package org.jgroups.protocols.pbcast;
import org.jgroups.Address;
import org.jgroups.Event;
import org.jgroups.Message;
import org.jgroups.View;
import org.jgroups.protocols.PingData;
import org.jgroups.util.Digest;
import org.jgroups.util.Promise;
import org.jgroups.util.Responses;
import org.jgroups.util.Util;
import java.util.*;
/**
* Client part of GMS. Whenever a new member wants to join a group, it starts in the CLIENT role.
* No multicasts to the group will be received and processed until the member has been joined and
* turned into a SERVER (either coordinator or participant, mostly just participant). This class
* only implements <code>Join</code> (called by clients who want to join a certain group, and
* <code>ViewChange</code> which is called by the coordinator that was contacted by this client, to
* tell the client what its initial membership is.
* @author Bela Ban
* @version $Revision: 1.78 $
*/
public class ClientGmsImpl extends GmsImpl {
protected final Promise<JoinRsp> join_promise=new Promise<>();
public ClientGmsImpl(GMS g) {
super(g);
}
public void init() throws Exception {
super.init();
join_promise.reset();
}
public void join(Address address,boolean useFlushIfPresent) {
joinInternal(address, false,useFlushIfPresent);
}
public void joinWithStateTransfer(Address local_addr, boolean useFlushIfPresent) {
joinInternal(local_addr,true,useFlushIfPresent);
}
/**
* Joins this process to a group. Determines the coordinator and sends a
* unicast handleJoin() message to it. The coordinator returns a JoinRsp and
* then broadcasts the new view, which contains a message digest and the
* current membership (including the joiner). The joiner is then supposed to
* install the new view and the digest and starts accepting mcast messages.
* Previous mcast messages were discarded (this is done in PBCAST).
* <p>
* If successful, impl is changed to an instance of ParticipantGmsImpl.
* Otherwise, we continue trying to send join() messages to the coordinator,
* until we succeed (or there is no member in the group. In this case, we
* create our own singleton group).
*
* @param mbr Our own address (assigned through SET_LOCAL_ADDRESS)
*/
protected void joinInternal(Address mbr, boolean joinWithStateTransfer, boolean useFlushIfPresent) {
long join_attempts=0;
leaving=false;
join_promise.reset();
while(!leaving) {
if(installViewIfValidJoinRsp(join_promise, false))
return;
long start=System.currentTimeMillis();
Responses responses=(Responses)gms.getDownProtocol().down(Event.FIND_INITIAL_MBRS_EVT);
// Sept 2008 (bela): break if we got a belated JoinRsp (https://jira.jboss.org/jira/browse/JGRP-687)
if(installViewIfValidJoinRsp(join_promise, false))
return;
responses.waitFor(gms.join_timeout);
responses.done();
long diff=System.currentTimeMillis() - start;
if(responses.isEmpty()) {
log.trace("%s: no members discovered after %d ms: creating cluster as first member", gms.local_addr, diff);
becomeSingletonMember(mbr);
return;
}
log.trace("%s: discovery took %d ms, members: %s", gms.local_addr, diff, responses);
List<Address> coords=getCoords(responses);
// We didn't get any coord responses; all responses were clients. If I'm the first of the sorted clients
// I'll become coordinator. The others will wait and then retry the discovery and join process
if(coords == null) { // e.g. because we have all clients only
if(firstOfAllClients(mbr, responses))
return;
continue;
}
if(coords.size() > 1) {
log.debug("%s: found multiple coords: %s", gms.local_addr, coords);
Collections.shuffle(coords); // so the code below doesn't always pick the same coord
}
join_attempts++;
for(Address coord: coords) {
log.debug("%s: sending JOIN(%s) to %s", gms.local_addr, mbr, coord);
sendJoinMessage(coord, mbr, joinWithStateTransfer, useFlushIfPresent);
if(installViewIfValidJoinRsp(join_promise, true))
return;
log.warn("%s: JOIN(%s) sent to %s timed out (after %d ms), on try %d",
gms.local_addr, mbr, coord, gms.join_timeout, join_attempts);
}
if(gms.max_join_attempts != 0 && join_attempts >= gms.max_join_attempts) {
log.warn("%s: too many JOIN attempts (%d): becoming singleton", gms.local_addr, join_attempts);
becomeSingletonMember(mbr);
return;
}
}
}
public void leave(Address mbr) {
leaving=true;
wrongMethod("leave");
}
public void handleJoinResponse(JoinRsp join_rsp) {
join_promise.setResult(join_rsp); // will wake up join() method
}
/* --------------------------- Private Methods ------------------------------------ */
// Installs a new view and sends an ack if (1) the join rsp is not null, (2) it is valid and (3) the view
// installation is successful. If true is returned, the JOIN process can be terminated, else it needs to
// go through discovery and JOIN-REQ again in a next iteration
protected boolean installViewIfValidJoinRsp(final Promise<JoinRsp> join_promise, boolean block_for_rsp) {
boolean success=false;
JoinRsp rsp=null;
try {
if(join_promise.hasResult())
rsp=join_promise.getResult(1, true);
else if(block_for_rsp)
rsp=join_promise.getResult(gms.join_timeout, true);
return success=rsp != null && isJoinResponseValid(rsp) && installView(rsp.getView(), rsp.getDigest());
}
finally {
if(success)
sendViewAck(rsp.getView().getCreator());
}
}
/** Handles the case where no coord responses were received. Returns true if we became the coord
* (caller needs to terminate the join() call), or false when the caller needs to continue */
protected boolean firstOfAllClients(final Address joiner, final Responses rsps) {
log.trace("%s: could not determine coordinator from rsps %s", gms.local_addr, rsps);
// so the member to become singleton member (and thus coord) is the first of all clients
SortedSet<Address> clients=new TreeSet<>();
clients.add(joiner); // add myself again (was removed by findInitialMembers())
for(PingData response: rsps)
clients.add(response.getAddress());
log.trace("%s: nodes to choose new coord from are: %s", gms.local_addr, clients);
Address new_coord=clients.first();
if(new_coord.equals(joiner)) {
log.trace("%s: I (%s) am the first of the nodes, will become coordinator", gms.local_addr, joiner);
becomeSingletonMember(joiner);
return true;
}
log.trace("%s: I (%s) am not the first of the nodes, waiting for another client to become coordinator",
gms.local_addr, joiner);
Util.sleep(500);
return false;
}
protected boolean isJoinResponseValid(final JoinRsp rsp) {
if(rsp.getFailReason() != null)
throw new SecurityException(rsp.getFailReason());
Digest tmp_digest=rsp.getDigest();
if(tmp_digest == null || tmp_digest.capacity() == 0) {
log.warn("%s: digest is empty: digest=%s", gms.local_addr, rsp.getDigest());
return false;
}
if(!tmp_digest.contains(gms.local_addr)) {
log.error("%s: digest in JOIN_RSP does not contain myself; join response: %s", gms.local_addr, rsp);
return false;
}
if(rsp.getView() == null) {
log.error("%s: JoinRsp has a null view, skipping it", gms.local_addr);
return false;
}
return true;
}
/**
* Called by join(). Installs the view returned by calling Coord.handleJoin() and becomes coordinator.
*/
private boolean installView(View new_view, Digest digest) {
if(!new_view.containsMember(gms.local_addr)) {
log.error("%s: I'm not member of %s, will not install view", gms.local_addr, new_view);
return false;
}
gms.installView(new_view, digest);
if(gms.impl == null || gms.impl instanceof ClientGmsImpl) // installView() should have set the role (impl)
gms.becomeParticipant();
gms.getUpProtocol().up(new Event(Event.BECOME_SERVER));
gms.getDownProtocol().down(new Event(Event.BECOME_SERVER));
return true;
}
protected static String print(List<PingData> rsps) {
StringBuilder sb=new StringBuilder();
for(PingData rsp: rsps)
sb.append(rsp.getAddress() + " ");
return sb.toString();
}
protected static String print(Responses rsps) {
StringBuilder sb=new StringBuilder();
for(PingData rsp: rsps)
sb.append(rsp.getAddress() + " ");
return sb.toString();
}
void sendJoinMessage(Address coord, Address mbr,boolean joinWithTransfer, boolean useFlushIfPresent) {
byte type=joinWithTransfer? GMS.GmsHeader.JOIN_REQ_WITH_STATE_TRANSFER : GMS.GmsHeader.JOIN_REQ;
GMS.GmsHeader hdr=new GMS.GmsHeader(type, mbr, useFlushIfPresent);
Message msg=new Message(coord).setFlag(Message.Flag.OOB, Message.Flag.INTERNAL).putHeader(gms.getId(), hdr);
gms.getDownProtocol().down(msg);
}
void sendViewAck(Address coord) {
Message view_ack=new Message(coord).setFlag(Message.Flag.OOB, Message.Flag.INTERNAL)
.putHeader(gms.getId(), new GMS.GmsHeader(GMS.GmsHeader.VIEW_ACK));
gms.getDownProtocol().down(view_ack); // send VIEW_ACK to sender of view
}
/** Returns all members whose PingData is flagged as coordinator */
private static List<Address> getCoords(Iterable<PingData> mbrs) {
if(mbrs == null)
return null;
List<Address> coords=null;
for(PingData mbr: mbrs) {
if(mbr.isCoord()) {
if(coords == null)
coords=new ArrayList<>();
if(!coords.contains(mbr.getAddress()))
coords.add(mbr.getAddress());
}
}
return coords;
}
void becomeSingletonMember(Address mbr) {
View new_view=View.create(mbr, 0, mbr); // create singleton view with mbr as only member
// set the initial digest (since I'm the first member)
Digest initial_digest=new Digest(mbr, 0, 0);
gms.installView(new_view, initial_digest);
gms.becomeCoordinator(); // not really necessary - installView() should do it
gms.getUpProtocol().up(new Event(Event.BECOME_SERVER));
gms.getDownProtocol().down(new Event(Event.BECOME_SERVER));
log.debug("%s: created cluster (first member). My view is %s, impl is %s",
gms.getLocalAddress(), gms.getViewId(), gms.getImpl().getClass().getName());
}
}