List of usage examples for org.apache.hadoop.yarn.server.nodemanager NodeManager getNMContext
@VisibleForTesting
public Context getNMContext()
From source file:com.splicemachine.test.SpliceTestYarnPlatform.java
License:Apache License
private static void waitForNMToRegister(NodeManager nm) throws Exception { int attempt = 60; ContainerManagerImpl cm = ((ContainerManagerImpl) nm.getNMContext().getContainerManager()); while (cm.getBlockNewContainerRequestsStatus() && attempt-- > 0) { Thread.sleep(2000);// ww w . j a va 2 s.c o m } }
From source file:org.apache.flink.yarn.YARNSessionCapacitySchedulerITCase.java
License:Apache License
/** * Test TaskManager failure and also if the vcores are set correctly (see issue FLINK-2213). *//* w w w . j a v a 2 s .c o m*/ @Test(timeout = 100000) // timeout after 100 seconds public void testTaskManagerFailure() { LOG.info("Starting testTaskManagerFailure()"); Runner runner = startWithArgs( new String[] { "-j", flinkUberjar.getAbsolutePath(), "-t", flinkLibFolder.getAbsolutePath(), "-n", "1", "-jm", "768", "-tm", "1024", "-s", "3", // set the slots 3 to check if the vCores are set properly! "-nm", "customName", "-Dfancy-configuration-value=veryFancy", "-Dyarn.maximum-failed-containers=3", "-D" + ConfigConstants.YARN_VCORES + "=2" }, "Number of connected TaskManagers changed to 1. Slots available: 3", RunTypes.YARN_SESSION); Assert.assertEquals(2, getRunningContainers()); // ------------------------ Test if JobManager web interface is accessible ------- YarnClient yc = null; try { yc = YarnClient.createYarnClient(); yc.init(yarnConfiguration); yc.start(); List<ApplicationReport> apps = yc.getApplications(EnumSet.of(YarnApplicationState.RUNNING)); Assert.assertEquals(1, apps.size()); // Only one running ApplicationReport app = apps.get(0); Assert.assertEquals("customName", app.getName()); String url = app.getTrackingUrl(); if (!url.endsWith("/")) { url += "/"; } if (!url.startsWith("http://")) { url = "http://" + url; } LOG.info("Got application URL from YARN {}", url); String response = TestBaseUtils.getFromHTTP(url + "taskmanagers/"); JsonNode parsedTMs = new ObjectMapper().readTree(response); ArrayNode taskManagers = (ArrayNode) parsedTMs.get("taskmanagers"); Assert.assertNotNull(taskManagers); Assert.assertEquals(1, taskManagers.size()); Assert.assertEquals(3, taskManagers.get(0).get("slotsNumber").asInt()); // get the configuration from webinterface & check if the dynamic properties from YARN show up there. String jsonConfig = TestBaseUtils.getFromHTTP(url + "jobmanager/config"); Map<String, String> parsedConfig = WebMonitorUtils.fromKeyValueJsonArray(jsonConfig); Assert.assertEquals("veryFancy", parsedConfig.get("fancy-configuration-value")); Assert.assertEquals("3", parsedConfig.get("yarn.maximum-failed-containers")); Assert.assertEquals("2", parsedConfig.get(ConfigConstants.YARN_VCORES)); // -------------- FLINK-1902: check if jobmanager hostname/port are shown in web interface // first, get the hostname/port String oC = outContent.toString(); Pattern p = Pattern.compile("Flink JobManager is now running on ([a-zA-Z0-9.-]+):([0-9]+)"); Matcher matches = p.matcher(oC); String hostname = null; String port = null; while (matches.find()) { hostname = matches.group(1).toLowerCase(); port = matches.group(2); } LOG.info("Extracted hostname:port: {} {}", hostname, port); Assert.assertEquals("unable to find hostname in " + jsonConfig, hostname, parsedConfig.get(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY)); Assert.assertEquals("unable to find port in " + jsonConfig, port, parsedConfig.get(ConfigConstants.JOB_MANAGER_IPC_PORT_KEY)); // test logfile access String logs = TestBaseUtils.getFromHTTP(url + "jobmanager/log"); Assert.assertTrue(logs.contains("Starting YARN ApplicationMaster")); Assert.assertTrue(logs.contains("Starting JobManager")); Assert.assertTrue(logs.contains("Starting JobManager Web Frontend")); } catch (Throwable e) { LOG.warn("Error while running test", e); Assert.fail(e.getMessage()); } // ------------------------ Kill container with TaskManager and check if vcores are set correctly ------- // find container id of taskManager: ContainerId taskManagerContainer = null; NodeManager nodeManager = null; UserGroupInformation remoteUgi = null; NMTokenIdentifier nmIdent = null; try { remoteUgi = UserGroupInformation.getCurrentUser(); } catch (IOException e) { LOG.warn("Unable to get curr user", e); Assert.fail(); } for (int nmId = 0; nmId < NUM_NODEMANAGERS; nmId++) { NodeManager nm = yarnCluster.getNodeManager(nmId); ConcurrentMap<ContainerId, Container> containers = nm.getNMContext().getContainers(); for (Map.Entry<ContainerId, Container> entry : containers.entrySet()) { String command = Joiner.on(" ").join(entry.getValue().getLaunchContext().getCommands()); if (command.contains(YarnTaskManager.class.getSimpleName())) { taskManagerContainer = entry.getKey(); nodeManager = nm; nmIdent = new NMTokenIdentifier(taskManagerContainer.getApplicationAttemptId(), null, "", 0); // allow myself to do stuff with the container // remoteUgi.addCredentials(entry.getValue().getCredentials()); remoteUgi.addTokenIdentifier(nmIdent); } } sleep(500); } Assert.assertNotNull("Unable to find container with TaskManager", taskManagerContainer); Assert.assertNotNull("Illegal state", nodeManager); try { List<NodeReport> nodeReports = yc.getNodeReports(NodeState.RUNNING); // we asked for one node with 2 vcores so we expect 2 vcores int userVcores = 0; for (NodeReport rep : nodeReports) { userVcores += rep.getUsed().getVirtualCores(); } Assert.assertEquals(2, userVcores); } catch (Exception e) { Assert.fail("Test failed: " + e.getMessage()); } yc.stop(); List<ContainerId> toStop = new LinkedList<ContainerId>(); toStop.add(taskManagerContainer); StopContainersRequest scr = StopContainersRequest.newInstance(toStop); try { nodeManager.getNMContext().getContainerManager().stopContainers(scr); } catch (Throwable e) { LOG.warn("Error stopping container", e); Assert.fail("Error stopping container: " + e.getMessage()); } // stateful termination check: // wait until we saw a container being killed and AFTERWARDS a new one launched boolean ok = false; do { LOG.debug("Waiting for correct order of events. Output: {}", errContent.toString()); String o = errContent.toString(); int killedOff = o.indexOf("Container killed by the ApplicationMaster"); if (killedOff != -1) { o = o.substring(killedOff); ok = o.indexOf("Launching TaskManager") > 0; } sleep(1000); } while (!ok); // send "stop" command to command line interface runner.sendStop(); // wait for the thread to stop try { runner.join(1000); } catch (InterruptedException e) { LOG.warn("Interrupted while stopping runner", e); } LOG.warn("stopped"); // ----------- Send output to logger System.setOut(originalStdout); System.setErr(originalStderr); String oC = outContent.toString(); String eC = errContent.toString(); LOG.info("Sending stdout content through logger: \n\n{}\n\n", oC); LOG.info("Sending stderr content through logger: \n\n{}\n\n", eC); // ------ Check if everything happened correctly Assert.assertTrue("Expect to see failed container", eC.contains("New messages from the YARN cluster")); Assert.assertTrue("Expect to see failed container", eC.contains("Container killed by the ApplicationMaster")); Assert.assertTrue("Expect to see new container started", eC.contains("Launching TaskManager") && eC.contains("on host")); // cleanup auth for the subsequent tests. remoteUgi.getTokenIdentifiers().remove(nmIdent); LOG.info("Finished testTaskManagerFailure()"); }
From source file:org.apache.flink.yarn.YarnTestBase.java
License:Apache License
public static int getRunningContainers() { int count = 0; for (int nmId = 0; nmId < NUM_NODEMANAGERS; nmId++) { NodeManager nm = yarnCluster.getNodeManager(nmId); ConcurrentMap<ContainerId, Container> containers = nm.getNMContext().getContainers(); count += containers.size();// w ww . ja v a2 s . c o m } return count; }