@Override // GemStoneAddition
public void init() throws Exception {
super.init();
synchronized (initial_mbrs) {
initial_mbrs.clear();
initial_mbrs_received = false;
}
join_promise.reset();
}
@Override // GemStoneAddition
public void handleJoinResponse(JoinRsp join_rsp) {
if (log.isDebugEnabled()) {
log.getLogWriter().info(ExternalStrings.DEBUG, "Received join response " + join_rsp);
}
while (SLOW_JOIN) { // test hook
log.getLogWriter().info(ExternalStrings.DEBUG, "delaying delivery of join response");
try {
Thread.sleep(5000);
} catch (InterruptedException e) {
return;
}
}
join_promise.setResult(join_rsp); // will wake up join() method
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "Set join response in promise");
}
/**
* 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).
*
* <p>When GMS.disable_initial_coord is set to true, then we won't become coordinator on receiving
* an initial membership of 0, but instead will retry (forever) until we get an initial membership
* of > 0.
*
* @param myAddr Our own address (assigned through SET_LOCAL_ADDRESS)
*/
@Override // GemStoneAddition
public boolean join(Address myAddr) {
Address coord = null;
Address coordSentJoin = null;
JoinRsp rsp;
Digest tmp_digest;
View tmp_view = null;
leaving = false;
long starttime = System.currentTimeMillis(); // GemStoneAddition
final int maxRetries = 6; // GemStoneAddition - need this figure later
int joinRetries = maxRetries;
join_promise.reset();
boolean debugFailFirst = Boolean.getBoolean("p2p.DEBUG_FAIL_FIRST");
while (!leaving) {
joinRetries--;
if (joinRetries < 0) {
// GemStoneAddition
// if not just anyone can be the coordinator and the elected \
// coordinator cannot process a join_req, assume it is
// gone and elect self as new coordinator
if (coord != null
&& gms.floatingCoordinatorDisabled
&& !gms.splitBrainDetectionEnabled // bug #41772 I DON'T THINK THIS IS TRUE WITH QUORUM
&& myAddr.preferredForCoordinator()) {
Address failedCoord = coord;
if (!myAddr.equals(failedCoord)) {
gms.log
.getLogWriter()
.warning(
ExternalStrings.ClientGmsImpl_COULD_NOT_JOIN_DISTRIBUTED_SYSTEM_0, failedCoord);
if (becomeGroupCoordinator(myAddr, failedCoord)) {
return true;
}
}
coord = null;
} else {
if (!join_promise
.hasResult()) { // GemStoneAddition - check one more time since we just waited for the
// retry
log.warn(
"Could not join distributed system using member "
+ coord
+ ". gms.floatingCoordinatorDisabled="
+ gms.floatingCoordinatorDisabled
+ ", mbr.canBeCoordinator()="
+ myAddr.preferredForCoordinator());
return false;
}
}
}
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "findInitialMembers");
missing_mbrs = null;
findInitialMembers();
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "findInitialMembers done");
if (debugFailFirst) {
log.getLogWriter()
.severe(ExternalStrings.ClientGmsImpl_IGNORING_FIRST_MEMBERSHIP_SET_FOR_DEBUGGING);
debugFailFirst = false;
continue;
}
if (!join_promise.hasResult()) {
if (log.isDebugEnabled())
log.debug(
"CGMS: initial_mbrs are " + initial_mbrs + "; missing_mbrs are " + missing_mbrs);
if (initial_mbrs.size() == 0) {
if (missing_mbrs == null || missing_mbrs.isEmpty()) { // GemStoneAddition
if (gms.disable_initial_coord) {
// GemStoneAddition - there was no timeout if disable_ic was set!!
checkForTimeout(starttime);
// if(trace)
log.warn(
"received an initial membership of 0, but cannot become coordinator "
+ // GemStoneAddition - was trace level output. Debug #34274
"(disable_initial_coord=true), will retry fetching the initial membership");
// continue;
}
// if we're getting no responses at all after two or more attempts, there's no-one
// out there.
else if (joinRetries <= (maxRetries - 2)) {
if (log.isDebugEnabled())
log.debug("no initial members discovered: creating group as first member");
becomeSingletonMember(myAddr);
return true;
}
}
if (joinRetries > 0) {
try { // GemStoneAddition
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "join sleep #1");
Util.sleep(gms.join_retry_timeout);
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "join sleep #1 done");
} catch (InterruptedException e) {
if (log.isDebugEnabled()) log.debug("interrupted; creating group as first member");
becomeSingletonMember(myAddr);
Thread.currentThread().interrupt(); // GemStoneAddition
return true;
}
}
continue;
}
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "determining coordinator");
coord = determineCoord(initial_mbrs);
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "done determining coordinator: " +
// coord);
// GemStoneAddition - don't elect a new coordinator on the first attempt
if (coord == null) {
if (joinRetries < maxRetries - 1) {
if (gms.handle_concurrent_startup == false) {
if (trace)
log.trace(
"handle_concurrent_startup is false; ignoring responses of initial clients");
becomeSingletonMember(myAddr);
return true;
}
if (trace) log.trace("could not determine coordinator from responses " + initial_mbrs);
// the member to become singleton member (and thus coord) is the first of all clients
Set clients = new TreeSet(); // sorted
clients.add(myAddr); // add myself again (was removed by findInitialMembers())
for (int i = 0; i < initial_mbrs.size(); i++) {
PingRsp pingRsp = (PingRsp) initial_mbrs.elementAt(i);
Address client_addr =
pingRsp.getCoordAddress(); // GemStoneAddition - was .getAddress();
if (client_addr != null) clients.add(client_addr);
}
if (trace)
log.trace(
"Choosing coordinator from these servers: "
+ clients
+ ".\nJoin retries remaining: "
+ joinRetries);
Address new_coord = (Address) clients.iterator().next();
if (new_coord.equals(myAddr)) {
if (!gms.disable_initial_coord) { // GemStoneAddition - respect disable_initial_coord
if (trace)
log.trace(
"I ("
+ myAddr
+ ") am the first of the clients, and will become coordinator");
becomeSingletonMember(myAddr);
// GemStoneAddition - tell the DM to wait a while
// if (gms.stack.jgmm != null) {
// long waitPeriod = Long.getLong("p2p.concStartupDelay",
// gms.join_timeout).longValue();
// if (trace)
// log.trace("Telling the manager to wait for " + waitPeriod + " ms");
// gms.stack.jgmm.establishChannelPause(waitPeriod);
// //try { Thread.sleep(waitPeriod); } catch (InterruptedException ie) {
// Thread.currentThread().interrupt(); }
// }
return true;
}
} else {
// if(trace)
// log.trace("I (" + myAddr + ") am not the first of the clients,
// " +
// "waiting for another client to become coordinator");
log.getLogWriter()
.info(
ExternalStrings.DEBUG,
"Waiting for "
+ new_coord
+ " to possibly become coordinator during concurrent startup. "
+ "My address is "
+ gms.local_addr);
// GemStoneAddition - if there is concurrent startup, the other
// client will be waiting the same amount of time, so increase the
// join attempts in this VM to let the other client finish
if (joinRetries < 2) {
checkForTimeout(starttime);
joinRetries++;
}
}
} // joinRetries < maxRetries-1
else {
if (trace)
log.trace(
"I ("
+ myAddr
+ ") am not the coordinator and it is too"
+ " soon for an election");
}
if (joinRetries > 0) {
try { // GemStoneAddition
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "join sleep #2");
Util.sleep(gms.join_retry_timeout);
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "join sleep #2 done");
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
return false; // treat as failure
}
}
continue;
} // coord==null
if (!coord.splitBrainEnabled() && this.gms.splitBrainDetectionEnabled) {
throw gms.stack.gfBasicFunctions.getGemFireConfigException(
"Group membership coordinator, "
+ coord
+ " does not have "
+ "network partition detection enabled, but this DistributedSystem has it enabled");
}
boolean c6 = (((IpAddress) coord).getIpAddress() instanceof Inet6Address);
boolean m6 = (((IpAddress) myAddr).getIpAddress() instanceof Inet6Address);
if (c6 && !m6) {
throw gms.stack.gfBasicFunctions.getGemFireConfigException(
"Group membership coordinator, "
+ coord
+ ", is using IPv6 but this "
+ "process is using IPv4");
}
if (m6 && !c6) {
throw gms.stack.gfBasicFunctions.getGemFireConfigException(
"Group membership coordinator, "
+ coord
+ ", is using IPv4 but this "
+ "process is using IPv6");
}
} // !join_promise.hasResult()
else {
coord = coordSentJoin;
}
if (coord == null) {
// GemStoneAddition
// fix for NPE when a false join response has been received. This could
// be from the retransmission of a response sent to a previous member
// using the same address. In this case we haven't even gotten to
// the point of sending a join request and the response should be ignored
join_promise.reset();
continue;
}
try {
if (coord.equals(gms.local_addr) && gms.local_addr.preferredForCoordinator()) {
// GemStoneAddition - we can't usurp a group that was using a coordinator
// that had the same address that this process is now using (bug #41772)
if (((IpAddress) coord).getUniqueID() != ((IpAddress) gms.local_addr).getUniqueID()) {
throw new ShunnedAddressException();
}
// this member was elected by others as the coordinator
if (becomeGroupCoordinator(gms.local_addr, null)) {
return true;
} else {
coord = null;
continue; // no need to sleep - there's another coordinator now
}
}
// GemStoneAddition - check to see if a result came in while get_mbrs was being processed.
// If there is one, just use it.
if (!join_promise.hasResult()) {
if (log.isDebugEnabled()) log.debug("sending join_req(" + myAddr + ") to " + coord);
coordSentJoin = coord;
sendJoinMessage(coord, myAddr);
}
if (log.isDebugEnabled()) {
log.debug("Waiting for join response");
}
// test hook for bug #49897
if (SLOW_JOIN) {
synchronized (SLOW_JOIN_LOCK) {
try {
SLOW_JOIN_LOCK.wait();
} catch (InterruptedException e) {
// the test will interrupt. Set the interrupt flag
// so the Promise.getResult() call will run into it
Thread.currentThread().interrupt();
}
}
}
rsp = (JoinRsp) join_promise.getResult(gms.join_timeout);
// bug #49897 - don't keep going if the thread's been interrupted
if (Thread.currentThread().isInterrupted()) {
throw gms.stack.gfBasicFunctions.getSystemConnectException(
ExternalStrings.ClientGmsImpl_JOIN_INTERRUPTED.toLocalizedString());
}
if (rsp == null) {
if (warn && !leaving /* GemStoneAddition */)
log.getLogWriter()
.warning(
ExternalStrings.ClientGmsImpl_JOIN_0__SENT_TO__1__TIMED_OUT_RETRYING,
new Object[] {myAddr, coord});
} else {
if (log.isDebugEnabled()) {
log.debug("Got join response: " + rsp);
}
// GemStoneAddition - birth view ID
if (rsp.view != null && rsp.view.getVid() != null) {
((IpAddress) gms.local_addr).setBirthViewId(rsp.view.getVid().getId());
}
if (rsp.getFailReason() != null) {
if (rsp.getFailReason().equals(JoinRsp.SHUNNED_ADDRESS)) {
throw new ShunnedAddressException();
}
if (rsp.getFailReason()
.contains("Rejecting the attempt of a member using an older version")
|| rsp.getFailReason().contains("15806")) {
throw gms.stack.gfBasicFunctions.getSystemConnectException(rsp.getFailReason());
}
throw gms.stack.gfBasicFunctions.getAuthenticationFailedException(rsp.getFailReason());
}
coord = rsp.getView().getCreator();
gms.notifyOfCoordinator(coord);
// 1. Install digest
tmp_digest = rsp.getDigest();
tmp_view = rsp.getView();
if (tmp_digest == null || tmp_view == null) {
if (log.isErrorEnabled())
log.error(ExternalStrings.ClientGmsImpl_DIGEST_OR_VIEW_OF_JOIN_RESPONSE_IS_NULL);
} else {
// GemStoneAddition - commented out inc of highSeqno because this
// does not work if the view is not multicast. In view of this there
// must be a retransmission protocol.
// tmp_digest.incrementHighSeqno(coord); // see DESIGN for an
// explanantion
if (log.isDebugEnabled()) log.debug("digest is " + tmp_digest);
gms.setDigest(tmp_digest);
// 2. Install view
if (log.isDebugEnabled())
log.debug(
"["
+ gms.local_addr
+ "]: JoinRsp="
+ rsp.getView()
+ " [size="
+ rsp.getView().size()
+ "]\n\n");
if (!installView(tmp_view)) {
if (log.isErrorEnabled())
log.error(
ExternalStrings.ClientGmsImpl_VIEW_INSTALLATION_FAILED_RETRYING_TO_JOIN_GROUP);
continue;
}
// send VIEW_ACK to sender of view
Message view_ack = new Message(coord, null, null);
GMS.GmsHeader tmphdr = new GMS.GmsHeader(GMS.GmsHeader.VIEW_ACK);
view_ack.putHeader(GMS.name, tmphdr);
// GemStoneAddition: bug #49261 - do not send the view
// so that the ack-collector will accept this as an ack
// for the view about to be cast by the coordinator
view_ack.setObject(null);
gms.passDown(new Event(Event.MSG, view_ack));
gms.passUp(new Event(Event.BECOME_SERVER));
gms.passDown(new Event(Event.BECOME_SERVER));
return true;
}
}
} catch (ShunnedAddressException e) {
System.setProperty("gemfire.jg-bind-port", "0"); // for testing
throw e;
} catch (RuntimeException e) {
String name = e.getClass().getSimpleName();
if (name.equals("AuthenticationFailedException")) {
if (log.isDebugEnabled()) {
log.debug("AuthenticationFailedException=" + e.toString());
}
throw e;
} else if (name.equals("SystemConnectException")) {
throw e;
}
if (log.getLogWriter().infoEnabled()) {
log.getLogWriter().info(ExternalStrings.UnexpectedException, e);
}
} catch (Exception e) {
if (log.getLogWriter().infoEnabled()) {
log.getLogWriter().info(ExternalStrings.UnexpectedException, e);
}
}
if (joinRetries > 0) {
try { // GemStoneAddition
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "join sleep #3");
Util.sleep(gms.join_retry_timeout);
// log.getLogWriterI18n().info(JGroupsStrings.DEBUG, "join sleep #3 done");
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
return false; // treat as failure
}
}
}
if (missing_mbrs != null && missing_mbrs.size() > 0) {
log.getLogWriter().info(ExternalStrings.DEBUG, "missing ping responses from " + missing_mbrs);
return false;
}
return true;
}