use of com.nokia.dempsy.cluster.ClusterInfoSession in project Dempsy by Dempsy.
the class TestDempsy method testExpandingAndContractingKeySpace.
@Test
public void testExpandingAndContractingKeySpace() throws Throwable {
Checker checker = new Checker() {
@Override
public void check(ApplicationContext context) throws Throwable {
ClusterInfoSession session = null;
try {
// start things and verify that the init method was called
Dempsy dempsy = (Dempsy) context.getBean("dempsy");
TestMp mp = (TestMp) getMp(dempsy, "test-app", "test-cluster1");
final ClusterId clusterId = new ClusterId("test-app", "test-cluster1");
// verify we haven't called it again, not that there's really
// a way to given the code
assertEquals(1, mp.startCalls.get());
// make sure that there are no Mps
MetricGetters statsCollector = (MetricGetters) dempsy.getCluster(new ClusterId("test-app", "test-cluster1")).getNodes().get(0).getStatsCollector();
// This will wait until the keySpace is up to the maxcount which is set (in the setup, below) to 100000
assertTrue(poll(baseTimeoutMillis, statsCollector, new Condition<MetricGetters>() {
@Override
public boolean conditionMet(MetricGetters sc) {
return 100000 == sc.getMessageProcessorCount();
}
}));
// now push the cluster into backup node.
ClusterInfoSession originalSession = dempsy.getCluster(new ClusterId("test-app", "test-cluster1")).getNodes().get(0).retouRteg().getClusterSession();
ClusterInfoSessionFactory factory = dempsy.getClusterSessionFactory();
session = TestUtils.stealShard(originalSession, factory, clusterId.asPath() + "/" + String.valueOf(0), baseTimeoutMillis);
// If we got here then the MpContainer is on standby and the number of Mps should
// drop to zero.
assertTrue(poll(baseTimeoutMillis, statsCollector, new Condition<MetricGetters>() {
@Override
public boolean conditionMet(MetricGetters sc) {
return 0 == sc.getMessageProcessorCount();
}
}));
Thread.sleep(10);
assertEquals(0, statsCollector.getMessageProcessorCount());
// this should give control back over to the original session.
session.stop();
session = null;
// If we got here then the MpContainer is no longer on standby and the number of Mps should
// go back to the original amount.
assertTrue(poll(baseTimeoutMillis, statsCollector, new Condition<MetricGetters>() {
@Override
public boolean conditionMet(MetricGetters sc) {
return 100000 == sc.getMessageProcessorCount();
}
}));
Thread.sleep(10);
assertEquals(100000, statsCollector.getMessageProcessorCount());
} finally {
if (session != null)
session.stop();
}
}
public String toString() {
return "testExpandingAndContractingKeySpace";
}
public void setup() {
KeySourceImpl.maxcount = 100000;
System.setProperty("min_nodes_for_cluster", "1");
System.setProperty("total_slots_for_cluster", "1");
}
};
runAllCombinations("SinglestageWithKeyStoreAndExecutorApplicationActx.xml", checker);
}
use of com.nokia.dempsy.cluster.ClusterInfoSession in project Dempsy by Dempsy.
the class TestUtils method stealShard.
/**
* This method will grab the slot requested. It requires that it is already held by
* the session provided and that the entry there contains a valid DefaultRouterSlotInfo
* which it will extract, modify and use to replace.
*
* This will be accomplished by disrupting the session and trying to grab the slot
* at the same time. It will try this over and over until it gets it, or until the
* number of tries is exceeded.
*
* @param originalSession is the session that will be disrupted in order to grab the shard.
* @param factory is the {@link ClusterInfoSessionFactory} that will be used to create a new
* session that can be used to grab the slot.
* @param shardPath is the path all the way to the directory containing the shard that you
* want stolen.
*
* @throws Assert when one of the test condition fails or grabbing the slot fails.
*/
public static ClusterInfoSession stealShard(final ClusterInfoSession originalSession, final ClusterInfoSessionFactory factory, final String shardPath, final long timeoutmillis) throws InterruptedException, ClusterInfoException {
// get the current slot data to use as a template
final DefaultRouterSlotInfo newSlot = (DefaultRouterSlotInfo) originalSession.getData(shardPath, null);
final AtomicBoolean stillRunning = new AtomicBoolean(true);
final AtomicBoolean failed = new AtomicBoolean(false);
final ClusterInfoSession session = factory.createSession();
Runnable slotGrabber = new Runnable() {
@Override
public void run() {
try {
Thread.currentThread().setPriority(Thread.MAX_PRIORITY);
boolean haveSlot = false;
while (!haveSlot && stillRunning.get()) {
newSlot.setDestination(new JunkDestination());
if (session.mkdir(shardPath, newSlot, DirMode.EPHEMERAL) != null)
haveSlot = true;
Thread.yield();
}
} catch (ClusterInfoException e) {
failed.set(true);
} catch (RuntimeException re) {
re.printStackTrace();
failed.set(true);
} finally {
stillRunning.set(false);
}
}
};
try {
new Thread(slotGrabber).start();
boolean onStandby = false;
long startTime = System.currentTimeMillis();
while (!onStandby && timeoutmillis >= (System.currentTimeMillis() - startTime)) {
((DisruptibleSession) originalSession).disrupt();
Thread.sleep(100);
if (!stillRunning.get())
onStandby = true;
}
assertTrue(onStandby);
assertFalse(failed.get());
} catch (InterruptedException ie) {
session.stop();
throw ie;
} catch (Error cie) {
session.stop();
throw cie;
} finally {
stillRunning.set(false);
}
return session;
}
use of com.nokia.dempsy.cluster.ClusterInfoSession in project Dempsy by Dempsy.
the class TestZookeeperClusterResilience method testSessionExpiredWithFullApp.
@Test
public void testSessionExpiredWithFullApp() throws Throwable {
// now lets startup the server.
ZookeeperTestServer server = null;
final AtomicReference<ZookeeperSession> sessionRef = new AtomicReference<ZookeeperSession>();
ZookeeperSession session = null;
final AtomicLong processCount = new AtomicLong(0);
Dempsy[] dempsy = new Dempsy[3];
try {
server = new ZookeeperTestServer();
server.start();
session = new ZookeeperSession("127.0.0.1:" + port, 5000) {
@Override
public WatcherProxy makeWatcherProxy(ClusterInfoWatcher w) {
processCount.incrementAndGet();
return super.makeWatcherProxy(w);
}
;
};
sessionRef.set(session);
final FullApplication app = new FullApplication();
ApplicationDefinition ad = app.getTopology();
// no calls yet
assertEquals(0, processCount.intValue());
dempsy[0] = getDempsyFor(new ClusterId(FullApplication.class.getSimpleName(), FullApplication.MyAdaptor.class.getSimpleName()), ad);
dempsy[0].setClusterSessionFactory(new ZookeeperSessionFactory("127.0.0.1:" + port, 5000));
dempsy[1] = getDempsyFor(new ClusterId(FullApplication.class.getSimpleName(), FullApplication.MyMp.class.getSimpleName()), ad);
dempsy[1].setClusterSessionFactory(new ZookeeperSessionFactory("127.0.0.1:" + port, 5000));
dempsy[2] = getDempsyFor(new ClusterId(FullApplication.class.getSimpleName(), FullApplication.MyRankMp.class.getSimpleName()), ad);
// dempsy[2].setClusterSessionFactory(new ZookeeperSessionFactory<ClusterInformation, SlotInformation>("127.0.0.1:" + port,5000));
dempsy[2].setClusterSessionFactory(new ClusterInfoSessionFactory() {
@Override
public ClusterInfoSession createSession() throws ClusterInfoException {
return sessionRef.get();
}
});
// start everything in reverse order
for (int i = 2; i >= 0; i--) dempsy[i].start();
// make sure the final count is incrementing
long curCount = app.finalMessageCount.get();
assertTrue(poll(30000, curCount, new Condition<Long>() {
@Override
public boolean conditionMet(Long o) {
return app.finalMessageCount.get() > (o + 100L);
}
}));
logger.trace("Killing zookeeper");
ZooKeeper origZk = session.zkref.get();
ZookeeperTestServer.forceSessionExpiration(origZk);
logger.trace("Killed zookeeper");
// wait for the current session to go invalid
assertTrue(poll(baseTimeoutMillis, origZk, new Condition<ZooKeeper>() {
@Override
public boolean conditionMet(ZooKeeper o) {
return !o.getState().isAlive();
}
}));
// make sure the final count is STILL incrementing
curCount = app.finalMessageCount.get();
assertTrue(poll(30000, curCount, new Condition<Long>() {
@Override
public boolean conditionMet(Long o) {
return app.finalMessageCount.get() > (o + 100L);
}
}));
} finally {
if (server != null)
server.shutdown();
if (session != null)
session.stop();
for (int i = 0; i < dempsy.length; i++) if (dempsy[i] != null)
dempsy[i].stop();
for (int i = 0; i < dempsy.length; i++) if (dempsy[i] != null)
assertTrue(dempsy[i].waitToBeStopped(baseTimeoutMillis));
}
}
use of com.nokia.dempsy.cluster.ClusterInfoSession in project Dempsy by Dempsy.
the class TestDempsy method testFailedClusterManagerDuringKeyStoreCalls.
@Test
public void testFailedClusterManagerDuringKeyStoreCalls() throws Throwable {
Checker checker = new Checker() {
@Override
public void check(ApplicationContext context) throws Throwable {
ClusterInfoSession session = null;
try {
// start things and verify that the init method was called
Dempsy dempsy = (Dempsy) context.getBean("dempsy");
TestMp mp = (TestMp) getMp(dempsy, "test-app", "test-cluster1");
final ClusterId clusterId = new ClusterId("test-app", "test-cluster1");
// verify we haven't called it again, not that there's really
// a way to given the code
assertEquals(1, mp.startCalls.get());
// make sure that there are no Mps
MetricGetters statsCollector = (MetricGetters) dempsy.getCluster(new ClusterId("test-app", "test-cluster1")).getNodes().get(0).getStatsCollector();
assertTrue(poll(baseTimeoutMillis, statsCollector, new Condition<MetricGetters>() {
@Override
public boolean conditionMet(MetricGetters sc) {
return 100000 == sc.getMessageProcessorCount();
}
}));
// now there's 100000 mps in the container created from the KeySource. So we steal the
// shard and force if offline but continuously disrupt it while it tries to come
// back up.
// now push the cluster into backup node.
ClusterInfoSession originalSession = dempsy.getCluster(new ClusterId("test-app", "test-cluster1")).getNodes().get(0).retouRteg().getClusterSession();
ClusterInfoSessionFactory factory = dempsy.getClusterSessionFactory();
String path = clusterId.asPath() + "/" + String.valueOf(0);
session = TestUtils.stealShard(originalSession, factory, path, baseTimeoutMillis);
DefaultRouterSlotInfo si = (DefaultRouterSlotInfo) session.getData(path, null);
// checks to see who actually has the slot.
assertTrue(si.getDestination() instanceof JunkDestination);
// we will keep disrupting the session but we should still end up with zero mps.
for (int i = 0; i < 100; i++) {
((DisruptibleSession) originalSession).disrupt();
Thread.sleep(1);
}
// now wait until we get to zero.
assertTrue(poll(baseTimeoutMillis, statsCollector, new Condition<MetricGetters>() {
@Override
public boolean conditionMet(MetricGetters sc) {
return 0 == sc.getMessageProcessorCount();
}
}));
Thread.sleep(10);
assertEquals(0, statsCollector.getMessageProcessorCount());
// ok. Now we will close the session that's holding the shard and allow the container
// to re-establish control of that shard. During the KeyStore reinstantiation of the
// MPs we will be disrupting the session.
session.stop();
for (int i = 0; i < 100; i++) {
((DisruptibleSession) originalSession).disrupt();
Thread.sleep(1);
}
// Now we should get back to 100000 Mps.
poll(baseTimeoutMillis, statsCollector, new Condition<MetricGetters>() {
@Override
public boolean conditionMet(MetricGetters sc) {
return 100000 == sc.getMessageProcessorCount();
}
});
assertEquals(100000, statsCollector.getMessageProcessorCount());
} finally {
if (session != null)
session.stop();
}
}
public String toString() {
return "testFailedClusterManagerDuringKeyStoreCalls";
}
public void setup() {
KeySourceImpl.maxcount = 100000;
System.setProperty("min_nodes_for_cluster", "1");
System.setProperty("total_slots_for_cluster", "1");
}
};
runAllCombinations("SinglestageWithKeyStoreAndExecutorApplicationActx.xml", checker);
}
use of com.nokia.dempsy.cluster.ClusterInfoSession in project Dempsy by Dempsy.
the class TestFullApp method testStartForceMpDisconnectStop.
@Test
public void testStartForceMpDisconnectStop() throws Throwable {
ClassPathXmlApplicationContext actx = null;
Dempsy dempsy = null;
try {
logger.debug("Starting up the appliction context ...");
actx = new ClassPathXmlApplicationContext(ctx);
actx.registerShutdownHook();
final FullApplication app = (FullApplication) actx.getBean("app");
dempsy = (Dempsy) actx.getBean("dempsy");
// Override the cluster session factory to keep track of the sessions asked for.
// This is so that I can grab the ZookeeperSession that's being instantiated by
// the MyMp cluster.
zookeeperCluster = null;
dempsy.setClusterSessionFactory(new ZookeeperSessionFactory(System.getProperty("zk_connect"), 5000) {
int sessionCount = 0;
@Override
public synchronized ClusterInfoSession createSession() throws ClusterInfoException {
sessionCount++;
ClusterInfoSession ret = super.createSession();
if (sessionCount == 2)
zookeeperCluster = (ZookeeperSession) ret;
return ret;
}
});
dempsy.start();
Dempsy.Application.Cluster cluster = dempsy.getCluster(new ClusterId(FullApplication.class.getSimpleName(), MyAdaptor.class.getSimpleName()));
Dempsy.Application.Cluster.Node node = cluster.getNodes().get(0);
final StatsCollector collector = node.getStatsCollector();
// this checks that the throughput works.
assertTrue(poll(baseTimeoutMillis * 5, app, new Condition<Object>() {
@Override
public boolean conditionMet(Object o) {
return app.finalMessageCount.get() > 10;
}
}));
assertNotNull(zookeeperCluster);
assertEquals(0, ((MetricGetters) collector).getDiscardedMessageCount());
assertEquals(0, ((MetricGetters) collector).getMessageFailedCount());
// ok ... so now we have stuff going all the way through. let's kick
// the middle Mp's zookeeper cluster and see what happens.
ZooKeeper origZk = zookeeperCluster.zkref.get();
long sessionid = origZk.getSessionId();
ZooKeeper killer = new ZooKeeper(System.getProperty("zk_connect"), 5000, new Watcher() {
@Override
public void process(WatchedEvent arg0) {
}
}, sessionid, null);
// tricks the server into expiring the other session
killer.close();
// // we should be getting failures now ...
// // but it's possible that it can reconnect prior to actually seeing an error so if this
// // fails frequently we need to remove this test.
// assertTrue(poll(baseTimeoutMillis, app, new Condition()
// {
// @Override
// public boolean conditionMet(Object o)
// {
// return collector.getMessageFailedCount() > 1;
// }
// }));
//... and then recover.
// get the MyMp prototype
cluster = dempsy.getCluster(new ClusterId(FullApplication.class.getSimpleName(), MyMp.class.getSimpleName()));
node = cluster.getNodes().get(0);
final MyMp prototype = (MyMp) node.getMpContainer().getPrototype();
// so let's see where we are
final long interimMessageCount = prototype.myMpReceived.get();
// and now we should eventually get more as the session recovers.
assertTrue(poll(baseTimeoutMillis * 5, app, new Condition<Object>() {
@Override
public boolean conditionMet(Object o) {
return prototype.myMpReceived.get() > interimMessageCount + 100;
}
}));
} finally {
if (dempsy != null)
dempsy.stop();
if (actx != null)
actx.close();
if (dempsy != null)
assertTrue(dempsy.waitToBeStopped(baseTimeoutMillis));
}
}
Aggregations