@Test
public void testWriteAheadLog() throws Exception {
final MutableInt flushCount = new MutableInt();
final MutableBoolean isClosed = new MutableBoolean(false);
LogicalPlan dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, testMeta.dir);
dag.setAttribute(OperatorContext.STORAGE_AGENT, new FSStorageAgent(testMeta.dir, null));
TestGeneratorInputOperator o1 = dag.addOperator("o1", TestGeneratorInputOperator.class);
StreamingContainerManager scm = new StreamingContainerManager(dag);
PhysicalPlan plan = scm.getPhysicalPlan();
Journal j = scm.getJournal();
ByteArrayOutputStream bos =
new ByteArrayOutputStream() {
@Override
public void flush() throws IOException {
super.flush();
flushCount.increment();
}
@Override
public void close() throws IOException {
super.close();
isClosed.setValue(true);
}
};
j.setOutputStream(new DataOutputStream(bos));
PTOperator o1p1 = plan.getOperators(dag.getMeta(o1)).get(0);
assertEquals(PTOperator.State.PENDING_DEPLOY, o1p1.getState());
String externalId = new MockContainer(scm, o1p1.getContainer()).container.getExternalId();
assertEquals("flush count", 1, flushCount.intValue());
o1p1.setState(PTOperator.State.ACTIVE);
assertEquals(PTOperator.State.ACTIVE, o1p1.getState());
assertEquals("flush count", 2, flushCount.intValue());
assertEquals("is closed", false, isClosed.booleanValue());
// this will close the stream. There are 2 calls to flush() during the close() - one in Kryo
// Output and one
// in FilterOutputStream
j.setOutputStream(null);
assertEquals("flush count", 4, flushCount.intValue());
assertEquals("is closed", true, isClosed.booleanValue());
// output stream is closed, so state will be changed without recording it in the journal
o1p1.setState(PTOperator.State.INACTIVE);
assertEquals(PTOperator.State.INACTIVE, o1p1.getState());
assertEquals("flush count", 4, flushCount.intValue());
ByteArrayInputStream bis = new ByteArrayInputStream(bos.toByteArray());
j.replay(new DataInputStream(bis));
assertEquals(PTOperator.State.ACTIVE, o1p1.getState());
InetSocketAddress addr1 = InetSocketAddress.createUnresolved("host1", 1);
PTContainer c1 = plan.getContainers().get(0);
c1.setState(PTContainer.State.ALLOCATED);
c1.host = "host1";
c1.bufferServerAddress = addr1;
c1.setAllocatedMemoryMB(2);
c1.setRequiredMemoryMB(1);
c1.setAllocatedVCores(3);
c1.setRequiredVCores(4);
j.setOutputStream(new DataOutputStream(bos));
j.write(c1.getSetContainerState());
c1.setExternalId(null);
c1.setState(PTContainer.State.NEW);
c1.setExternalId(null);
c1.host = null;
c1.bufferServerAddress = null;
bis = new ByteArrayInputStream(bos.toByteArray());
j.replay(new DataInputStream(bis));
assertEquals(externalId, c1.getExternalId());
assertEquals(PTContainer.State.ALLOCATED, c1.getState());
assertEquals("host1", c1.host);
assertEquals(addr1, c1.bufferServerAddress);
assertEquals(1, c1.getRequiredMemoryMB());
assertEquals(2, c1.getAllocatedMemoryMB());
assertEquals(3, c1.getAllocatedVCores());
assertEquals(4, c1.getRequiredVCores());
j.write(scm.getSetOperatorProperty("o1", "maxTuples", "100"));
o1.setMaxTuples(10);
j.setOutputStream(null);
bis = new ByteArrayInputStream(bos.toByteArray());
j.replay(new DataInputStream(bis));
assertEquals(100, o1.getMaxTuples());
j.setOutputStream(new DataOutputStream(bos));
scm.setOperatorProperty("o1", "maxTuples", "10");
assertEquals(10, o1.getMaxTuples());
o1.setMaxTuples(100);
assertEquals(100, o1.getMaxTuples());
j.setOutputStream(null);
bis = new ByteArrayInputStream(bos.toByteArray());
j.replay(new DataInputStream(bis));
assertEquals(10, o1.getMaxTuples());
j.setOutputStream(new DataOutputStream(bos));
scm.setPhysicalOperatorProperty(o1p1.getId(), "maxTuples", "50");
}
private void testRestartApp(StorageAgent agent, String appPath1) throws Exception {
FileUtils.deleteDirectory(new File(testMeta.dir)); // clean any state from previous run
String appId1 = "app1";
String appId2 = "app2";
String appPath2 = testMeta.dir + "/" + appId2;
LogicalPlan dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_ID, appId1);
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath1);
dag.setAttribute(LogicalPlan.APPLICATION_ATTEMPT_ID, 1);
dag.setAttribute(OperatorContext.STORAGE_AGENT, agent);
dag.addOperator("o1", StatsListeningOperator.class);
FSRecoveryHandler recoveryHandler =
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false));
StreamingContainerManager.getInstance(recoveryHandler, dag, false);
// test restore initial snapshot + log
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath1);
StreamingContainerManager scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
PhysicalPlan plan = scm.getPhysicalPlan();
dag = plan.getLogicalPlan(); // original plan
Assert.assertNotNull("operator", dag.getOperatorMeta("o1"));
PTOperator o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
long[] ids =
new FSStorageAgent(appPath1 + "/" + LogicalPlan.SUBDIR_CHECKPOINTS, new Configuration())
.getWindowIds(o1p1.getId());
Assert.assertArrayEquals(new long[] {o1p1.getRecoveryCheckpoint().getWindowId()}, ids);
Assert.assertNull(o1p1.getContainer().getExternalId());
// trigger journal write
o1p1.getContainer().setExternalId("cid1");
scm.writeJournal(o1p1.getContainer().getSetContainerState());
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath2);
dag.setAttribute(LogicalPlan.APPLICATION_ID, appId2);
StramClient sc = new StramClient(new Configuration(), dag);
try {
sc.start();
sc.copyInitialState(new Path(appPath1));
} finally {
sc.stop();
}
scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
plan = scm.getPhysicalPlan();
dag = plan.getLogicalPlan();
assertEquals("modified appId", appId2, dag.getValue(LogicalPlan.APPLICATION_ID));
assertEquals("modified appPath", appPath2, dag.getValue(LogicalPlan.APPLICATION_PATH));
Assert.assertNotNull("operator", dag.getOperatorMeta("o1"));
o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
assertEquals("journal copied", "cid1", o1p1.getContainer().getExternalId());
ids =
new FSStorageAgent(appPath2 + "/" + LogicalPlan.SUBDIR_CHECKPOINTS, new Configuration())
.getWindowIds(o1p1.getId());
Assert.assertArrayEquals(
"checkpoints copied", new long[] {o1p1.getRecoveryCheckpoint().getWindowId()}, ids);
}
/**
* Test serialization of the container manager with mock execution layer.
*
* @throws Exception
*/
private void testContainerManager(StorageAgent agent) throws Exception {
FileUtils.deleteDirectory(new File(testMeta.dir)); // clean any state from previous run
LogicalPlan dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, testMeta.dir);
dag.setAttribute(OperatorContext.STORAGE_AGENT, agent);
StatsListeningOperator o1 = dag.addOperator("o1", StatsListeningOperator.class);
FSRecoveryHandler recoveryHandler =
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false));
StreamingContainerManager scm =
StreamingContainerManager.getInstance(recoveryHandler, dag, false);
File expFile = new File(recoveryHandler.getDir(), FSRecoveryHandler.FILE_SNAPSHOT);
Assert.assertTrue("snapshot file " + expFile, expFile.exists());
PhysicalPlan plan = scm.getPhysicalPlan();
assertEquals("number required containers", 1, plan.getContainers().size());
PTOperator o1p1 = plan.getOperators(dag.getMeta(o1)).get(0);
@SuppressWarnings(
"UnusedAssignment") /* sneaky: the constructor does some changes to the container */
MockContainer mc = new MockContainer(scm, o1p1.getContainer());
PTContainer originalContainer = o1p1.getContainer();
Assert.assertNotNull(o1p1.getContainer().bufferServerAddress);
assertEquals(PTContainer.State.ACTIVE, o1p1.getContainer().getState());
assertEquals("state " + o1p1, PTOperator.State.PENDING_DEPLOY, o1p1.getState());
// test restore initial snapshot + log
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, testMeta.dir);
scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
dag = scm.getLogicalPlan();
plan = scm.getPhysicalPlan();
o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
assertEquals("post restore state " + o1p1, PTOperator.State.PENDING_DEPLOY, o1p1.getState());
o1 = (StatsListeningOperator) o1p1.getOperatorMeta().getOperator();
assertEquals(
"containerId", originalContainer.getExternalId(), o1p1.getContainer().getExternalId());
assertEquals("stats listener", 1, o1p1.statsListeners.size());
assertEquals("number stats calls", 0, o1.processStatsCnt); // stats are not logged
assertEquals("post restore 1", PTContainer.State.ALLOCATED, o1p1.getContainer().getState());
assertEquals(
"post restore 1",
originalContainer.bufferServerAddress,
o1p1.getContainer().bufferServerAddress);
StreamingContainerAgent sca = scm.getContainerAgent(originalContainer.getExternalId());
Assert.assertNotNull("allocated container restored " + originalContainer, sca);
assertEquals(
"memory usage allocated container",
(int) OperatorContext.MEMORY_MB.defaultValue,
sca.container.getAllocatedMemoryMB());
// YARN-1490 - simulate container terminated on AM recovery
scm.scheduleContainerRestart(originalContainer.getExternalId());
assertEquals("memory usage of failed container", 0, sca.container.getAllocatedMemoryMB());
Checkpoint firstCheckpoint = new Checkpoint(3, 0, 0);
mc = new MockContainer(scm, o1p1.getContainer());
checkpoint(scm, o1p1, firstCheckpoint);
mc.stats(o1p1.getId())
.deployState(OperatorHeartbeat.DeployState.ACTIVE)
.currentWindowId(3)
.checkpointWindowId(3);
mc.sendHeartbeat();
assertEquals("state " + o1p1, PTOperator.State.ACTIVE, o1p1.getState());
// logical plan modification triggers snapshot
CreateOperatorRequest cor = new CreateOperatorRequest();
cor.setOperatorFQCN(GenericTestOperator.class.getName());
cor.setOperatorName("o2");
CreateStreamRequest csr = new CreateStreamRequest();
csr.setSourceOperatorName("o1");
csr.setSourceOperatorPortName("outport");
csr.setSinkOperatorName("o2");
csr.setSinkOperatorPortName("inport1");
FutureTask<?> lpmf = scm.logicalPlanModification(Lists.newArrayList(cor, csr));
while (!lpmf.isDone()) {
scm.monitorHeartbeat();
}
Assert.assertNull(lpmf.get()); // unmask exception, if any
Assert.assertSame("dag references", dag, scm.getLogicalPlan());
assertEquals("number operators after plan modification", 2, dag.getAllOperators().size());
// set operator state triggers journal write
o1p1.setState(PTOperator.State.INACTIVE);
Checkpoint offlineCheckpoint = new Checkpoint(10, 0, 0);
// write checkpoint while AM is out,
// it needs to be picked up as part of restore
checkpoint(scm, o1p1, offlineCheckpoint);
// test restore
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, testMeta.dir);
scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
Assert.assertNotSame("dag references", dag, scm.getLogicalPlan());
assertEquals(
"number operators after restore", 2, scm.getLogicalPlan().getAllOperators().size());
dag = scm.getLogicalPlan();
plan = scm.getPhysicalPlan();
o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
assertEquals("post restore state " + o1p1, PTOperator.State.INACTIVE, o1p1.getState());
o1 = (StatsListeningOperator) o1p1.getOperatorMeta().getOperator();
assertEquals("stats listener", 1, o1p1.statsListeners.size());
assertEquals("number stats calls post restore", 1, o1.processStatsCnt);
assertEquals("post restore 1", PTContainer.State.ACTIVE, o1p1.getContainer().getState());
assertEquals(
"post restore 1",
originalContainer.bufferServerAddress,
o1p1.getContainer().bufferServerAddress);
// offline checkpoint detection
assertEquals(
"checkpoints after recovery",
Lists.newArrayList(firstCheckpoint, offlineCheckpoint),
o1p1.checkpoints);
}