public void testActionCheck() throws Exception {
JPAService jpaService = Services.get().get(JPAService.class);
WorkflowJobBean job =
this.addRecordToWfJobTable(WorkflowJob.Status.RUNNING, WorkflowInstance.Status.RUNNING);
WorkflowActionBean action =
this.addRecordToWfActionTable(job.getId(), "1", WorkflowAction.Status.PREP);
WorkflowActionGetJPAExecutor wfActionGetCmd = new WorkflowActionGetJPAExecutor(action.getId());
new ActionStartXCommand(action.getId(), "map-reduce").call();
action = jpaService.execute(wfActionGetCmd);
ActionExecutorContext context =
new ActionXCommand.ActionExecutorContext(job, action, false, false);
MapReduceActionExecutor actionExecutor = new MapReduceActionExecutor();
JobConf conf =
actionExecutor.createBaseHadoopConf(context, XmlUtils.parseXml(action.getConf()));
String user = conf.get("user.name");
JobClient jobClient =
Services.get().get(HadoopAccessorService.class).createJobClient(user, conf);
String launcherId = action.getExternalId();
final RunningJob launcherJob = jobClient.getJob(JobID.forName(launcherId));
waitFor(
120 * 1000,
new Predicate() {
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(action.getId()).call();
action = jpaService.execute(wfActionGetCmd);
String mapperId = action.getExternalId();
String childId = action.getExternalChildIDs();
assertTrue(launcherId.equals(mapperId));
final RunningJob mrJob = jobClient.getJob(JobID.forName(childId));
waitFor(
120 * 1000,
new Predicate() {
public boolean evaluate() throws Exception {
return mrJob.isComplete();
}
});
assertTrue(mrJob.isSuccessful());
new ActionCheckXCommand(action.getId()).call();
action = jpaService.execute(wfActionGetCmd);
assertEquals("SUCCEEDED", action.getExternalStatus());
}
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());
}
public void testGet() throws Exception {
WorkflowActionBean bean =
addRecordToWfActionTable("workflowId", "testAction", WorkflowAction.Status.PREP, "", true);
WorkflowActionBean retBean;
// GET_ACTION_ID_TYPE_LASTCHECK
retBean =
WorkflowActionQueryExecutor.getInstance()
.get(WorkflowActionQuery.GET_ACTION_ID_TYPE_LASTCHECK, bean.getId());
assertEquals(bean.getId(), retBean.getId());
assertEquals(bean.getType(), retBean.getType());
assertEquals(bean.getLastCheckTime(), retBean.getLastCheckTime());
// GET_ACTION_FAIL
retBean =
WorkflowActionQueryExecutor.getInstance()
.get(WorkflowActionQuery.GET_ACTION_FAIL, bean.getId());
assertEquals(bean.getId(), retBean.getId());
assertEquals(bean.getJobId(), retBean.getJobId());
assertEquals(bean.getName(), retBean.getName());
assertEquals(bean.getStatusStr(), retBean.getStatusStr());
assertEquals(bean.getPending(), retBean.getPending());
assertEquals(bean.getType(), retBean.getType());
assertEquals(bean.getLogToken(), retBean.getLogToken());
assertEquals(bean.getTransition(), retBean.getTransition());
assertEquals(bean.getErrorCode(), retBean.getErrorCode());
assertEquals(bean.getErrorMessage(), retBean.getErrorMessage());
assertNull(retBean.getConf());
assertNull(retBean.getSlaXml());
assertNull(retBean.getData());
assertNull(retBean.getStats());
assertNull(retBean.getExternalChildIDs());
// GET_ACTION_SIGNAL
retBean =
WorkflowActionQueryExecutor.getInstance()
.get(WorkflowActionQuery.GET_ACTION_SIGNAL, bean.getId());
assertEquals(bean.getId(), retBean.getId());
assertEquals(bean.getJobId(), retBean.getJobId());
assertEquals(bean.getName(), retBean.getName());
assertEquals(bean.getStatusStr(), retBean.getStatusStr());
assertEquals(bean.getPending(), retBean.getPending());
assertEquals(bean.getPendingAge().getTime(), retBean.getPendingAge().getTime());
assertEquals(bean.getType(), retBean.getType());
assertEquals(bean.getLogToken(), retBean.getLogToken());
assertEquals(bean.getTransition(), retBean.getTransition());
assertEquals(bean.getErrorCode(), retBean.getErrorCode());
assertEquals(bean.getErrorMessage(), retBean.getErrorMessage());
assertEquals(bean.getExecutionPath(), retBean.getExecutionPath());
assertEquals(bean.getSignalValue(), retBean.getSignalValue());
assertEquals(bean.getSlaXml(), retBean.getSlaXml());
assertNull(retBean.getConf());
assertNull(retBean.getData());
assertNull(retBean.getStats());
assertNull(retBean.getExternalChildIDs());
// GET_ACTION_CHECK
retBean =
WorkflowActionQueryExecutor.getInstance()
.get(WorkflowActionQuery.GET_ACTION_CHECK, bean.getId());
assertEquals(bean.getId(), retBean.getId());
assertEquals(bean.getJobId(), retBean.getJobId());
assertEquals(bean.getName(), retBean.getName());
assertEquals(bean.getStatusStr(), retBean.getStatusStr());
assertEquals(bean.getPending(), retBean.getPending());
assertEquals(bean.getPendingAge().getTime(), retBean.getPendingAge().getTime());
assertEquals(bean.getType(), retBean.getType());
assertEquals(bean.getLogToken(), retBean.getLogToken());
assertEquals(bean.getTransition(), retBean.getTransition());
assertEquals(bean.getRetries(), retBean.getRetries());
assertEquals(bean.getUserRetryCount(), retBean.getUserRetryCount());
assertEquals(bean.getUserRetryMax(), retBean.getUserRetryMax());
assertEquals(bean.getUserRetryInterval(), retBean.getUserRetryInterval());
assertEquals(bean.getTrackerUri(), retBean.getTrackerUri());
assertEquals(bean.getStartTime().getTime(), retBean.getStartTime().getTime());
assertEquals(bean.getEndTime().getTime(), retBean.getEndTime().getTime());
assertEquals(bean.getLastCheckTime().getTime(), retBean.getLastCheckTime().getTime());
assertEquals(bean.getErrorCode(), retBean.getErrorCode());
assertEquals(bean.getErrorMessage(), retBean.getErrorMessage());
assertEquals(bean.getExternalId(), retBean.getExternalId());
assertEquals(bean.getExternalStatus(), retBean.getExternalStatus());
assertEquals(bean.getExternalChildIDs(), retBean.getExternalChildIDs());
assertEquals(bean.getConf(), retBean.getConf());
assertNull(retBean.getData());
assertNull(retBean.getStats());
assertNull(retBean.getSlaXml());
// GET_ACTION_END
retBean =
WorkflowActionQueryExecutor.getInstance()
.get(WorkflowActionQuery.GET_ACTION_END, bean.getId());
assertEquals(bean.getId(), retBean.getId());
assertEquals(bean.getJobId(), retBean.getJobId());
assertEquals(bean.getName(), retBean.getName());
assertEquals(bean.getStatusStr(), retBean.getStatusStr());
assertEquals(bean.getPending(), retBean.getPending());
assertEquals(bean.getPendingAge().getTime(), retBean.getPendingAge().getTime());
assertEquals(bean.getType(), retBean.getType());
assertEquals(bean.getLogToken(), retBean.getLogToken());
assertEquals(bean.getTransition(), retBean.getTransition());
assertEquals(bean.getRetries(), retBean.getRetries());
assertEquals(bean.getTrackerUri(), retBean.getTrackerUri());
assertEquals(bean.getUserRetryCount(), retBean.getUserRetryCount());
assertEquals(bean.getUserRetryMax(), retBean.getUserRetryMax());
assertEquals(bean.getUserRetryInterval(), retBean.getUserRetryInterval());
assertEquals(bean.getExternalId(), retBean.getExternalId());
assertEquals(bean.getExternalStatus(), retBean.getExternalStatus());
assertEquals(bean.getExternalChildIDs(), retBean.getExternalChildIDs());
assertEquals(bean.getStartTime().getTime(), retBean.getStartTime().getTime());
assertEquals(bean.getEndTime().getTime(), retBean.getEndTime().getTime());
assertEquals(bean.getErrorCode(), retBean.getErrorCode());
assertEquals(bean.getErrorMessage(), retBean.getErrorMessage());
assertEquals(bean.getConf(), retBean.getConf());
assertEquals(bean.getData(), retBean.getData());
assertEquals(bean.getStats(), retBean.getStats());
assertNull(retBean.getSlaXml());
// GET_ACTION_COMPLETED
retBean =
WorkflowActionQueryExecutor.getInstance()
.get(WorkflowActionQuery.GET_ACTION_COMPLETED, bean.getId());
assertEquals(bean.getId(), retBean.getId());
assertEquals(bean.getJobId(), retBean.getJobId());
assertEquals(bean.getStatusStr(), retBean.getStatusStr());
assertEquals(bean.getType(), retBean.getType());
assertEquals(bean.getLogToken(), retBean.getLogToken());
assertNull(retBean.getSlaXml());
assertNull(retBean.getConf());
assertNull(retBean.getData());
assertNull(retBean.getStats());
assertNull(retBean.getExternalChildIDs());
// GET_ACTION (entire obj)
retBean =
WorkflowActionQueryExecutor.getInstance().get(WorkflowActionQuery.GET_ACTION, bean.getId());
assertEquals(bean.getId(), retBean.getId());
assertEquals(bean.getJobId(), retBean.getJobId());
assertEquals(bean.getName(), retBean.getName());
assertEquals(bean.getStatusStr(), retBean.getStatusStr());
assertEquals(bean.getPending(), retBean.getPending());
assertEquals(bean.getPendingAge().getTime(), retBean.getPendingAge().getTime());
assertEquals(bean.getType(), retBean.getType());
assertEquals(bean.getLogToken(), retBean.getLogToken());
assertEquals(bean.getTransition(), retBean.getTransition());
assertEquals(bean.getRetries(), retBean.getRetries());
assertEquals(bean.getUserRetryCount(), retBean.getUserRetryCount());
assertEquals(bean.getUserRetryMax(), retBean.getUserRetryMax());
assertEquals(bean.getUserRetryInterval(), retBean.getUserRetryInterval());
assertEquals(bean.getStartTime().getTime(), retBean.getStartTime().getTime());
assertEquals(bean.getEndTime().getTime(), retBean.getEndTime().getTime());
assertEquals(bean.getCreatedTime().getTime(), retBean.getCreatedTime().getTime());
assertEquals(bean.getLastCheckTime().getTime(), retBean.getLastCheckTime().getTime());
assertEquals(bean.getErrorCode(), retBean.getErrorCode());
assertEquals(bean.getErrorMessage(), retBean.getErrorMessage());
assertEquals(bean.getExecutionPath(), retBean.getExecutionPath());
assertEquals(bean.getSignalValue(), retBean.getSignalValue());
assertEquals(bean.getCred(), retBean.getCred());
assertEquals(bean.getConf(), retBean.getConf());
assertEquals(bean.getSlaXml(), retBean.getSlaXml());
assertEquals(bean.getData(), retBean.getData());
assertEquals(bean.getStats(), retBean.getStats());
assertEquals(bean.getExternalChildIDs(), retBean.getExternalChildIDs());
}
/**
* Provides functionality to test non transient failures.
*
* @param errorType the error type. (start.non-transient, end.non-transient)
* @param expStatus1 expected status. (START_MANUAL, END_MANUAL)
* @param expErrorMsg expected error message.
* @throws Exception
*/
private void _testNonTransient(
String errorType, WorkflowActionBean.Status expStatus1, String expErrorMsg) throws Exception {
String workflowPath = getTestCaseFileUri("workflow.xml");
Reader reader = IOUtils.getResourceAsReader("wf-ext-schema-valid.xml", -1);
Writer writer = new FileWriter(new File(getTestCaseDir(), "workflow.xml"));
IOUtils.copyCharStream(reader, writer);
final DagEngine engine = new DagEngine("u");
Configuration conf = new XConfiguration();
conf.set(OozieClient.APP_PATH, workflowPath);
conf.set(OozieClient.USER_NAME, getTestUser());
conf.set(OozieClient.LOG_TOKEN, "t");
conf.set("signal-value", "OK");
conf.set("external-status", "ok");
conf.set("error", errorType);
final String jobId = engine.submitJob(conf, true);
waitFor(
5000,
new Predicate() {
public boolean evaluate() throws Exception {
return (engine.getJob(jobId).getStatus() == WorkflowJob.Status.SUSPENDED);
}
});
final WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
store.beginTrx();
List<WorkflowActionBean> actions = store.getActionsForWorkflow(jobId, true);
int n = actions.size();
WorkflowActionBean action = actions.get(n - 1);
assertEquals("TEST_ERROR", action.getErrorCode());
assertEquals(expErrorMsg, action.getErrorMessage());
assertEquals(expStatus1, action.getStatus());
assertTrue(action.isPending() == false);
assertTrue(engine.getJob(jobId).getStatus() == WorkflowJob.Status.SUSPENDED);
String actionConf = action.getConf();
String fixedActionConf = actionConf.replaceAll(errorType, "none");
action.setConf(fixedActionConf);
store.updateAction(action);
store.commitTrx();
store.closeTrx();
engine.resume(jobId);
waitFor(
5000,
new Predicate() {
public boolean evaluate() throws Exception {
return (engine.getJob(jobId).getStatus() == WorkflowJob.Status.SUCCEEDED);
}
});
assertEquals(WorkflowJob.Status.SUCCEEDED, engine.getJob(jobId).getStatus());
final WorkflowStore store2 = Services.get().get(WorkflowStoreService.class).create();
store2.beginTrx();
actions = store2.getActionsForWorkflow(jobId, false);
action = actions.get(0);
assertEquals(null, action.getErrorCode());
assertEquals(null, action.getErrorMessage());
assertEquals(WorkflowActionBean.Status.OK, action.getStatus());
store2.commitTrx();
store2.closeTrx();
}
