/**
* 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);
}
@Test
public void testOperatorShutdown() {
dag.setAttribute(OperatorContext.STORAGE_AGENT, new MemoryStorageAgent());
GenericTestOperator o1 = dag.addOperator("o1", GenericTestOperator.class);
GenericTestOperator o2 = dag.addOperator("o2", GenericTestOperator.class);
GenericTestOperator o3 = dag.addOperator("o3", GenericTestOperator.class);
dag.addStream("stream1", o1.outport1, o2.inport1);
dag.addStream("stream2", o2.outport1, o3.inport1);
dag.setAttribute(
o2, OperatorContext.PARTITIONER, new StatelessPartitioner<GenericTestOperator>(2));
StreamingContainerManager scm = new StreamingContainerManager(dag);
PhysicalPlan physicalPlan = scm.getPhysicalPlan();
Map<PTContainer, MockContainer> mockContainers = new HashMap<>();
for (PTContainer c : physicalPlan.getContainers()) {
MockContainer mc = new MockContainer(scm, c);
mockContainers.put(c, mc);
}
// deploy all containers
for (Map.Entry<PTContainer, MockContainer> ce : mockContainers.entrySet()) {
ce.getValue().deploy();
}
for (Map.Entry<PTContainer, MockContainer> ce : mockContainers.entrySet()) {
// skip buffer server purge in monitorHeartbeat
ce.getKey().bufferServerAddress = null;
}
PTOperator o1p1 = physicalPlan.getOperators(dag.getMeta(o1)).get(0);
MockContainer mc1 = mockContainers.get(o1p1.getContainer());
MockOperatorStats o1p1mos = mc1.stats(o1p1.getId());
o1p1mos.currentWindowId(1).checkpointWindowId(1).deployState(DeployState.ACTIVE);
mc1.sendHeartbeat();
PTOperator o2p1 = physicalPlan.getOperators(dag.getMeta(o2)).get(0);
MockContainer mc2 = mockContainers.get(o2p1.getContainer());
MockOperatorStats o2p1mos = mc2.stats(o2p1.getId());
o2p1mos.currentWindowId(1).checkpointWindowId(1).deployState(DeployState.ACTIVE);
mc2.sendHeartbeat();
Assert.assertEquals("2 partitions", 2, physicalPlan.getOperators(dag.getMeta(o2)).size());
PTOperator o2p2 = physicalPlan.getOperators(dag.getMeta(o2)).get(1);
MockContainer mc3 = mockContainers.get(o2p2.getContainer());
MockOperatorStats o2p2mos = mc3.stats(o2p2.getId());
o2p2mos.currentWindowId(1).checkpointWindowId(1).deployState(DeployState.ACTIVE);
mc3.sendHeartbeat();
PTOperator o3p1 = physicalPlan.getOperators(dag.getMeta(o3)).get(0);
MockContainer mc4 = mockContainers.get(o3p1.getContainer());
MockOperatorStats o3p1mos = mc4.stats(o3p1.getId());
o3p1mos.currentWindowId(1).checkpointWindowId(1).deployState(DeployState.ACTIVE);
mc4.sendHeartbeat();
// unifier
PTOperator unifier = physicalPlan.getMergeOperators(dag.getMeta(o2)).get(0);
MockContainer mc5 = mockContainers.get(unifier.getContainer());
MockOperatorStats unifierp1mos = mc5.stats(unifier.getId());
unifierp1mos.currentWindowId(1).checkpointWindowId(1).deployState(DeployState.ACTIVE);
mc5.sendHeartbeat();
o1p1mos.currentWindowId(2).deployState(DeployState.SHUTDOWN);
mc1.sendHeartbeat();
scm.monitorHeartbeat();
Assert.assertEquals("committedWindowId", -1, scm.getCommittedWindowId());
scm.monitorHeartbeat(); // committedWindowId updated in next cycle
Assert.assertEquals("committedWindowId", 1, scm.getCommittedWindowId());
scm.processEvents();
Assert.assertEquals(
"containers at committedWindowId=1", 5, physicalPlan.getContainers().size());
// checkpoint window 2
o1p1mos.checkpointWindowId(2);
mc1.sendHeartbeat();
scm.monitorHeartbeat();
Assert.assertEquals("committedWindowId", 1, scm.getCommittedWindowId());
o2p1mos.currentWindowId(2).checkpointWindowId(2);
o2p2mos.currentWindowId(2).checkpointWindowId(2);
o3p1mos.currentWindowId(2).checkpointWindowId(2);
unifierp1mos.currentWindowId(2).checkpointWindowId(2);
mc2.sendHeartbeat();
mc3.sendHeartbeat();
mc4.sendHeartbeat();
mc5.sendHeartbeat();
scm.monitorHeartbeat();
// Operators are shutdown when both operators reach window Id 2
Assert.assertEquals(0, o1p1.getContainer().getOperators().size());
Assert.assertEquals(0, o2p1.getContainer().getOperators().size());
Assert.assertEquals(0, physicalPlan.getContainers().size());
}