Search in sources :

Example 1 with MockComponentFactory

use of com.datatorrent.stram.StramLocalCluster.MockComponentFactory in project apex-core by apache.

the class StramLocalClusterTest method testRecovery.

@Test
@SuppressWarnings("SleepWhileInLoop")
public void testRecovery() throws Exception {
    AsyncFSStorageAgent agent = new AsyncFSStorageAgent(testMeta.getPath(), null);
    agent.setSyncCheckpoint(true);
    dag.setAttribute(OperatorContext.STORAGE_AGENT, agent);
    TestGeneratorInputOperator node1 = dag.addOperator("o1", TestGeneratorInputOperator.class);
    // data will be added externally from test
    node1.setMaxTuples(0);
    GenericTestOperator node2 = dag.addOperator("o2", GenericTestOperator.class);
    dag.addStream("o1o2", node1.outport, node2.inport1);
    dag.validate();
    dag.getAttributes().put(LogicalPlan.CHECKPOINT_WINDOW_COUNT, 2);
    final ManualScheduledExecutorService wclock = new ManualScheduledExecutorService(1);
    MockComponentFactory mcf = new MockComponentFactory() {

        @Override
        public WindowGenerator setupWindowGenerator() {
            WindowGenerator wingen = StramTestSupport.setupWindowGenerator(wclock);
            wingen.setCheckpointCount(2, 0);
            return wingen;
        }
    };
    StramLocalCluster localCluster = new StramLocalCluster(dag, mcf);
    localCluster.setPerContainerBufferServer(true);
    // driven by test
    localCluster.setHeartbeatMonitoringEnabled(false);
    localCluster.runAsync();
    PTOperator ptNode1 = localCluster.findByLogicalNode(dag.getMeta(node1));
    PTOperator ptNode2 = localCluster.findByLogicalNode(dag.getMeta(node2));
    LocalStreamingContainer c0 = StramTestSupport.waitForActivation(localCluster, ptNode1);
    Map<Integer, Node<?>> nodeMap = c0.getNodes();
    Assert.assertEquals("number operators", 1, nodeMap.size());
    TestGeneratorInputOperator n1 = (TestGeneratorInputOperator) nodeMap.get(ptNode1.getId()).getOperator();
    Assert.assertNotNull(n1);
    LocalStreamingContainer c2 = StramTestSupport.waitForActivation(localCluster, ptNode2);
    Map<Integer, Node<?>> c2NodeMap = c2.getNodes();
    Assert.assertEquals("number operators downstream", 1, c2NodeMap.size());
    GenericTestOperator n2 = (GenericTestOperator) c2NodeMap.get(localCluster.findByLogicalNode(dag.getMeta(node2)).getId()).getOperator();
    Assert.assertNotNull(n2);
    // input data
    String tuple1 = "tuple1";
    n1.addTuple(tuple1);
    OperatorContext n1Context = c0.getNodeContext(ptNode1.getId());
    Assert.assertEquals("initial window id", -1, n1Context.getLastProcessedWindowId());
    // checkpoint window
    wclock.tick(1);
    wclock.tick(1);
    Assert.assertEquals("current window", 2, wclock.getCurrentTimeMillis());
    OperatorContext o2Context = c2.getNodeContext(ptNode2.getId());
    Assert.assertNotNull("context ", o2Context);
    StramTestSupport.waitForWindowComplete(o2Context, 1);
    Assert.assertEquals("o2 received ", tuple1, n2.inport1Tuple);
    wclock.tick(1);
    Assert.assertEquals("current window", 3, wclock.getCurrentTimeMillis());
    // checkpoint between window 1 and 2
    StramTestSupport.waitForWindowComplete(o2Context, 2);
    // propagate checkpoints to master
    c0.triggerHeartbeat();
    // wait for heartbeat cycle to complete
    c0.waitForHeartbeat(5000);
    Assert.assertEquals("checkpoint " + ptNode1, 1, ptNode1.getRecentCheckpoint().windowId);
    c2.triggerHeartbeat();
    //Thread.yield();
    // yield without using yield for heartbeat cycle
    Thread.sleep(1);
    c2.waitForHeartbeat(5000);
    Assert.assertEquals("checkpoint " + ptNode2, 1, ptNode2.getRecentCheckpoint().windowId);
    Assert.assertEquals("checkpoints " + ptNode1, Arrays.asList(new Checkpoint[] { new Checkpoint(1L, 0, 0) }), ptNode1.checkpoints);
    Assert.assertEquals("checkpoints " + ptNode2, Arrays.asList(new Checkpoint[] { new Checkpoint(1L, 0, 0) }), ptNode2.checkpoints);
    //
    // simulate container failure (operator o1)
    //
    localCluster.failContainer(c0);
    // replacement container starts empty
    // operators will deploy after downstream operator was removed
    LocalStreamingContainer c0Replaced = StramTestSupport.waitForActivation(localCluster, ptNode1);
    c0Replaced.triggerHeartbeat();
    // next heartbeat after setup
    c0Replaced.waitForHeartbeat(5000);
    Assert.assertNotSame("old container", c0, c0Replaced);
    Assert.assertNotSame("old container", c0.getContainerId(), c0Replaced.getContainerId());
    // verify change in downstream container
    LOG.debug("triggering c2 heartbeat processing");
    StreamingContainerAgent c2Agent = localCluster.getContainerAgent(c2);
    // wait for downstream re-deploy to complete
    long startTms = System.currentTimeMillis();
    while (c2Agent.hasPendingWork() && StramTestSupport.DEFAULT_TIMEOUT_MILLIS > System.currentTimeMillis() - startTms) {
        Thread.sleep(200);
        c2.triggerHeartbeat();
        LOG.debug("Waiting for {} to complete pending work.", c2.getContainerId());
    }
    Assert.assertEquals(c2.getContainerId() + " operators after redeploy " + c2.getNodes(), 1, c2.getNodes().size());
    // verify downstream operator re-deployed in existing container
    Assert.assertEquals("active " + ptNode2, c2, StramTestSupport.waitForActivation(localCluster, ptNode2));
    GenericTestOperator o2Recovered = (GenericTestOperator) c2NodeMap.get(localCluster.findByLogicalNode(dag.getMeta(node2)).getId()).getOperator();
    Assert.assertNotNull("redeployed " + ptNode2, o2Recovered);
    Assert.assertNotSame("new instance " + ptNode2, n2, o2Recovered);
    Assert.assertEquals("restored state " + ptNode2, tuple1, o2Recovered.inport1Tuple);
    TestGeneratorInputOperator o1Recovered = (TestGeneratorInputOperator) c0Replaced.getNodes().get(ptNode1.getId()).getOperator();
    Assert.assertNotNull(o1Recovered);
    OperatorContext o1RecoveredContext = c0Replaced.getNodeContext(ptNode1.getId());
    Assert.assertNotNull("active " + ptNode1, o1RecoveredContext);
    wclock.tick(1);
    Assert.assertEquals("current window", 4, wclock.getCurrentTimeMillis());
    // refresh context after operator re-deploy
    o2Context = c2.getNodeContext(ptNode2.getId());
    Assert.assertNotNull("active " + ptNode2, o2Context);
    StramTestSupport.waitForWindowComplete(o1RecoveredContext, 3);
    StramTestSupport.waitForWindowComplete(o2Context, 3);
    // checkpoint window
    wclock.tick(1);
    Assert.assertEquals("current window", 5, wclock.getCurrentTimeMillis());
    String tuple2 = "tuple2";
    o1Recovered.addTuple(tuple2);
    StramTestSupport.waitForWindowComplete(o1RecoveredContext, 4);
    StramTestSupport.waitForWindowComplete(o2Context, 4);
    // check data flow after recovery
    Assert.assertEquals("retrieved tuple (after recovery) " + ptNode2, tuple2, o2Recovered.inport1Tuple);
    // propagate checkpoints to master
    c0Replaced.triggerHeartbeat();
    c0Replaced.waitForHeartbeat(5000);
    c2.triggerHeartbeat();
    c2.waitForHeartbeat(5000);
    // purge checkpoints
    // checkpoint purging
    localCluster.dnmgr.monitorHeartbeat(false);
    Assert.assertEquals("checkpoints " + ptNode1, Arrays.asList(new Checkpoint[] { new Checkpoint(3L, 0, 0) }), ptNode1.checkpoints);
    Assert.assertEquals("checkpoints " + ptNode2, Arrays.asList(new Checkpoint[] { new Checkpoint(3L, 0, 0) }), ptNode2.checkpoints);
    localCluster.shutdown();
}
Also used : PTOperator(com.datatorrent.stram.plan.physical.PTOperator) LocalStreamingContainer(com.datatorrent.stram.StramLocalCluster.LocalStreamingContainer) Node(com.datatorrent.stram.engine.Node) AsyncFSStorageAgent(com.datatorrent.common.util.AsyncFSStorageAgent) TestGeneratorInputOperator(com.datatorrent.stram.engine.TestGeneratorInputOperator) Checkpoint(com.datatorrent.stram.api.Checkpoint) MockComponentFactory(com.datatorrent.stram.StramLocalCluster.MockComponentFactory) GenericTestOperator(com.datatorrent.stram.engine.GenericTestOperator) WindowGenerator(com.datatorrent.stram.engine.WindowGenerator) OperatorContext(com.datatorrent.stram.engine.OperatorContext) ManualScheduledExecutorService(com.datatorrent.stram.support.ManualScheduledExecutorService) Test(org.junit.Test)

Aggregations

AsyncFSStorageAgent (com.datatorrent.common.util.AsyncFSStorageAgent)1 LocalStreamingContainer (com.datatorrent.stram.StramLocalCluster.LocalStreamingContainer)1 MockComponentFactory (com.datatorrent.stram.StramLocalCluster.MockComponentFactory)1 Checkpoint (com.datatorrent.stram.api.Checkpoint)1 GenericTestOperator (com.datatorrent.stram.engine.GenericTestOperator)1 Node (com.datatorrent.stram.engine.Node)1 OperatorContext (com.datatorrent.stram.engine.OperatorContext)1 TestGeneratorInputOperator (com.datatorrent.stram.engine.TestGeneratorInputOperator)1 WindowGenerator (com.datatorrent.stram.engine.WindowGenerator)1 PTOperator (com.datatorrent.stram.plan.physical.PTOperator)1 ManualScheduledExecutorService (com.datatorrent.stram.support.ManualScheduledExecutorService)1 Test (org.junit.Test)1