id=")
.append(send_conn_id);
if (hdr.first) sb.append(", first");
sb.append(')');
log.trace(sb);
}
try {
down_prot.down(evt);
num_msgs_sent++;
num_bytes_sent += msg.getLength();
} catch (Throwable t) {
log.warn("failed sending the message", t);
}
return null; // we already passed the msg down
case Event.VIEW_CHANGE: // remove connections to peers that are not members anymore !
View view = (View) evt.getArg();
List<Address> new_members = view.getMembers();
Set<Address> non_members = new HashSet<Address>(send_table.keySet());
non_members.addAll(recv_table.keySet());
synchronized (members) {
members.clear();
if (new_members != null) members.addAll(new_members);
non_members.removeAll(members);
if (cache != null) {
cache.removeAll(members);
}
}
if (!non_members.isEmpty()) {
if (log.isTraceEnabled()) log.trace("removing non members " + non_members);
for (Address non_mbr : non_members) removeConnection(non_mbr);
}
break;
case Event.SET_LOCAL_ADDRESS:
local_addr = (Address) evt.getArg();
break;
}
return down_prot.down(evt); // Pass on to the layer below us
}
public Object down(Event evt) {
switch (evt.getType()) {
case Event.MSG: // Add UnicastHeader, add to AckSenderWindow and pass down
Message msg = (Message) evt.getArg();
Address dst = msg.getDest();
/* only handle unicast messages */
if (dst == null || msg.isFlagSet(Message.NO_RELIABILITY)) break;
if (!running) {
if (log.isTraceEnabled())
log.trace("discarded message as start() has not yet been called, message: " + msg);
return null;
}
SenderEntry entry = send_table.get(dst);
if (entry == null) {
entry = new SenderEntry(getNewConnectionId());
SenderEntry existing = send_table.putIfAbsent(dst, entry);
if (existing != null) entry = existing;
else {
if (log.isTraceEnabled())
log.trace(
local_addr
+ ": created connection to "
+ dst
+ " (conn_id="
+ entry.send_conn_id
+ ")");
if (cache != null && !members.contains(dst)) cache.add(dst);
}
}
short send_conn_id = entry.send_conn_id;
long seqno = entry.sent_msgs_seqno.getAndIncrement();
long sleep = 10;
while (running) {
try {
msg.putHeader(
this.id,
Unicast2Header.createDataHeader(seqno, send_conn_id, seqno == DEFAULT_FIRST_SEQNO));
entry.sent_msgs.add(seqno, msg); // add *including* UnicastHeader, adds to retransmitter
if (conn_expiry_timeout > 0) entry.update();
break;
} catch (Throwable t) {
if (!running) break;
if (log.isWarnEnabled()) log.warn("failed sending message", t);
Util.sleep(sleep);
sleep = Math.min(5000, sleep * 2);
}
}
if (log.isTraceEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append(local_addr)
.append(" --> DATA(")
.append(dst)
.append(": #")
.append(seqno)
.append(", conn_id=")
.append(send_conn_id);
if (seqno == DEFAULT_FIRST_SEQNO) sb.append(", first");
sb.append(')');
log.trace(sb);
}
try {
down_prot.down(evt);
num_msgs_sent++;
} catch (Throwable t) {
log.warn("failed sending the message", t);
}
return null; // we already passed the msg down
case Event.VIEW_CHANGE: // remove connections to peers that are not members anymore !
View view = (View) evt.getArg();
List<Address> new_members = view.getMembers();
Set<Address> non_members = new HashSet<Address>(send_table.keySet());
non_members.addAll(recv_table.keySet());
members = new_members;
non_members.removeAll(new_members);
if (cache != null) cache.removeAll(new_members);
if (!non_members.isEmpty()) {
if (log.isTraceEnabled()) log.trace("removing non members " + non_members);
for (Address non_mbr : non_members) removeConnection(non_mbr);
}
break;
case Event.SET_LOCAL_ADDRESS:
local_addr = (Address) evt.getArg();
break;
}
return down_prot.down(evt); // Pass on to the layer below us
}
protected Object handleUpEvent(Event evt) throws Exception {
switch (evt.getType()) {
case Event.MSG:
if (msg_listener != null) msg_listener.receive((Message) evt.getArg());
break;
case Event.GET_APPLSTATE: // reply with GET_APPLSTATE_OK
byte[] tmp_state = null;
if (msg_listener != null) {
ByteArrayOutputStream output = new ByteArrayOutputStream(1024);
msg_listener.getState(output);
tmp_state = output.toByteArray();
}
return new StateTransferInfo(null, 0L, tmp_state);
case Event.GET_STATE_OK:
if (msg_listener != null) {
StateTransferResult result = (StateTransferResult) evt.getArg();
if (result.hasBuffer()) {
ByteArrayInputStream input = new ByteArrayInputStream(result.getBuffer());
msg_listener.setState(input);
}
}
break;
case Event.STATE_TRANSFER_OUTPUTSTREAM:
OutputStream os = (OutputStream) evt.getArg();
if (msg_listener != null && os != null) {
msg_listener.getState(os);
}
break;
case Event.STATE_TRANSFER_INPUTSTREAM:
InputStream is = (InputStream) evt.getArg();
if (msg_listener != null && is != null) msg_listener.setState(is);
break;
case Event.VIEW_CHANGE:
View v = (View) evt.getArg();
List<Address> new_mbrs = v.getMembers();
setMembers(new_mbrs);
if (membership_listener != null) membership_listener.viewAccepted(v);
break;
case Event.SET_LOCAL_ADDRESS:
if (log.isTraceEnabled())
log.trace("setting local_addr (" + local_addr + ") to " + evt.getArg());
local_addr = (Address) evt.getArg();
break;
case Event.SUSPECT:
if (membership_listener != null) membership_listener.suspect((Address) evt.getArg());
break;
case Event.BLOCK:
if (membership_listener != null) membership_listener.block();
break;
case Event.UNBLOCK:
if (membership_listener != null) membership_listener.unblock();
break;
}
return null;
}
/**
* Broadcasts the new view and digest, and waits for acks from all members in the list given as
* argument. If the list is null, we take the members who are part of new_view
*/
public void castViewChange(
View new_view, Digest digest, JoinRsp jr, Collection<Address> newMembers) {
if (log.isTraceEnabled())
log.trace(local_addr + ": mcasting view " + new_view + " (" + new_view.size() + " mbrs)\n");
// Send down a local TMP_VIEW event. This is needed by certain layers (e.g. NAKACK) to compute
// correct digest
// in case client's next request (e.g. getState()) reaches us *before* our own view change
// multicast.
// Check NAKACK's TMP_VIEW handling for details
down_prot.up(new Event(Event.TMP_VIEW, new_view));
down_prot.down(new Event(Event.TMP_VIEW, new_view));
List<Address> ackMembers = new ArrayList<Address>(new_view.getMembers());
if (newMembers != null && !newMembers.isEmpty()) ackMembers.removeAll(newMembers);
Message view_change_msg = new Message(); // bcast to all members
GmsHeader hdr = new GmsHeader(GmsHeader.VIEW, new_view);
hdr.my_digest = digest;
view_change_msg.putHeader(this.id, hdr);
// If we're the only member the VIEW is broadcast to, let's simply install the view directly,
// without
// sending the VIEW multicast ! Or else N-1 members drop the multicast anyway...
if (local_addr != null && ackMembers.size() == 1 && ackMembers.get(0).equals(local_addr)) {
// we need to add the message to the retransmit window (e.g. in NAKACK), so (1) it can be
// retransmitted and
// (2) we increment the seqno (otherwise, we'd return an incorrect digest)
down_prot.down(new Event(Event.ADD_TO_XMIT_TABLE, view_change_msg));
impl.handleViewChange(new_view, digest);
} else {
if (!ackMembers.isEmpty()) ack_collector.reset(ackMembers);
down_prot.down(new Event(Event.MSG, view_change_msg));
try {
if (!ackMembers.isEmpty()) {
ack_collector.waitForAllAcks(view_ack_collection_timeout);
if (log.isTraceEnabled())
log.trace(
local_addr
+ ": received all "
+ ack_collector.expectedAcks()
+ " ACKs from members for view "
+ new_view.getVid());
}
} catch (TimeoutException e) {
if (log_collect_msgs && log.isWarnEnabled()) {
log.warn(
local_addr
+ ": failed to collect all ACKs (expected="
+ ack_collector.expectedAcks()
+ ") for view "
+ new_view.getViewId()
+ " after "
+ view_ack_collection_timeout
+ "ms, missing ACKs from "
+ ack_collector.printMissing());
}
}
}
if (jr != null && (newMembers != null && !newMembers.isEmpty())) {
ack_collector.reset(new ArrayList<Address>(newMembers));
for (Address joiner : newMembers) {
sendJoinResponse(jr, joiner);
}
try {
ack_collector.waitForAllAcks(view_ack_collection_timeout);
if (log.isTraceEnabled())
log.trace(
local_addr
+ ": received all ACKs ("
+ ack_collector.expectedAcks()
+ ") from joiners for view "
+ new_view.getVid());
} catch (TimeoutException e) {
if (log_collect_msgs && log.isWarnEnabled()) {
log.warn(
local_addr
+ ": failed to collect all ACKs (expected="
+ ack_collector.expectedAcks()
+ ") for unicast view "
+ new_view
+ " after "
+ view_ack_collection_timeout
+ "ms, missing ACKs from "
+ ack_collector.printMissing());
}
}
}
}
/**
* Sets the new view and sends a VIEW_CHANGE event up and down the stack. If the view is a
* MergeView (subclass of View), then digest will be non-null and has to be set before installing
* the view.
*/
public void installView(View new_view, Digest digest) {
ViewId vid = new_view.getVid();
List<Address> mbrs = new_view.getMembers();
ltime =
Math.max(
vid.getId(),
ltime); // compute the logical time, regardless of whether the view is accepted
// Discards view with id lower than or equal to our own. Will be installed without check if it
// is the first view
if (view != null) {
ViewId view_id = view.getViewId();
int rc = vid.compareToIDs(view_id);
if (rc <= 0) {
if (log.isWarnEnabled()
&& rc < 0
&& log_view_warnings) { // only scream if view is smaller, silently discard same views
log.warn(
local_addr
+ ": received view < current view;"
+ " discarding it (current vid: "
+ view_id
+ ", new vid: "
+ vid
+ ')');
}
return;
}
}
/* Check for self-inclusion: if I'm not part of the new membership, I just discard it.
This ensures that messages sent in view V1 are only received by members of V1 */
if (!mbrs.contains(local_addr)) {
if (log.isWarnEnabled() && log_view_warnings)
log.warn(local_addr + ": not member of view " + new_view.getViewId() + "; discarding it");
return;
}
if (digest != null) {
if (new_view instanceof MergeView) mergeDigest(digest);
else setDigest(digest);
}
if (log.isDebugEnabled()) log.debug(local_addr + ": installing view " + new_view);
Event view_event;
synchronized (members) {
view = new View(new_view.getVid(), new_view.getMembers());
view_event = new Event(Event.VIEW_CHANGE, new_view);
// Set the membership. Take into account joining members
if (!mbrs.isEmpty()) {
members.set(mbrs);
tmp_members.set(members);
joining.removeAll(mbrs); // remove all members in mbrs from joining
// remove all elements from 'leaving' that are not in 'mbrs'
leaving.retainAll(mbrs);
tmp_members.add(joining); // add members that haven't yet shown up in the membership
tmp_members.remove(
leaving); // remove members that haven't yet been removed from the membership
// add to prev_members
for (Address addr : mbrs) {
if (!prev_members.contains(addr)) prev_members.add(addr);
}
}
Address coord = determineCoordinator();
if (coord != null && coord.equals(local_addr) && !haveCoordinatorRole()) {
becomeCoordinator();
} else {
if (haveCoordinatorRole() && !local_addr.equals(coord)) {
becomeParticipant();
merge_ack_collector.reset(null); // we don't need this one anymore
}
}
}
// - Changed order of passing view up and down (http://jira.jboss.com/jira/browse/JGRP-347)
// - Changed it back (bela Sept 4 2007): http://jira.jboss.com/jira/browse/JGRP-564
// - Moved sending up view_event out of the synchronized block (bela Nov 2011)
down_prot.down(view_event); // needed e.g. by failure detector or UDP
up_prot.up(view_event);
List<Address> tmp_mbrs = new_view.getMembers();
ack_collector.retainAll(tmp_mbrs);
merge_ack_collector.retainAll(tmp_mbrs);
if (new_view instanceof MergeView) merger.forceCancelMerge();
if (stats) {
num_views++;
prev_views.add(new Tuple<View, Long>(new_view, System.currentTimeMillis()));
}
}
/**
* Callback method <br>
* Called by the ProtocolStack when a message is received.
*
* @param evt the event carrying the message from the protocol stack
*/
public Object up(Event evt) {
switch (evt.getType()) {
case Event.MSG:
Message msg = (Message) evt.getArg();
if (stats) {
received_msgs++;
received_bytes += msg.getLength();
}
// discard local messages (sent by myself to me)
if (discard_own_messages
&& local_addr != null
&& msg.getSrc() != null
&& local_addr.equals(msg.getSrc())) return null;
break;
case Event.VIEW_CHANGE:
View tmp = (View) evt.getArg();
if (tmp instanceof MergeView) my_view = new View(tmp.getViewId(), tmp.getMembers());
else my_view = tmp;
// Bela&Vladimir Oct 27th,2006 (JGroups 2.4): we need to set connected=true because a client
// can
// call channel.getView() in viewAccepted() callback invoked on this thread (see
// Event.VIEW_CHANGE handling below)
// not good: we are only connected when we returned from connect() - bela June 22 2007
// Changed: when a channel gets a view of which it is a member then it should be
// connected even if connect() hasn't returned yet ! (bela Noc 2010)
if (state != State.CONNECTED) state = State.CONNECTED;
break;
case Event.CONFIG:
Map<String, Object> cfg = (Map<String, Object>) evt.getArg();
if (cfg != null) {
if (cfg.containsKey("state_transfer")) {
state_transfer_supported = (Boolean) cfg.get("state_transfer");
}
if (cfg.containsKey("flush_supported")) {
flush_supported = (Boolean) cfg.get("flush_supported");
}
}
break;
case Event.GET_STATE_OK:
StateTransferResult result = (StateTransferResult) evt.getArg();
if (up_handler != null) {
try {
Object retval = up_handler.up(evt);
state_promise.setResult(new StateTransferResult());
return retval;
} catch (Throwable t) {
state_promise.setResult(new StateTransferResult(t));
}
}
if (receiver != null) {
try {
if (result.hasBuffer()) {
byte[] tmp_state = result.getBuffer();
ByteArrayInputStream input = new ByteArrayInputStream(tmp_state);
receiver.setState(input);
}
state_promise.setResult(result);
} catch (Throwable t) {
state_promise.setResult(new StateTransferResult(t));
}
}
break;
case Event.STATE_TRANSFER_INPUTSTREAM_CLOSED:
state_promise.setResult((StateTransferResult) evt.getArg());
break;
case Event.STATE_TRANSFER_INPUTSTREAM:
// Oct 13,2006 moved to down() when Event.STATE_TRANSFER_INPUTSTREAM_CLOSED is received
// state_promise.setResult(is != null? Boolean.TRUE : Boolean.FALSE);
if (up_handler != null) return up_handler.up(evt);
InputStream is = (InputStream) evt.getArg();
if (is != null && receiver != null) {
try {
receiver.setState(is);
} catch (Throwable t) {
throw new RuntimeException("failed calling setState() in state requester", t);
}
}
break;
case Event.STATE_TRANSFER_OUTPUTSTREAM:
if (receiver != null && evt.getArg() != null) {
try {
receiver.getState((OutputStream) evt.getArg());
} catch (Exception e) {
throw new RuntimeException("failed calling getState() in state provider", e);
}
}
break;
case Event.GET_LOCAL_ADDRESS:
return local_addr;
default:
break;
}
// If UpHandler is installed, pass all events to it and return (UpHandler is e.g. a building
// block)
if (up_handler != null) return up_handler.up(evt);
if (receiver != null) return invokeCallback(evt.getType(), evt.getArg());
return null;
}
Address determineCoordinator() {
List<Address> mbrs = my_view != null ? my_view.getMembers() : null;
if (mbrs == null) return null;
if (!mbrs.isEmpty()) return mbrs.iterator().next();
return null;
}
@SuppressWarnings("unchecked")
public Object down(Event evt) {
switch (evt.getType()) {
case Event.FIND_INITIAL_MBRS: // sent by GMS layer
return findMembers(null, true, false); // triggered by JOIN process (ClientGmsImpl)
case Event.FIND_MBRS:
return findMembers(
(List<Address>) evt.getArg(), false, false); // triggered by MERGE2/MERGE3
// case Event.TMP_VIEW:
case Event.VIEW_CHANGE:
List<Address> tmp;
View old_view = view;
view = (View) evt.getArg();
if ((tmp = view.getMembers()) != null) {
synchronized (members) {
members.clear();
members.addAll(tmp);
}
}
current_coord = !members.isEmpty() ? members.get(0) : null;
is_coord = current_coord != null && local_addr != null && current_coord.equals(local_addr);
Object retval = down_prot.down(evt);
if (send_cache_on_join && !isDynamic() && is_coord) {
List<Address> curr_mbrs, left_mbrs, new_mbrs;
synchronized (members) {
curr_mbrs = new ArrayList<>(members);
left_mbrs = old_view != null ? Util.leftMembers(old_view.getMembers(), members) : null;
new_mbrs = old_view != null ? Util.newMembers(old_view.getMembers(), members) : null;
}
startCacheDissemination(curr_mbrs, left_mbrs, new_mbrs); // separate task
}
return retval;
case Event.BECOME_SERVER: // called after client has joined and is fully working group member
down_prot.down(evt);
is_server = true;
return null;
case Event.SET_LOCAL_ADDRESS:
local_addr = (Address) evt.getArg();
return down_prot.down(evt);
case Event.CONNECT:
case Event.CONNECT_WITH_STATE_TRANSFER:
case Event.CONNECT_USE_FLUSH:
case Event.CONNECT_WITH_STATE_TRANSFER_USE_FLUSH:
is_leaving = false;
cluster_name = (String) evt.getArg();
Object ret = down_prot.down(evt);
handleConnect();
return ret;
case Event.DISCONNECT:
is_leaving = true;
handleDisconnect();
return down_prot.down(evt);
default:
return down_prot.down(evt); // Pass on to the layer below us
}
}