/** This also implements the Apache Commons Daemon interface's start */
public synchronized void start() {
FlumeConfiguration conf = FlumeConfiguration.get();
// the simple report interface
simpleReportManager.add(vmInfo);
simpleReportManager.add(sysInfo);
simpleReportManager.add(
new Reportable() {
@Override
public String getName() {
return FlumeNode.this.getName();
}
@Override
public ReportEvent getMetrics() {
return FlumeNode.this.getReport();
}
@Override
public Map<String, Reportable> getSubMetrics() {
return ReportUtil.noChildren();
}
});
// the full report interface
ReportManager.get().add(vmInfo);
ReportManager.get().add(sysInfo);
ReportManager.get().add(this);
if (startHttp) {
int nodePort = conf.getNodeStatusPort();
String bindAddress = "0.0.0.0";
ContextCreator cc =
new ContextCreator() {
@Override
public void addContexts(ContextHandlerCollection handlers) {
handlers.addHandler(InternalHttpServer.createLogAppContext());
handlers.addHandler(InternalHttpServer.createStackSevletContext());
String webAppRoot = FlumeConfiguration.get().getNodeWebappRoot();
InternalHttpServer.addHandlersFromPaths(handlers, new File(webAppRoot));
}
};
http = InternalHttpServer.startFindPortHttpServer(cc, bindAddress, nodePort);
}
if (reportPusher != null) {
reportPusher.start();
}
if (liveMan != null) {
liveMan.start();
}
if (chokeMan != null) {
// JVM exits if only daemons threads remain.
chokeMan.setDaemon(true);
chokeMan.start();
}
}
/**
* Tests to make sure the report sink receives data.
*
* @throws InterruptedException
*/
@Test
public void testReportSink() throws FlumeSpecException, IOException, InterruptedException {
String spec =
"{benchinject(\"foo\") => {benchreport(\"report\", \"[ console , counter(\\\"test\\\") ]\") => null } }";
EventSink snk = FlumeBuilder.buildSink(new ReportTestingContext(), spec);
snk.open();
snk.append(new EventImpl(new byte[0]));
snk.append(new EventImpl(new byte[0]));
snk.close();
CounterSink ctr = (CounterSink) ReportManager.get().getReportable("test");
Assert.assertEquals(1, ctr.getCount());
}
@Test
public void testAutoRoll() throws IOException, InterruptedException {
RollSink snk = new RollSink(new ReportTestingContext(), "counter(\"foo\")", 2000, 10000); // two
// second sleeper, but check period is really long
Event e = new EventImpl("this is a test message".getBytes());
snk.open();
snk.append(e);
CounterSink cnt = (CounterSink) ReportManager.get().getReportable("foo");
Clock.sleep(3000); // sleep 3s
// the roller automatically flushed!
assertEquals(1, cnt.getCount());
snk.close();
}
/**
* This tests the new failover builder that uses specs strings as arguments and instantiates them!
*
* @throws InterruptedException
*/
@Test
public void testBackoffFailoverBuilder() throws IOException, InterruptedException {
SinkBuilder bld = BackOffFailOverSink.builder();
EventSink snk =
bld.build(
new ReportTestingContext(),
"{intervalFlakeyAppend(2) => counter(\"pri\") } ",
"counter(\"sec\")");
snk.open();
Event e = new EventImpl("foo".getBytes());
snk.append(e);
snk.append(e);
snk.append(e);
snk.append(e);
snk.append(e);
snk.close();
CounterSink priCnt = (CounterSink) ReportManager.get().getReportable("pri");
CounterSink secCnt = (CounterSink) ReportManager.get().getReportable("sec");
// these are timing based, may fail.
Assert.assertEquals(1, priCnt.getCount());
Assert.assertEquals(4, secCnt.getCount());
}
public void doTestLogicalNodesConcurrentDFOMans(final int threads, final int events, int timeout)
throws IOException, InterruptedException, FlumeSpecException {
FlumeTestHarness.setupLocalWriteDir();
FlumeMaster master = new FlumeMaster();
FlumeNode node = new FlumeNode(new DirectMasterRPC(master), false, false);
final Reportable[] dfos = new Reportable[threads];
for (int i = 0; i < threads; i++) {
String name = "test." + i;
String report = "report." + i;
int count = events + i;
String src = "asciisynth(" + count + ",100)";
String snk = "{ diskFailover => counter(\"" + report + "\") } ";
node.getLogicalNodeManager().testingSpawn(name, src, snk);
dfos[i] = node.getLogicalNodeManager().get(name);
}
// TODO (jon) using sleep is cheating to give all threads a chance to start.
// Test seems flakey without this due to a race condition.
Thread.sleep(500);
// wait for all to be done.
waitForEmptyDFOs(node, timeout);
// check to make sure everyone got the right number of events
boolean success = true;
for (int i = 0; i < threads; i++) {
LOG.info(dfos[i].getMetrics());
}
for (int i = 0; i < threads; i++) {
CounterSink cnt = (CounterSink) ReportManager.get().getReportable("report." + i);
LOG.info(i + " expected " + (events + i) + " and got " + cnt.getCount());
success &= ((events + i) == cnt.getCount());
assertEquals(events + i, cnt.getCount());
}
assertTrue("Counts did not line up", success);
FlumeTestHarness.cleanupLocalWriteDir();
}
/**
* Bang on the wal mechanism as hard was you want with number of concurrent threads, number of
* events per thread. Timeout in millis,
*/
public void doTestConcurrentDFOMans(final int threads, final int events, int timeout)
throws IOException, InterruptedException {
final CountDownLatch started = new CountDownLatch(threads);
final CountDownLatch done = new CountDownLatch(threads);
final DiskFailoverManager[] dfos = new DiskFailoverManager[threads];
for (int i = 0; i < threads; i++) {
final int idx = i;
new Thread("Concurrent-" + i) {
@Override
public void run() {
try {
File f1 = FileUtil.mktempdir();
AccumulatorSink cnt1 = new AccumulatorSink("count." + idx);
DiskFailoverManager dfoMan = new NaiveFileFailoverManager(f1);
dfos[idx] = dfoMan; // save for checking.
// short trigger causes lots of rolls
EventSink snk =
new DiskFailoverDeco(
cnt1, LogicalNodeContext.testingContext(), dfoMan, new TimeTrigger(100), 50);
ReportManager.get().add(cnt1);
// make each parallel instance send a slightly different number of
// messages.
EventSource src = new NoNlASCIISynthSource(events + idx, 100);
src.open();
snk.open();
started.countDown();
EventUtil.dumpAll(src, snk);
src.close();
snk.close(); // this triggers a flush of current file!?
FileUtil.rmr(f1);
} catch (Exception e) {
LOG.error(e, e);
} finally {
done.countDown();
}
}
}.start();
}
started.await();
boolean ok = done.await(timeout, TimeUnit.MILLISECONDS);
assertTrue("Test timed out", ok);
for (int i = 0; i < threads; i++) {
AccumulatorSink cnt = (AccumulatorSink) ReportManager.get().getReportable("count." + i);
// check for the slightly different counts based on thread.
int exp = events + i;
LOG.info("count." + i + " expected " + exp + " and got " + cnt.getCount());
assertEquals(exp, (int) cnt.getCount());
// check dfo reports to see if they are sane.
ReportEvent rpt = dfos[i].getMetrics();
LOG.info(rpt);
long failovered = rpt.getLongMetric(DiskFailoverManager.A_MSG_WRITING);
assertEquals(events + i, failovered);
}
}
public void serve() throws IOException {
if (cfg.getMasterStore().equals(ZK_CFG_STORE)) {
try {
ZooKeeperService.getAndInit(cfg);
} catch (InterruptedException e) {
throw new IOException("Unexpected interrupt when starting ZooKeeper", e);
}
}
ReportManager.get().add(vmInfo);
ReportManager.get().add(sysInfo);
if (doHttp) {
String webPath = FlumeNode.getWebPath(cfg);
this.http =
new StatusHttpServer("flumeconfig", webPath, "0.0.0.0", cfg.getMasterHttpPort(), false);
http.addServlet(jerseyMasterServlet(), "/master/*");
http.start();
}
controlServer = new MasterClientServer(this, FlumeConfiguration.get());
configServer = new MasterAdminServer(this, FlumeConfiguration.get());
/*
* We instantiate both kinds of report servers below, but no resources are
* allocated till we call serve() on them.
*/
avroReportServer = new AvroReportServer(FlumeConfiguration.get().getReportServerPort());
thriftReportServer = new ThriftReportServer(FlumeConfiguration.get().getReportServerPort());
ReportManager.get().add(this);
try {
controlServer.serve();
configServer.serve();
/*
* Start the Avro/Thrift ReportServer based on the flag set in the
* configuration file.
*/
if (cfg.getReportServerRPC() == cfg.RPC_TYPE_AVRO) {
avroReportServer.serve();
} else {
thriftReportServer.serve();
}
} catch (TTransportException e1) {
throw new IOException("Error starting control or config server", e1);
}
cmdman.start();
ackman.start();
specman.start();
// TODO (jon) clean shutdown
reaper =
new Thread("Lost node reaper") {
@Override
public void run() {
try {
while (true) {
Thread.sleep(FlumeConfiguration.get().getConfigHeartbeatPeriod());
statman.checkup();
}
} catch (InterruptedException e) {
LOG.error("Reaper thread unexpectedly interrupted:" + e.getMessage());
LOG.debug("Lost node reaper unexpectedly interrupted", e);
}
}
};
reaper.start();
}