public void sendDiscoveryRequest(String cluster_name, Promise promise, ViewId view_id)
throws Exception {
PingData data = null;
PhysicalAddress physical_addr =
(PhysicalAddress) down(new Event(Event.GET_PHYSICAL_ADDRESS, local_addr));
if (view_id == null) {
List<PhysicalAddress> physical_addrs = Arrays.asList(physical_addr);
data = new PingData(local_addr, null, false, UUID.get(local_addr), physical_addrs);
}
PingHeader hdr = new PingHeader(PingHeader.GET_MBRS_REQ, data, cluster_name);
hdr.view_id = view_id;
Collection<PhysicalAddress> cluster_members = fetchClusterMembers(cluster_name);
if (cluster_members == null) {
Message msg = new Message(null); // multicast msg
msg.setFlag(Message.OOB);
msg.putHeader(getId(), hdr);
sendMcastDiscoveryRequest(msg);
} else {
if (cluster_members.isEmpty()) { // if we don't find any members, return immediately
if (promise != null) promise.setResult(null);
} else {
for (final Address addr : cluster_members) {
if (addr.equals(physical_addr)) // no need to send the request to myself
continue;
final Message msg = new Message(addr, null, null);
msg.setFlag(Message.OOB);
msg.putHeader(this.id, hdr);
if (log.isTraceEnabled())
log.trace("[FIND_INITIAL_MBRS] sending discovery request to " + msg.getDest());
if (!sendDiscoveryRequestsInParallel()) {
down_prot.down(new Event(Event.MSG, msg));
} else {
timer.execute(
new Runnable() {
public void run() {
try {
down_prot.down(new Event(Event.MSG, msg));
} catch (Exception ex) {
if (log.isErrorEnabled())
log.error("failed sending discovery request to " + addr + ": " + ex);
}
}
});
}
}
}
}
}
/**
* Sends the new view and digest to all subgroup coordinors in coords. Each coord will in turn
*
* <ol>
* <li>broadcast the new view and digest to all the members of its subgroup (MergeView)
* <li>on reception of the view, if it is a MergeView, each member will set the digest and
* install the new view
* </ol>
*/
private void sendMergeView(
Collection<Address> coords, MergeData combined_merge_data, MergeId merge_id) {
if (coords == null || combined_merge_data == null) return;
View view = combined_merge_data.view;
Digest digest = combined_merge_data.digest;
if (view == null || digest == null) {
if (log.isErrorEnabled())
log.error("view or digest is null, cannot send consolidated merge view/digest");
return;
}
if (log.isDebugEnabled())
log.debug(
gms.local_addr + ": sending merge view " + view.getVid() + " to coordinators " + coords);
gms.merge_ack_collector.reset(coords);
int size = gms.merge_ack_collector.size();
long timeout = gms.view_ack_collection_timeout;
long start = System.currentTimeMillis();
for (Address coord : coords) {
Message msg = new Message(coord, null, null);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.INSTALL_MERGE_VIEW);
hdr.view = view;
hdr.my_digest = digest;
hdr.merge_id = merge_id;
msg.putHeader(gms.getId(), hdr);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
// [JGRP-700] - FLUSH: flushing should span merge
// if flush is in stack wait for acks from separated island coordinators
if (gms.flushProtocolInStack) {
try {
gms.merge_ack_collector.waitForAllAcks(timeout);
long stop = System.currentTimeMillis();
if (log.isTraceEnabled())
log.trace(
"received all ACKs ("
+ size
+ ") for merge view "
+ view
+ " in "
+ (stop - start)
+ "ms");
} catch (TimeoutException e) {
log.warn(
gms.local_addr
+ ": failed to collect all ACKs for merge ("
+ size
+ ") for view "
+ view
+ " after "
+ timeout
+ "ms, missing ACKs from "
+ gms.merge_ack_collector.printMissing());
}
}
}
/**
* Sends a MERGE_REQ to all coords and populates a list of MergeData (in merge_rsps). Returns
* after coords.size() response have been received, or timeout msecs have elapsed (whichever is
* first).
*
* <p>If a subgroup coordinator rejects the MERGE_REQ (e.g. because of participation in a
* different merge), <em>that member will be removed from coords !</em>
*
* @param coords A map of coordinatgor addresses and associated membership lists
* @param new_merge_id The new merge id
* @param timeout Max number of msecs to wait for the merge responses from the subgroup coords
*/
private boolean getMergeDataFromSubgroupCoordinators(
Map<Address, Collection<Address>> coords, MergeId new_merge_id, long timeout) {
boolean gotAllResponses;
long start = System.currentTimeMillis();
merge_rsps.reset(coords.keySet());
if (log.isDebugEnabled())
log.debug(gms.local_addr + ": sending MERGE_REQ to " + coords.keySet());
for (Map.Entry<Address, Collection<Address>> entry : coords.entrySet()) {
Address coord = entry.getKey();
Collection<Address> mbrs = entry.getValue();
Message msg = new Message(coord, null, null);
msg.setFlag(Message.OOB);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.MERGE_REQ, mbrs);
hdr.mbr = gms.local_addr;
hdr.merge_id = new_merge_id;
msg.putHeader(gms.getId(), hdr);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
// wait until num_rsps_expected >= num_rsps or timeout elapsed
merge_rsps.waitForAllResponses(timeout);
gotAllResponses = merge_rsps.hasAllResponses();
long stop = System.currentTimeMillis();
if (log.isDebugEnabled())
log.debug(
gms.local_addr
+ ": collected "
+ merge_rsps.size()
+ " merge response(s) in "
+ (stop - start)
+ " ms");
return gotAllResponses;
}
public void retransmit(long first_seqno, long last_seqno, Address sender) {
if (last_seqno < first_seqno) return;
Unicast2Header hdr = Unicast2Header.createXmitReqHeader(first_seqno, last_seqno);
Message xmit_req = new Message(sender, null, null);
xmit_req.putHeader(this.id, hdr);
down_prot.down(new Event(Event.MSG, xmit_req));
}
protected void sendStableMessage(Address dest, short conn_id, long low, long high) {
Message stable_msg = new Message(dest, null, null);
Unicast2Header hdr = Unicast2Header.createStableHeader(conn_id, low, high);
stable_msg.putHeader(this.id, hdr);
stable_msg.setFlag(Message.OOB);
if (log.isTraceEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append(local_addr)
.append(" --> STABLE(")
.append(dest)
.append(": ")
.append(low)
.append("-")
.append(high)
.append(", conn_id=")
.append(conn_id)
.append(")");
log.trace(sb.toString());
}
down_prot.down(new Event(Event.MSG, stable_msg));
ReceiverEntry entry = recv_table.get(dest);
NakReceiverWindow win = entry != null ? entry.received_msgs : null;
if (win != null) win.stable(win.getHighestDelivered());
}
private void sendViewAck(Address dest) {
Message view_ack = new Message(dest, null, null);
view_ack.setFlag(Message.OOB);
GmsHeader tmphdr = new GmsHeader(GmsHeader.VIEW_ACK);
view_ack.putHeader(this.id, tmphdr);
down_prot.down(new Event(Event.MSG, view_ack));
}
/**
* We need to resend our first message with our conn_id
*
* @param sender
* @param seqno Resend the non null messages in the range [lowest .. seqno]
*/
protected void handleResendingOfFirstMessage(Address sender, long seqno) {
if (log.isTraceEnabled())
log.trace(local_addr + " <-- SEND_FIRST_SEQNO(" + sender + "," + seqno + ")");
SenderEntry entry = send_table.get(sender);
Table<Message> win = entry != null ? entry.sent_msgs : null;
if (win == null) {
if (log.isErrorEnabled())
log.error(local_addr + ": sender window for " + sender + " not found");
return;
}
boolean first_sent = false;
for (long i = win.getLow() + 1; i <= seqno; i++) {
Message rsp = win.get(i);
if (rsp == null) continue;
if (first_sent) {
down_prot.down(new Event(Event.MSG, rsp));
} else {
first_sent = true;
// We need to copy the UnicastHeader and put it back into the message because Message.copy()
// doesn't copy
// the headers and therefore we'd modify the original message in the sender retransmission
// window
// (https://jira.jboss.org/jira/browse/JGRP-965)
Message copy = rsp.copy();
Unicast2Header hdr = (Unicast2Header) copy.getHeader(this.id);
Unicast2Header newhdr = hdr.copy();
newhdr.first = true;
copy.putHeader(this.id, newhdr);
down_prot.down(new Event(Event.MSG, copy));
}
}
}
/**
* Multicasts a GET_DIGEST_REQ to all current members and waits for all responses (GET_DIGEST_RSP)
* or N ms.
*
* @return
*/
private Digest fetchDigestsFromAllMembersInSubPartition(List<Address> current_mbrs) {
if (current_mbrs == null) return null;
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.GET_DIGEST_REQ);
Message get_digest_req = new Message();
get_digest_req.setFlag(Message.OOB);
get_digest_req.putHeader(gms.getId(), hdr);
long max_wait_time = gms.merge_timeout > 0 ? gms.merge_timeout / 2 : 2000L;
digest_collector.reset(current_mbrs);
// add my own digest first
Digest digest = (Digest) gms.getDownProtocol().down(Event.GET_DIGEST_EVT);
digest_collector.add(gms.local_addr, digest);
gms.getDownProtocol().down(new Event(Event.MSG, get_digest_req));
digest_collector.waitForAllResponses(max_wait_time);
if (log.isDebugEnabled()) {
if (digest_collector.hasAllResponses())
log.debug(gms.local_addr + ": fetched all digests for " + current_mbrs);
else
log.debug(
gms.local_addr
+ ": fetched incomplete digests (after timeout of "
+ max_wait_time
+ ") ms for "
+ current_mbrs);
}
Map<Address, Digest> responses = new HashMap<Address, Digest>(digest_collector.getResults());
MutableDigest retval = new MutableDigest(responses.size());
for (Digest dig : responses.values()) {
if (dig != null) retval.add(dig);
}
return retval;
}
/**
* If merge_id is not equal to this.merge_id then discard. Else cast the view/digest to all
* members of this group.
*/
public void handleMergeView(final MergeData data, final MergeId merge_id) {
if (!matchMergeId(merge_id)) {
if (log.isErrorEnabled())
log.error("merge_ids don't match (or are null); merge view discarded");
return;
}
// only send to our *current* members, if we have A and B being merged (we are B), then we would
// *not*
// receive a VIEW_ACK from A because A doesn't see us in the pre-merge view yet and discards the
// view
// [JGRP-700] - FLUSH: flushing should span merge
// we have to send new view only to current members and we should not wait
// for view acks from newly merged mebers
List<Address> newViewMembers = new Vector<Address>(data.view.getMembers());
newViewMembers.removeAll(gms.members.getMembers());
gms.castViewChangeWithDest(data.view, data.digest, null, newViewMembers);
// if we have flush in stack send ack back to merge coordinator
if (gms.flushProtocolInStack) {
Message ack = new Message(data.getSender(), null, null);
ack.setFlag(Message.OOB);
GMS.GmsHeader ack_hdr = new GMS.GmsHeader(GMS.GmsHeader.INSTALL_MERGE_VIEW_OK);
ack.putHeader(gms.getId(), ack_hdr);
gms.getDownProtocol().down(new Event(Event.MSG, ack));
}
cancelMerge(merge_id);
}
public void sendGetMembersRequest(String cluster_name, Promise promise, boolean return_views_only)
throws Exception {
PhysicalAddress physical_addr =
(PhysicalAddress) down_prot.down(new Event(Event.GET_PHYSICAL_ADDRESS, local_addr));
PingData data =
new PingData(local_addr, null, false, UUID.get(local_addr), Arrays.asList(physical_addr));
PingHeader hdr = new PingHeader(PingHeader.GET_MBRS_REQ, data, cluster_name);
hdr.return_view_only = return_views_only;
Set<PhysicalAddress> combined_target_members = new HashSet<PhysicalAddress>(initial_hosts);
combined_target_members.addAll(dynamic_hosts);
for (final Address addr : combined_target_members) {
if (addr.equals(physical_addr)) continue;
final Message msg = new Message(addr, null, null);
msg.setFlag(Message.OOB);
msg.putHeader(this.id, hdr);
if (log.isTraceEnabled())
log.trace("[FIND_INITIAL_MBRS] sending PING request to " + msg.getDest());
timer.execute(
new Runnable() {
public void run() {
try {
down_prot.down(new Event(Event.MSG, msg));
} catch (Exception ex) {
if (log.isErrorEnabled())
log.error("failed sending discovery request to " + addr + ": " + ex);
}
}
});
}
}
protected void sendRequestForFirstSeqno(Address dest, long seqno_received) {
Message msg = new Message(dest).setFlag(Message.Flag.OOB, Message.Flag.INTERNAL);
UnicastHeader hdr = UnicastHeader.createSendFirstSeqnoHeader(seqno_received);
msg.putHeader(this.id, hdr);
if (log.isTraceEnabled())
log.trace(local_addr + " --> SEND_FIRST_SEQNO(" + dest + "," + seqno_received + ")");
down_prot.down(new Event(Event.MSG, msg));
}
protected void sendMergeRejectedResponse(Address sender, MergeId merge_id) {
Message msg = new Message(sender).setFlag(Message.Flag.OOB, Message.Flag.INTERNAL);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.MERGE_RSP);
hdr.merge_rejected = true;
hdr.merge_id = merge_id;
msg.putHeader(gms.getId(), hdr);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
/**
* Fetches the digests from all members and installs them again. Used only for diagnosis and
* support; don't use this otherwise !
*/
void fixDigests() {
Digest digest = fetchDigestsFromAllMembersInSubPartition(gms.view.getMembers());
Message msg = new Message();
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.INSTALL_DIGEST);
hdr.my_digest = digest;
msg.putHeader(gms.getId(), hdr);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
protected void sendMergeRejectedResponse(Address sender, MergeId merge_id) {
Message msg = new Message(sender, null, null);
msg.setFlag(Message.OOB);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.MERGE_RSP);
hdr.merge_rejected = true;
hdr.merge_id = merge_id;
msg.putHeader(gms.getId(), hdr);
if (log.isDebugEnabled()) log.debug("merge response=" + hdr);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
/** Send back a response containing view and digest to sender */
private void sendMergeResponse(Address sender, View view, Digest digest, MergeId merge_id) {
Message msg = new Message(sender).setFlag(Message.Flag.OOB, Message.Flag.INTERNAL);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.MERGE_RSP);
hdr.merge_id = merge_id;
hdr.view = view;
hdr.my_digest = digest;
msg.putHeader(gms.getId(), hdr);
if (log.isTraceEnabled()) log.trace(gms.local_addr + ": sending merge response=" + hdr);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
public void retransmit(SeqnoList missing, Address sender) {
Unicast2Header hdr = Unicast2Header.createXmitReqHeader();
Message retransmit_msg = new Message(sender, null, missing);
retransmit_msg.setFlag(Message.OOB);
if (log.isTraceEnabled())
log.trace(local_addr + ": sending XMIT_REQ (" + missing + ") to " + sender);
retransmit_msg.putHeader(this.id, hdr);
down_prot.down(new Event(Event.MSG, retransmit_msg));
xmit_reqs_sent.addAndGet(missing.size());
}
private void sendMergeCancelledMessage(Collection<Address> coords, MergeId merge_id) {
if (coords == null || merge_id == null) return;
for (Address coord : coords) {
Message msg = new Message(coord);
// msg.setFlag(Message.Flag.OOB);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.CANCEL_MERGE);
hdr.merge_id = merge_id;
msg.putHeader(gms.getId(), hdr);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
}
private void sendMergeCancelledMessage(Collection<Address> coords, MergeId merge_id) {
if (coords == null || merge_id == null) return;
for (Address coord : coords) {
Message msg = new Message(coord, null, null);
// msg.setFlag(Message.OOB);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.CANCEL_MERGE);
hdr.merge_id = merge_id;
msg.putHeader(gms.getId(), hdr);
if (log.isDebugEnabled()) log.debug(gms.local_addr + ": sending cancel merge to " + coord);
gms.getDownProtocol().down(new Event(Event.MSG, msg));
}
}
protected void sendDiscoveryResponse(
Address logical_addr,
List<PhysicalAddress> physical_addrs,
boolean is_server,
boolean return_view_only,
String logical_name,
final Address sender) {
PingData data;
if (return_view_only) {
data = new PingData(logical_addr, view, is_server, null, null);
} else {
ViewId view_id = view != null ? view.getViewId() : null;
data = new PingData(logical_addr, null, view_id, is_server, logical_name, physical_addrs);
}
final Message rsp_msg = new Message(sender, null, null);
rsp_msg.setFlag(Message.OOB);
final PingHeader rsp_hdr = new PingHeader(PingHeader.GET_MBRS_RSP, data);
rsp_msg.putHeader(this.id, rsp_hdr);
if (stagger_timeout > 0) {
int view_size = view != null ? view.size() : 10;
int rank = Util.getRank(view, local_addr); // returns 0 if view or local_addr are null
long sleep_time =
rank == 0
? Util.random(stagger_timeout)
: stagger_timeout * rank / view_size - (stagger_timeout / view_size);
timer.schedule(
new Runnable() {
public void run() {
if (log.isTraceEnabled())
log.trace(
local_addr
+ ": received GET_MBRS_REQ from "
+ sender
+ ", sending staggered response "
+ rsp_hdr);
down_prot.down(new Event(Event.MSG, rsp_msg));
}
},
sleep_time,
TimeUnit.MILLISECONDS);
return;
}
if (log.isTraceEnabled())
log.trace("received GET_MBRS_REQ from " + sender + ", sending response " + rsp_hdr);
down_prot.down(new Event(Event.MSG, rsp_msg));
}
/**
* We need to resend our first message with our conn_id
*
* @param sender
* @param seqno Resend messages in the range [lowest .. seqno]
*/
private void handleResendingOfFirstMessage(Address sender, long seqno) {
if (log.isTraceEnabled())
log.trace(local_addr + " <-- SEND_FIRST_SEQNO(" + sender + "," + seqno + ")");
SenderEntry entry = send_table.get(sender);
AckSenderWindow win = entry != null ? entry.sent_msgs : null;
if (win == null) {
if (log.isErrorEnabled())
log.error(local_addr + ": sender window for " + sender + " not found");
return;
}
long lowest = win.getLowest();
Message rsp = win.get(lowest);
if (rsp == null) return;
// We need to copy the UnicastHeader and put it back into the message because Message.copy()
// doesn't copy
// the headers and therefore we'd modify the original message in the sender retransmission
// window
// (https://jira.jboss.org/jira/browse/JGRP-965)
Message copy = rsp.copy();
Unicast2Header hdr = (Unicast2Header) copy.getHeader(this.id);
Unicast2Header newhdr = hdr.copy();
newhdr.first = true;
copy.putHeader(this.id, newhdr);
if (log.isTraceEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append(local_addr)
.append(" --> DATA(")
.append(copy.getDest())
.append(": #")
.append(newhdr.seqno)
.append(", conn_id=")
.append(newhdr.conn_id);
if (newhdr.first) sb.append(", first");
sb.append(')');
log.trace(sb);
}
down_prot.down(new Event(Event.MSG, copy));
if (++lowest > seqno) return;
for (long i = lowest; i <= seqno; i++) {
rsp = win.get(i);
if (rsp != null) down_prot.down(new Event(Event.MSG, rsp));
}
}
protected void sendStableMessage(Address dest, short conn_id, long hd, long hr) {
Message stable_msg = new Message(dest, null, null);
Unicast2Header hdr = Unicast2Header.createStableHeader(conn_id, hd, hr);
stable_msg.putHeader(this.id, hdr);
stable_msg.setFlag(Message.OOB);
if (log.isTraceEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append(local_addr)
.append(" --> STABLE(")
.append(dest)
.append(": ")
.append(hd)
.append("-")
.append(hr)
.append(", conn_id=")
.append(conn_id)
.append(")");
log.trace(sb.toString());
}
down_prot.down(new Event(Event.MSG, stable_msg));
}
protected void sendAck(Address dst, long seqno) {
if (!running) // if we are disconnected, then don't send any acks which throw exceptions on
// shutdown
return;
Message ack = new Message(dst);
ack.putHeader(this.id, UnicastHeader.createAckHeader(seqno));
if (log.isTraceEnabled())
log.trace(
new StringBuilder()
.append(local_addr)
.append(" --> ACK(")
.append(dst)
.append(": #")
.append(seqno)
.append(')'));
try {
down_prot.down(new Event(Event.MSG, ack));
num_acks_sent++;
} catch (Throwable t) {
log.error("failed sending ACK(" + seqno + ") to " + dst, t);
}
}
/**
* If merge_id is not equal to this.merge_id then discard. Else cast the view/digest to all
* members of this group.
*/
public void handleMergeView(final MergeData data, final MergeId merge_id) {
if (!matchMergeId(merge_id)) {
if (log.isTraceEnabled())
log.trace(
gms.local_addr
+ ": merge_ids (mine: "
+ this.merge_id
+ ", received: "
+ merge_id
+ ") don't match; merge view "
+ data.view.getViewId()
+ " is discarded");
return;
}
// only send to our *current* members, if we have A and B being merged (we are B), then we would
// *not*
// want to block on a VIEW_ACK from A because A doesn't see us in the pre-merge view yet and
// discards the view
List<Address> newViewMembers = new ArrayList<Address>(data.view.getMembers());
newViewMembers.removeAll(gms.members.getMembers());
try {
gms.castViewChange(data.view, data.digest, null, newViewMembers);
// if we have flush in stack send ack back to merge coordinator
if (gms.flushProtocolInStack) { // [JGRP-700] - FLUSH: flushing should span merge
Message ack =
new Message(data.getSender()).setFlag(Message.Flag.OOB, Message.Flag.INTERNAL);
GMS.GmsHeader ack_hdr = new GMS.GmsHeader(GMS.GmsHeader.INSTALL_MERGE_VIEW_OK);
ack.putHeader(gms.getId(), ack_hdr);
gms.getDownProtocol().down(new Event(Event.MSG, ack));
}
} finally {
cancelMerge(merge_id);
}
}
/**
* Caller by the layer above this layer. Usually we just put this Message into the send queue and
* let one or more worker threads handle it. A worker thread then removes the Message from the
* send queue, performs a conversion and adds the modified Message to the send queue of the layer
* below it, by calling Down).
*/
public void down(Event evt) {
Message msg;
Object dest_addr;
if (evt.getType() != Event.MSG) { // unless it is a message handle it and respond
handleDownEvent(evt);
return;
}
// ****************** profiling ******************
/*if(num_msgs == 0) {
start=System.currentTimeMillis();
num_msgs++;
}
else if(num_msgs >= 1000) {
stop=System.currentTimeMillis();
long total_time=stop-start;
double msgs_per_msec=num_msgs / (double)total_time;
if(Trace.trace)
Trace.info("UDP.down.profile()",
"total_time=" + total_time + ", msgs/ms=" + msgs_per_msec);
num_msgs=0;
}
else {
num_msgs++;
}*/
// ****************** profiling ******************
msg = (Message) evt.getArg();
if (udp_hdr != null && udp_hdr.group_addr != null) {
// added patch by Roland Kurmann (March 20 2003)
msg.putHeader(name, udp_hdr);
}
dest_addr = msg.getDest();
// Because we don't call Protocol.passDown(), we notify the observer directly (e.g.
// PerfObserver).
// This way, we still have performance numbers for UDP
if (observer != null) {
observer.passDown(evt);
}
if (dest_addr == null) { // 'null' means send to all group members
if (ip_mcast) {
if (mcast_addr == null) {
Trace.error(
"UDP.down()",
"dest address of message is null, and "
+ "sending to default address fails as mcast_addr is null, too !"
+ " Discarding message "
+ Util.printEvent(evt));
return;
}
// if we want to use IP multicast, then set the destination of the message
msg.setDest(mcast_addr);
} else {
// sends a separate UDP message to each address
sendMultipleUdpMessages(msg, members);
return;
}
}
try {
sendUdpMessage(msg);
} catch (Exception e) {
Trace.error("UDP.down()", "exception=" + e + ", msg=" + msg + ", mcast_addr=" + mcast_addr);
}
}
/**
* 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());
}
}
}
}
public void sendJoinResponse(JoinRsp rsp, Address dest) {
Message m = new Message(dest, null, null);
GMS.GmsHeader hdr = new GMS.GmsHeader(GMS.GmsHeader.JOIN_RSP, rsp);
m.putHeader(this.id, hdr);
getDownProtocol().down(new Event(Event.MSG, m));
}
@SuppressWarnings("unchecked")
public Object up(Event evt) {
switch (evt.getType()) {
case Event.MSG:
Message msg = (Message) evt.getArg();
GmsHeader hdr = (GmsHeader) msg.getHeader(this.id);
if (hdr == null) break;
switch (hdr.type) {
case GmsHeader.JOIN_REQ:
view_handler.add(
new Request(Request.JOIN, hdr.mbr, false, null, hdr.useFlushIfPresent));
break;
case GmsHeader.JOIN_REQ_WITH_STATE_TRANSFER:
view_handler.add(
new Request(
Request.JOIN_WITH_STATE_TRANSFER, hdr.mbr, false, null, hdr.useFlushIfPresent));
break;
case GmsHeader.JOIN_RSP:
impl.handleJoinResponse(hdr.join_rsp);
break;
case GmsHeader.LEAVE_REQ:
if (hdr.mbr == null) return null;
view_handler.add(new Request(Request.LEAVE, hdr.mbr, false));
break;
case GmsHeader.LEAVE_RSP:
impl.handleLeaveResponse();
break;
case GmsHeader.VIEW:
View new_view = hdr.view;
if (new_view == null) return null;
Address coord = msg.getSrc();
if (!new_view.containsMember(coord)) {
sendViewAck(
coord); // we need to send the ack first, otherwise the connection is removed
impl.handleViewChange(new_view, hdr.my_digest);
} else {
impl.handleViewChange(new_view, hdr.my_digest);
sendViewAck(coord); // send VIEW_ACK to sender of view
}
break;
case GmsHeader.VIEW_ACK:
Address sender = msg.getSrc();
ack_collector.ack(sender);
return null; // don't pass further up
case GmsHeader.MERGE_REQ:
impl.handleMergeRequest(msg.getSrc(), hdr.merge_id, hdr.mbrs);
break;
case GmsHeader.MERGE_RSP:
MergeData merge_data =
new MergeData(msg.getSrc(), hdr.view, hdr.my_digest, hdr.merge_rejected);
if (log.isTraceEnabled()) {
log.trace(
local_addr
+ ": got merge response from "
+ msg.getSrc()
+ ", merge_id="
+ hdr.merge_id
+ ", merge data is "
+ merge_data);
}
impl.handleMergeResponse(merge_data, hdr.merge_id);
break;
case GmsHeader.INSTALL_MERGE_VIEW:
impl.handleMergeView(
new MergeData(msg.getSrc(), hdr.view, hdr.my_digest), hdr.merge_id);
break;
case GmsHeader.INSTALL_DIGEST:
Digest tmp = hdr.my_digest;
down_prot.down(new Event(Event.MERGE_DIGEST, tmp));
break;
case GmsHeader.INSTALL_MERGE_VIEW_OK:
// [JGRP-700] - FLUSH: flushing should span merge
merge_ack_collector.ack(msg.getSrc());
break;
case GmsHeader.CANCEL_MERGE:
// [JGRP-524] - FLUSH and merge: flush doesn't wrap entire merge process
impl.handleMergeCancelled(hdr.merge_id);
break;
case GmsHeader.GET_DIGEST_REQ:
// only handle this request if it was sent by the coordinator (or at least a member) of
// the current cluster
synchronized (members) {
if (!members.contains(msg.getSrc())) break;
}
// discard my own request:
if (msg.getSrc().equals(local_addr)) return null;
if (hdr.merge_id != null
&& !(merger.matchMergeId(hdr.merge_id) || merger.setMergeId(null, hdr.merge_id)))
return null;
// fetch only my own digest
Digest digest = (Digest) down_prot.down(new Event(Event.GET_DIGEST, local_addr));
if (digest != null) {
GmsHeader rsp_hdr = new GmsHeader(GmsHeader.GET_DIGEST_RSP);
rsp_hdr.my_digest = digest;
Message get_digest_rsp = new Message(msg.getSrc(), null, null);
get_digest_rsp.setFlag(Message.OOB);
get_digest_rsp.putHeader(this.id, rsp_hdr);
down_prot.down(new Event(Event.MSG, get_digest_rsp));
}
break;
case GmsHeader.GET_DIGEST_RSP:
Digest digest_rsp = hdr.my_digest;
impl.handleDigestResponse(msg.getSrc(), digest_rsp);
break;
default:
if (log.isErrorEnabled()) log.error("GmsHeader with type=" + hdr.type + " not known");
}
return null; // don't pass up
case Event.SUSPECT:
Object retval = up_prot.up(evt);
Address suspected = (Address) evt.getArg();
view_handler.add(new Request(Request.SUSPECT, suspected, true));
ack_collector.suspect(suspected);
merge_ack_collector.suspect(suspected);
return retval;
case Event.UNSUSPECT:
impl.unsuspect((Address) evt.getArg());
return null; // discard
case Event.MERGE:
view_handler.add(
new Request(Request.MERGE, null, false, (Map<Address, View>) evt.getArg()));
return null; // don't pass up
case Event.IS_MERGE_IN_PROGRESS:
return merger.isMergeInProgress();
}
return up_prot.up(evt);
}
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
}
