/**
* Create a services.
*
* <p>The built in services are initialized.
*
* @throws ServiceException thrown if any of the built in services could not initialize.
*/
public Services() throws ServiceException {
if (SERVICES != null) {
XLog log = XLog.getLog(getClass());
log.warn(XLog.OPS, "Previous services singleton active, destroying it");
SERVICES.destroy();
SERVICES = null;
}
setServiceInternal(XLogService.class, false);
setServiceInternal(ConfigurationService.class, true);
conf = get(ConfigurationService.class).getConf();
systemId = conf.get(CONF_SYSTEM_ID, ("oozie-" + System.getProperty("user.name")));
if (systemId.length() > MAX_SYSTEM_ID_LEN) {
systemId = systemId.substring(0, MAX_SYSTEM_ID_LEN);
XLog.getLog(getClass())
.warn(
"System ID [{0}] exceeds maximun lenght [{1}], trimming",
systemId, MAX_SYSTEM_ID_LEN);
}
setSystemMode(SYSTEM_MODE.valueOf(conf.get(CONF_SYSTEM_MODE, SYSTEM_MODE.NORMAL.toString())));
runtimeDir = createRuntimeDir();
}
private void setServiceInternal(Class<? extends Service> klass, boolean logging)
throws ServiceException {
try {
Service newService = (Service) ReflectionUtils.newInstance(klass, null);
Service oldService = services.get(newService.getInterface());
if (oldService != null) {
oldService.destroy();
}
if (logging) {
XLog log = new XLog(LogFactory.getLog(getClass()));
log.trace(
"Initializing service[{0}] class[{1}]",
newService.getInterface(), newService.getClass());
}
newService.init(this);
services.put(newService.getInterface(), newService);
} catch (ServiceException ex) {
XLog.getLog(getClass()).fatal(ex.getMessage(), ex);
destroy();
throw ex;
}
}
@Override
protected void tearDown() throws Exception {
LocalOozie.stop();
services.destroy();
super.tearDown();
}
@Override
protected void tearDown() throws Exception {
services.destroy();
super.tearDown();
}
public void testActionCheckTransientDuringMRAction() throws Exception {
// When using YARN, skip this test because it relies on shutting down the job tracker, which
// isn't used in YARN
if (createJobConf().get("yarn.resourcemanager.address") != null) {
return;
}
services.destroy();
// Make the max number of retries lower so the test won't take as long
final int maxRetries = 2;
setSystemProperty("oozie.action.retries.max", Integer.toString(maxRetries));
services = new Services();
// Disable ActionCheckerService so it doesn't interfere by triggering any extra
// ActionCheckXCommands
setClassesToBeExcluded(
services.getConf(), new String[] {"org.apache.oozie.service.ActionCheckerService"});
services.init();
final JPAService jpaService = Services.get().get(JPAService.class);
WorkflowJobBean job0 =
this.addRecordToWfJobTable(WorkflowJob.Status.RUNNING, WorkflowInstance.Status.RUNNING);
final String jobId = job0.getId();
WorkflowActionBean action0 =
this.addRecordToWfActionTable(jobId, "1", WorkflowAction.Status.PREP);
final String actionId = action0.getId();
final WorkflowActionGetJPAExecutor wfActionGetCmd = new WorkflowActionGetJPAExecutor(actionId);
new ActionStartXCommand(actionId, "map-reduce").call();
final WorkflowActionBean action1 = jpaService.execute(wfActionGetCmd);
String originalLauncherId = action1.getExternalId();
ActionExecutorContext context =
new ActionXCommand.ActionExecutorContext(job0, action1, false, false);
MapReduceActionExecutor actionExecutor = new MapReduceActionExecutor();
JobConf conf =
actionExecutor.createBaseHadoopConf(context, XmlUtils.parseXml(action1.getConf()));
String user = conf.get("user.name");
JobClient jobClient =
Services.get().get(HadoopAccessorService.class).createJobClient(user, conf);
final RunningJob launcherJob = jobClient.getJob(JobID.forName(originalLauncherId));
waitFor(
120 * 1000,
new Predicate() {
@Override
public boolean evaluate() throws Exception {
return launcherJob.isComplete();
}
});
assertTrue(launcherJob.isSuccessful());
Map<String, String> actionData =
LauncherMapperHelper.getActionData(getFileSystem(), context.getActionDir(), conf);
assertTrue(LauncherMapperHelper.hasIdSwap(actionData));
new ActionCheckXCommand(action1.getId()).call();
WorkflowActionBean action2 = jpaService.execute(wfActionGetCmd);
String originalMapperId = action2.getExternalChildIDs();
assertFalse(originalLauncherId.equals(originalMapperId));
// At this point, the launcher job has finished and the map-reduce action has started (but not
// finished)
// Now, shutdown the job tracker to pretend it has gone down during the map-reduce job
executeWhileJobTrackerIsShutdown(
new ShutdownJobTrackerExecutable() {
@Override
public void execute() throws Exception {
assertEquals(0, action1.getRetries());
new ActionCheckXCommand(actionId).call();
waitFor(
30 * 1000,
new Predicate() {
@Override
public boolean evaluate() throws Exception {
WorkflowActionBean action1a = jpaService.execute(wfActionGetCmd);
return (action1a.getRetries() > 0);
}
});
waitFor(
180 * 1000,
new Predicate() {
@Override
public boolean evaluate() throws Exception {
WorkflowActionBean action1a = jpaService.execute(wfActionGetCmd);
return (action1a.getRetries() == 0);
}
});
WorkflowActionBean action1b = jpaService.execute(wfActionGetCmd);
assertEquals(0, action1b.getRetries());
assertEquals("START_MANUAL", action1b.getStatusStr());
WorkflowJobBean job1 = jpaService.execute(new WorkflowJobGetJPAExecutor(jobId));
assertEquals("SUSPENDED", job1.getStatusStr());
// At this point, the action has gotten a transient error, even after maxRetries tries
// so the workflow has been
// SUSPENDED
}
});
// Now, lets bring the job tracker back up and resume the workflow (which will restart the
// current action)
// It should now continue and finish with SUCCEEDED
new ResumeXCommand(jobId).call();
WorkflowJobBean job2 = jpaService.execute(new WorkflowJobGetJPAExecutor(jobId));
assertEquals("RUNNING", job2.getStatusStr());
sleep(500);
new ActionCheckXCommand(actionId).call();
WorkflowActionBean action3 = jpaService.execute(wfActionGetCmd);
String launcherId = action3.getExternalId();
assertFalse(originalLauncherId.equals(launcherId));
final RunningJob launcherJob2 = jobClient.getJob(JobID.forName(launcherId));
waitFor(
120 * 1000,
new Predicate() {
@Override
public boolean evaluate() throws Exception {
return launcherJob2.isComplete();
}
});
assertTrue(launcherJob2.isSuccessful());
actionData = LauncherMapperHelper.getActionData(getFileSystem(), context.getActionDir(), conf);
assertTrue(LauncherMapperHelper.hasIdSwap(actionData));
new ActionCheckXCommand(actionId).call();
WorkflowActionBean action4 = jpaService.execute(wfActionGetCmd);
String mapperId = action4.getExternalChildIDs();
assertFalse(originalMapperId.equals(mapperId));
final RunningJob mrJob = jobClient.getJob(JobID.forName(mapperId));
waitFor(
120 * 1000,
new Predicate() {
@Override
public boolean evaluate() throws Exception {
return mrJob.isComplete();
}
});
assertTrue(mrJob.isSuccessful());
new ActionCheckXCommand(actionId).call();
WorkflowActionBean action5 = jpaService.execute(wfActionGetCmd);
assertEquals("SUCCEEDED", action5.getExternalStatus());
}
