public void testFlushWithCrashedParticipants() throws Exception {
JChannel c1 = null;
JChannel c2 = null;
JChannel c3 = null;
try {
c1 = createChannel(true, 3, "C1");
changeProps(c1);
c1.connect("testFlushWithCrashedFlushCoordinator");
c2 = createChannel(c1, "C2");
changeProps(c2);
c2.connect("testFlushWithCrashedFlushCoordinator");
c3 = createChannel(c1, "C3");
changeProps(c3);
c3.connect("testFlushWithCrashedFlushCoordinator");
// and then kill members other than flush coordinator
Util.shutdown(c3);
Util.shutdown(c1);
// start flush
Util.startFlush(c2);
c2.stopFlush();
Util.waitUntilAllChannelsHaveSameSize(10000, 500, c2);
// cluster should not hang and one remaining member should have a correct view
assertTrue("correct view size", c2.getView().size() == 1);
} finally {
Util.close(c3, c2, c1);
}
}
/**
* Connects this channel to a group and gets a state from a specified state provider.
*
* <p>This method invokes {@code connect()} and then {@code getState}.
*
* <p>If the FLUSH protocol is in the channel's stack definition, only one flush round is executed
* for both connecting and fetching the state rather than two flushes if we invoke {@code connect}
* and {@code getState} in succession.
*
* <p>If the channel is already connected, an error message will be printed to the error log. If
* the channel is closed a ChannelClosed exception will be thrown.
*
* @param cluster_name The cluster name to connect to. Cannot be null.
* @param target The state provider. If null, the state will be fetched from the coordinator,
* unless this channel is the coordinator.
* @param timeout The timeout for the state transfer.
* @exception Exception The protocol stack cannot be started, or the JOIN failed
* @exception IllegalStateException The channel is closed or disconnected
* @exception StateTransferException State transfer was not successful
*/
public synchronized void connect(
String cluster_name, Address target, long timeout, boolean useFlushIfPresent)
throws Exception {
if (!_preConnect(cluster_name)) return;
if (cluster_name == null) { // only connect if we are not a unicast channel
state = State.CONNECTED;
return;
}
boolean canFetchState = false;
try {
Event connect_event =
useFlushIfPresent
? new Event(Event.CONNECT_WITH_STATE_TRANSFER_USE_FLUSH, cluster_name)
: new Event(Event.CONNECT_WITH_STATE_TRANSFER, cluster_name);
_connect(connect_event);
state = State.CONNECTED;
notifyChannelConnected(this);
canFetchState = getView() != null && getView().size() > 1;
// if I am not the only member in cluster then
if (canFetchState) getState(target, timeout, false); // fetch state from target
} finally {
// stopFlush if we fetched the state or failed to connect...
if ((flushSupported() && useFlushIfPresent) && (canFetchState || state != State.CONNECTED))
stopFlush();
}
}
public void startFlush(boolean automatic_resume) throws Exception {
if (!flushSupported())
throw new IllegalStateException(
"Flush is not supported, add pbcast.FLUSH protocol to your configuration");
try {
down(new Event(Event.SUSPEND));
} catch (Exception e) {
throw new Exception("Flush failed", e.getCause());
} finally {
if (automatic_resume) stopFlush();
}
}
protected void getState(Address target, long timeout, Callable<Boolean> flushInvoker)
throws Exception {
checkClosedOrNotConnected();
if (!state_transfer_supported)
throw new IllegalStateException(
"fetching state will fail as state transfer is not supported. "
+ "Add one of the state transfer protocols to your configuration");
if (target == null) target = determineCoordinator();
if (target != null && local_addr != null && target.equals(local_addr)) {
log.trace(
local_addr
+ ": cannot get state from myself ("
+ target
+ "): probably the first member");
return;
}
boolean initiateFlush = flushSupported() && flushInvoker != null;
if (initiateFlush) {
boolean successfulFlush = false;
try {
successfulFlush = flushInvoker.call();
} catch (Throwable e) {
successfulFlush = false; // http://jira.jboss.com/jira/browse/JGRP-759
}
if (!successfulFlush)
throw new IllegalStateException(
"Node " + local_addr + " could not flush the cluster for state retrieval");
}
state_promise.reset();
StateTransferInfo state_info = new StateTransferInfo(target, timeout);
long start = System.currentTimeMillis();
down(new Event(Event.GET_STATE, state_info));
StateTransferResult result = state_promise.getResult(state_info.timeout);
if (initiateFlush) stopFlush();
if (result == null)
throw new StateTransferException(
"timeout during state transfer (" + (System.currentTimeMillis() - start) + "ms)");
if (result.hasException())
throw new StateTransferException("state transfer failed", result.getException());
}
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);
}
}
public void testFlushWithCrashedParticipant() throws Exception {
JChannel c1 = null;
JChannel c2 = null;
JChannel c3 = null;
try {
c1 = createChannel(true, 3, "C1");
changeProps(c1);
c1.connect("testFlushWithCrashedParticipant");
c2 = createChannel(c1, "C2");
changeProps(c2);
c2.connect("testFlushWithCrashedParticipant");
c3 = createChannel(c1, "C3");
changeProps(c3);
c3.connect("testFlushWithCrashedParticipant");
System.out.println("shutting down C3");
Util.shutdown(c3); // kill a flush participant
System.out.println("C2: starting flush");
boolean rc = Util.startFlush(c2);
System.out.println("flush " + (rc ? " was successful" : "failed"));
assert rc;
System.out.println("stopping flush");
c2.stopFlush();
System.out.println("waiting for view to contain C1 and C2");
Util.waitUntilAllChannelsHaveSameSize(10000, 500, c1, c2);
// cluster should not hang and two remaining members should have a correct view
System.out.println("C1: view=" + c1.getView() + "\nC2: view=" + c2.getView());
assertTrue("correct view size", c1.getView().size() == 2);
assertTrue("correct view size", c2.getView().size() == 2);
} finally {
Util.close(c3, c2, c1);
}
}
/** Tests http://jira.jboss.com/jira/browse/JGRP-661 */
public void testPartialFlush() throws Exception {
JChannel c1 = null;
JChannel c2 = null;
try {
c1 = createChannel(true, 2);
c1.setReceiver(new SimpleReplier(c1, true));
c1.connect("testPartialFlush");
c2 = createChannel(c1);
c2.setReceiver(new SimpleReplier(c2, false));
c2.connect("testPartialFlush");
List<Address> members = new ArrayList<Address>();
members.add(c2.getAddress());
boolean flushedOk = Util.startFlush(c2, members);
assertTrue("Partial flush worked", flushedOk);
c2.stopFlush(members);
} finally {
Util.close(c2, c1);
}
}
