Example #1
0
  /**
   * Provides functionality to test for set*Data calls not being made by the Action Handler.
   *
   * @param avoidParam set*Data function call to avoid.
   * @param expActionErrorCode the expected action error code.
   * @throws Exception
   */
  private void _testDataNotSet(String avoidParam, String expActionErrorCode) 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("external-status", "ok");
    conf.set("signal-value", "based_on_action_status");
    conf.set(avoidParam, "true");

    final String jobId = engine.submitJob(conf, true);

    final WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
    store.beginTrx();
    Thread.sleep(2000);

    waitFor(
        5000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            WorkflowJobBean bean = store.getWorkflow(jobId, false);
            return (bean.getWorkflowInstance().getStatus() == WorkflowInstance.Status.FAILED);
          }
        });
    store.commitTrx();
    store.closeTrx();

    final WorkflowStore store2 = Services.get().get(WorkflowStoreService.class).create();
    store2.beginTrx();
    assertEquals(
        WorkflowInstance.Status.FAILED,
        store2.getWorkflow(jobId, false).getWorkflowInstance().getStatus());
    assertEquals(WorkflowJob.Status.FAILED, engine.getJob(jobId).getStatus());

    List<WorkflowActionBean> actions = store2.getActionsForWorkflow(jobId, false);
    WorkflowActionBean action = null;
    for (WorkflowActionBean bean : actions) {
      if (bean.getType().equals("test")) {
        action = bean;
        break;
      }
    }
    assertNotNull(action);
    assertEquals(expActionErrorCode, action.getErrorCode());
    store2.commitTrx();
    store2.closeTrx();
  }
Example #2
0
  /**
   * Provides functionality to test user retry
   *
   * @param errorType the error type. (start.non-transient, end.non-transient)
   * @param externalStatus the external status to set.
   * @param signalValue the signal value to set.
   * @throws Exception
   */
  private void _testErrorWithUserRetry(String errorType, String externalStatus, String signalValue)
      throws Exception {
    String workflowPath = getTestCaseFileUri("workflow.xml");
    Reader reader = IOUtils.getResourceAsReader("wf-ext-schema-valid-user-retry.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("error", errorType);
    conf.set("external-status", externalStatus);
    conf.set("signal-value", signalValue);

    final String jobId = engine.submitJob(conf, true);

    final JPAService jpaService = Services.get().get(JPAService.class);
    final WorkflowJobGetJPAExecutor wfJobGetCmd = new WorkflowJobGetJPAExecutor(jobId);

    final WorkflowActionsGetForJobJPAExecutor actionsGetExecutor =
        new WorkflowActionsGetForJobJPAExecutor(jobId);
    waitFor(
        5000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            List<WorkflowActionBean> actions = jpaService.execute(actionsGetExecutor);
            WorkflowActionBean action = null;
            for (WorkflowActionBean bean : actions) {
              if (bean.getType().equals("test")) {
                action = bean;
                break;
              }
            }
            return (action != null && action.getUserRetryCount() == 2);
          }
        });

    List<WorkflowActionBean> actions = jpaService.execute(actionsGetExecutor);
    WorkflowActionBean action = null;
    for (WorkflowActionBean bean : actions) {
      if (bean.getType().equals("test")) {
        action = bean;
        break;
      }
    }
    assertNotNull(action);
    assertEquals(2, action.getUserRetryCount());
  }
Example #3
0
  /**
   * Provides functionality to test kill node message
   *
   * @throws Exception
   */
  public void testKillNodeErrorMessage() throws Exception {
    String workflowPath = getTestCaseFileUri("workflow.xml");
    Reader reader = IOUtils.getResourceAsReader("wf-test-kill-node-message.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("error", "end.error");
    conf.set("external-status", "FAILED/KILLED");
    conf.set("signal-value", "fail");

    final String jobId = engine.submitJob(conf, true);

    final JPAService jpaService = Services.get().get(JPAService.class);
    final WorkflowJobGetJPAExecutor wfJobGetCmd = new WorkflowJobGetJPAExecutor(jobId);

    waitFor(
        50000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            WorkflowJobBean job = jpaService.execute(wfJobGetCmd);
            return (job.getWorkflowInstance().getStatus() == WorkflowInstance.Status.KILLED);
          }
        });

    WorkflowJobBean job = jpaService.execute(wfJobGetCmd);
    assertEquals(WorkflowJob.Status.KILLED, job.getStatus());

    WorkflowActionsGetForJobJPAExecutor wfActionsGetCmd =
        new WorkflowActionsGetForJobJPAExecutor(jobId);
    List<WorkflowActionBean> actions = jpaService.execute(wfActionsGetCmd);

    int n = actions.size();
    WorkflowActionBean action = null;
    for (WorkflowActionBean bean : actions) {
      if (bean.getType().equals("test")) {
        action = bean;
        break;
      }
    }
    assertNotNull(action);
    assertEquals("TEST_ERROR", action.getErrorCode());
    assertEquals("end", action.getErrorMessage());
    assertEquals(WorkflowAction.Status.ERROR, action.getStatus());
  }
Example #4
0
  protected Void call(WorkflowStore store) throws StoreException, CommandException {
    WorkflowJobBean workflow = store.getWorkflow(jobId, false);
    setLogInfo(workflow);
    WorkflowActionBean action = store.getAction(id, false);
    setLogInfo(action);
    if (action.isPending()
        && (action.getStatus() == WorkflowActionBean.Status.DONE
            || action.getStatus() == WorkflowActionBean.Status.END_RETRY
            || action.getStatus() == WorkflowActionBean.Status.END_MANUAL)) {
      if (workflow.getStatus() == WorkflowJob.Status.RUNNING) {

        ActionExecutor executor =
            Services.get().get(ActionService.class).getExecutor(action.getType());
        Configuration conf = workflow.getWorkflowInstance().getConf();
        int maxRetries = conf.getInt(OozieClient.ACTION_MAX_RETRIES, executor.getMaxRetries());
        long retryInterval =
            conf.getLong(OozieClient.ACTION_RETRY_INTERVAL, executor.getRetryInterval());
        executor.setMaxRetries(maxRetries);
        executor.setRetryInterval(retryInterval);

        if (executor != null) {
          boolean isRetry = false;
          if (action.getStatus() == WorkflowActionBean.Status.END_RETRY
              || action.getStatus() == WorkflowActionBean.Status.END_MANUAL) {
            isRetry = true;
          }
          ActionExecutorContext context =
              new ActionCommand.ActionExecutorContext(workflow, action, isRetry);
          try {

            XLog.getLog(getClass())
                .debug(
                    "End, name [{0}] type [{1}] status[{2}] external status [{3}] signal value [{4}]",
                    action.getName(),
                    action.getType(),
                    action.getStatus(),
                    action.getExternalStatus(),
                    action.getSignalValue());
            WorkflowInstance wfInstance = workflow.getWorkflowInstance();
            DagELFunctions.setActionInfo(wfInstance, action);
            workflow.setWorkflowInstance(wfInstance);
            incrActionCounter(action.getType(), 1);

            Instrumentation.Cron cron = new Instrumentation.Cron();
            cron.start();
            executor.end(context, action);
            cron.stop();
            addActionCron(action.getType(), cron);

            if (!context.isEnded()) {
              XLog.getLog(getClass())
                  .warn(
                      XLog.OPS,
                      "Action Ended, ActionExecutor [{0}] must call setEndData()",
                      executor.getType());
              action.setErrorInfo(
                  END_DATA_MISSING, "Execution Ended, but End Data Missing from Action");
              failJob(context);
              store.updateAction(action);
              store.updateWorkflow(workflow);
              return null;
            }
            action.setRetries(0);
            action.setEndTime(new Date());
            store.updateAction(action);
            store.updateWorkflow(workflow);
            Status slaStatus = null;
            switch (action.getStatus()) {
              case OK:
                slaStatus = Status.SUCCEEDED;
                break;
              case KILLED:
                slaStatus = Status.KILLED;
                break;
              case FAILED:
                slaStatus = Status.FAILED;
                break;
              case ERROR:
                XLog.getLog(getClass()).info("ERROR is considered as FAILED for SLA");
                slaStatus = Status.KILLED;
                break;
              default: // TODO: What will happen for other Action
                // status
                slaStatus = Status.FAILED;
                break;
            }
            SLADbOperations.writeStausEvent(
                action.getSlaXml(), action.getId(), store, slaStatus, SlaAppType.WORKFLOW_ACTION);
            queueCallable(new NotificationCommand(workflow, action));
            XLog.getLog(getClass())
                .debug(
                    "Queuing commands for action "
                        + id
                        + " status "
                        + action.getStatus()
                        + ", Set pending="
                        + action.getPending());
            queueCallable(new SignalCommand(workflow.getId(), id));
          } catch (ActionExecutorException ex) {
            XLog.getLog(getClass())
                .warn(
                    "Error ending action [{0}]. ErrorType [{1}], ErrorCode [{2}], Message [{3}]",
                    action.getName(), ex.getErrorType(), ex.getErrorCode(), ex.getMessage());
            action.setErrorInfo(ex.getErrorCode(), ex.getMessage());
            action.setEndTime(null);
            switch (ex.getErrorType()) {
              case TRANSIENT:
                if (!handleTransient(context, executor, WorkflowAction.Status.END_RETRY)) {
                  handleNonTransient(context, executor, WorkflowAction.Status.END_MANUAL);
                  action.setPendingAge(new Date());
                  action.setRetries(0);
                }
                action.setEndTime(null);
                break;
              case NON_TRANSIENT:
                handleNonTransient(context, executor, WorkflowAction.Status.END_MANUAL);
                action.setEndTime(null);
                break;
              case ERROR:
                handleError(context, executor, COULD_NOT_END, false, WorkflowAction.Status.ERROR);
                queueCallable(new SignalCommand(workflow.getId(), id));
                break;
              case FAILED:
                failJob(context);
                break;
            }
            store.updateAction(action);
            store.updateWorkflow(workflow);
          }
        } else {
          throw new CommandException(ErrorCode.E0802, action.getType());
        }
      } else {
        XLog.getLog(getClass())
            .warn(
                "Job state is not {0}. Skipping ActionEnd Execution",
                WorkflowJob.Status.RUNNING.toString());
      }
    } else {
      XLog.getLog(getClass())
          .debug(
              "Action pending={0}, status={1}. Skipping ActionEnd Execution",
              action.getPending(), action.getStatusStr());
    }
    return null;
  }
  public void testGetUpdateQuery() throws Exception {
    EntityManager em = jpaService.getEntityManager();
    WorkflowJobBean job =
        this.addRecordToWfJobTable(WorkflowJob.Status.RUNNING, WorkflowInstance.Status.RUNNING);
    WorkflowActionBean bean =
        addRecordToWfActionTable(job.getId(), "1", WorkflowAction.Status.PREP);

    // UPDATE_ACTION
    Query query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION, bean, em);
    assertEquals(query.getParameterValue("conf"), bean.getConfBlob());
    assertEquals(query.getParameterValue("consoleUrl"), bean.getConsoleUrl());
    assertEquals(query.getParameterValue("data"), bean.getDataBlob());
    assertEquals(query.getParameterValue("stats"), bean.getStatsBlob());
    assertEquals(query.getParameterValue("externalChildIDs"), bean.getExternalChildIDsBlob());
    assertEquals(query.getParameterValue("errorCode"), bean.getErrorCode());
    assertEquals(query.getParameterValue("errorMessage"), bean.getErrorMessage());
    assertEquals(query.getParameterValue("externalId"), bean.getExternalId());
    assertEquals(query.getParameterValue("externalStatus"), bean.getExternalStatus());
    assertEquals(query.getParameterValue("name"), bean.getName());
    assertEquals(query.getParameterValue("cred"), bean.getCred());
    assertEquals(query.getParameterValue("retries"), bean.getRetries());
    assertEquals(query.getParameterValue("trackerUri"), bean.getTrackerUri());
    assertEquals(query.getParameterValue("transition"), bean.getTransition());
    assertEquals(query.getParameterValue("type"), bean.getType());
    assertEquals(query.getParameterValue("endTime"), bean.getEndTimestamp());
    assertEquals(query.getParameterValue("executionPath"), bean.getExecutionPath());
    assertEquals(query.getParameterValue("lastCheckTime"), bean.getLastCheckTimestamp());
    assertEquals(query.getParameterValue("logToken"), bean.getLogToken());
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAge());
    assertEquals(query.getParameterValue("signalValue"), bean.getSignalValue());
    assertEquals(query.getParameterValue("slaXml"), bean.getSlaXmlBlob());
    assertEquals(query.getParameterValue("startTime"), bean.getStartTimestamp());
    assertEquals(query.getParameterValue("status"), bean.getStatus().toString());
    assertEquals(query.getParameterValue("wfId"), bean.getWfId());
    assertEquals(query.getParameterValue("id"), bean.getId());

    // UPDATE_ACTION_FOR_LAST_CHECKED_TIME
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION, bean, em);
    assertEquals(query.getParameterValue("lastCheckTime"), bean.getLastCheckTimestamp());
    assertEquals(query.getParameterValue("id"), bean.getId());

    // UPDATE_ACTION_PENDING
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION_PENDING, bean, em);
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAgeTimestamp());
    assertEquals(query.getParameterValue("id"), bean.getId());

    // UPDATE_ACTION_STATUS_PENDING
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION_STATUS_PENDING, bean, em);
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAgeTimestamp());
    assertEquals(query.getParameterValue("status"), bean.getStatus().toString());
    assertEquals(query.getParameterValue("id"), bean.getId());
    // UPDATE_ACTION_PENDING_TRANS
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION_PENDING_TRANS, bean, em);
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAgeTimestamp());
    assertEquals(query.getParameterValue("transition"), bean.getTransition());
    assertEquals(query.getParameterValue("id"), bean.getId());
    // UPDATE_ACTION_PENDING_TRANS_ERROR
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION_PENDING_TRANS_ERROR, bean, em);
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAgeTimestamp());
    assertEquals(query.getParameterValue("transition"), bean.getTransition());
    assertEquals(query.getParameterValue("errorCode"), bean.getErrorCode());
    assertEquals(query.getParameterValue("errorMessage"), bean.getErrorMessage());
    assertEquals(query.getParameterValue("id"), bean.getId());
    // UPDATE_ACTION_START
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION_START, bean, em);
    assertEquals(query.getParameterValue("startTime"), bean.getStartTimestamp());
    assertEquals(query.getParameterValue("externalChildIDs"), bean.getExternalChildIDsBlob());
    assertEquals(query.getParameterValue("conf"), bean.getConfBlob());
    assertEquals(query.getParameterValue("errorCode"), bean.getErrorCode());
    assertEquals(query.getParameterValue("errorMessage"), bean.getErrorMessage());
    assertEquals(query.getParameterValue("externalId"), bean.getExternalId());
    assertEquals(query.getParameterValue("trackerUri"), bean.getTrackerUri());
    assertEquals(query.getParameterValue("consoleUrl"), bean.getConsoleUrl());
    assertEquals(query.getParameterValue("lastCheckTime"), bean.getLastCheckTimestamp());
    assertEquals(query.getParameterValue("status"), bean.getStatus().toString());
    assertEquals(query.getParameterValue("externalStatus"), bean.getExternalStatus());
    assertEquals(query.getParameterValue("data"), bean.getDataBlob());
    assertEquals(query.getParameterValue("retries"), bean.getRetries());
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAgeTimestamp());
    assertEquals(query.getParameterValue("userRetryCount"), bean.getUserRetryCount());
    assertEquals(query.getParameterValue("id"), bean.getId());
    // UPDATE_ACTION_CHECK
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION_CHECK, bean, em);
    assertEquals(query.getParameterValue("externalChildIDs"), bean.getExternalChildIDsBlob());
    assertEquals(query.getParameterValue("externalStatus"), bean.getExternalStatus());
    assertEquals(query.getParameterValue("status"), bean.getStatus().toString());
    assertEquals(query.getParameterValue("data"), bean.getDataBlob());
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("errorCode"), bean.getErrorCode());
    assertEquals(query.getParameterValue("errorMessage"), bean.getErrorMessage());
    assertEquals(query.getParameterValue("lastCheckTime"), bean.getLastCheckTimestamp());
    assertEquals(query.getParameterValue("retries"), bean.getRetries());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAgeTimestamp());
    assertEquals(query.getParameterValue("startTime"), bean.getStartTimestamp());
    assertEquals(query.getParameterValue("stats"), bean.getStatsBlob());
    assertEquals(query.getParameterValue("userRetryCount"), bean.getUserRetryCount());
    assertEquals(query.getParameterValue("id"), bean.getId());
    // UPDATE_ACTION_END
    query =
        WorkflowActionQueryExecutor.getInstance()
            .getUpdateQuery(WorkflowActionQuery.UPDATE_ACTION_END, bean, em);
    assertEquals(query.getParameterValue("errorCode"), bean.getErrorCode());
    assertEquals(query.getParameterValue("errorMessage"), bean.getErrorMessage());
    assertEquals(query.getParameterValue("retries"), bean.getRetries());
    assertEquals(query.getParameterValue("endTime"), bean.getEndTimestamp());
    assertEquals(query.getParameterValue("status"), bean.getStatus().toString());
    assertEquals(query.getParameterValue("retries"), bean.getRetries());
    assertEquals(query.getParameterValue("pending"), bean.getPending());
    assertEquals(query.getParameterValue("pendingAge"), bean.getPendingAgeTimestamp());
    assertEquals(query.getParameterValue("signalValue"), bean.getSignalValue());
    assertEquals(query.getParameterValue("userRetryCount"), bean.getUserRetryCount());
    assertEquals(query.getParameterValue("externalStatus"), bean.getExternalStatus());
    assertEquals(query.getParameterValue("stats"), bean.getStatsBlob());
    assertEquals(query.getParameterValue("id"), bean.getId());
    em.close();
  }
  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());
  }
Example #7
0
  public void testFsFailover() throws Exception {
    Reader reader = IOUtils.getResourceAsReader("failover-fs-wf.xml", -1);
    Writer writer = new FileWriter(getTestCaseDir() + "/workflow.xml");
    IOUtils.copyCharStream(reader, writer);

    final OozieClient wfClient = LocalOozie.getClient();
    Properties conf = wfClient.createConfiguration();
    conf.setProperty(OozieClient.APP_PATH, getTestCaseDir());
    conf.setProperty(OozieClient.USER_NAME, getTestUser());
    conf.setProperty(OozieClient.GROUP_NAME, getTestGroup());
    injectKerberosInfo(conf);
    final Path source = new Path(getFsTestCaseDir(), "fsfailover-source");
    getFileSystem().mkdirs(source);
    final Path target = new Path(getFsTestCaseDir().toString(), "fsfailover-target");
    conf.setProperty("source", source.toString());
    conf.setProperty("target", target.toUri().getPath());

    setSystemProperty(FaultInjection.FAULT_INJECTION, "true");
    setSystemProperty(SkipCommitFaultInjection.ACTION_FAILOVER_FAULT_INJECTION, "true");

    final String jobId1 = wfClient.submit(conf);
    wfClient.start(jobId1);

    waitFor(
        10 * 1000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            return getFileSystem().exists(target);
          }
        });
    assertTrue(getFileSystem().exists(target));

    waitFor(
        10 * 1000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            return FaultInjection.isActive("org.apache.oozie.command.SkipCommitFaultInjection");
          }
        });
    assertFalse(FaultInjection.isActive("org.apache.oozie.command.SkipCommitFaultInjection"));

    assertEquals(WorkflowJob.Status.RUNNING, wfClient.getJobInfo(jobId1).getStatus());

    WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();

    List<WorkflowActionBean> actions = store.getActionsForWorkflow(jobId1, false);
    WorkflowActionBean action = actions.get(0);
    assertEquals(WorkflowAction.Status.PREP, action.getStatus());

    setSystemProperty(FaultInjection.FAULT_INJECTION, "false");
    setSystemProperty(SkipCommitFaultInjection.ACTION_FAILOVER_FAULT_INJECTION, "false");

    ActionStartCommand actionStartCommand =
        new ActionStartCommand(action.getId(), action.getType());
    actionStartCommand.call();

    store = Services.get().get(WorkflowStoreService.class).create();
    actions = store.getActionsForWorkflow(jobId1, false);
    action = actions.get(0);
    assertEquals(WorkflowAction.Status.DONE, action.getStatus());

    waitFor(
        5 * 1000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            return wfClient.getJobInfo(jobId1).getStatus() == WorkflowJob.Status.SUCCEEDED;
          }
        });
    assertEquals(WorkflowJob.Status.SUCCEEDED, wfClient.getJobInfo(jobId1).getStatus());

    final String jobId2 = wfClient.submit(conf);

    wfClient.start(jobId2);
    waitFor(
        10 * 1000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            return wfClient.getJobInfo(jobId2).getStatus() == WorkflowJob.Status.KILLED;
          }
        });
    assertEquals(WorkflowJob.Status.KILLED, wfClient.getJobInfo(jobId2).getStatus());
  }
Example #8
0
  @Override
  protected Void execute() throws CommandException {
    LOG.debug("STARTED SignalCommand for jobid=" + jobId + ", actionId=" + actionId);
    WorkflowInstance workflowInstance = wfJob.getWorkflowInstance();
    workflowInstance.setTransientVar(WorkflowStoreService.WORKFLOW_BEAN, wfJob);
    boolean completed = false;
    boolean skipAction = false;
    if (wfAction == null) {
      if (wfJob.getStatus() == WorkflowJob.Status.PREP) {
        try {
          completed = workflowInstance.start();
        } catch (WorkflowException e) {
          throw new CommandException(e);
        }
        wfJob.setStatus(WorkflowJob.Status.RUNNING);
        wfJob.setStartTime(new Date());
        wfJob.setWorkflowInstance(workflowInstance);
        // 1. Add SLA status event for WF-JOB with status STARTED
        // 2. Add SLA registration events for all WF_ACTIONS
        SLADbXOperations.writeStausEvent(
            wfJob.getSlaXml(), jobId, Status.STARTED, SlaAppType.WORKFLOW_JOB);
        writeSLARegistrationForAllActions(
            workflowInstance.getApp().getDefinition(),
            wfJob.getUser(),
            wfJob.getGroup(),
            wfJob.getConf());
        queue(new NotificationXCommand(wfJob));
      } else {
        throw new CommandException(ErrorCode.E0801, wfJob.getId());
      }
    } else {
      String skipVar =
          workflowInstance.getVar(
              wfAction.getName() + WorkflowInstance.NODE_VAR_SEPARATOR + ReRunCommand.TO_SKIP);
      if (skipVar != null) {
        skipAction = skipVar.equals("true");
      }
      try {
        completed = workflowInstance.signal(wfAction.getExecutionPath(), wfAction.getSignalValue());
      } catch (WorkflowException e) {
        throw new CommandException(e);
      }
      wfJob.setWorkflowInstance(workflowInstance);
      wfAction.resetPending();
      if (!skipAction) {
        wfAction.setTransition(workflowInstance.getTransition(wfAction.getName()));
      }
      try {
        jpaService.execute(new WorkflowActionUpdateJPAExecutor(wfAction));
      } catch (JPAExecutorException je) {
        throw new CommandException(je);
      }
    }

    if (completed) {
      try {
        for (String actionToKillId : WorkflowStoreService.getActionsToKill(workflowInstance)) {
          WorkflowActionBean actionToKill;

          actionToKill = jpaService.execute(new WorkflowActionGetJPAExecutor(actionToKillId));

          actionToKill.setPending();
          actionToKill.setStatus(WorkflowActionBean.Status.KILLED);
          jpaService.execute(new WorkflowActionUpdateJPAExecutor(actionToKill));
          queue(new ActionKillXCommand(actionToKill.getId(), actionToKill.getType()));
        }

        for (String actionToFailId : WorkflowStoreService.getActionsToFail(workflowInstance)) {
          WorkflowActionBean actionToFail =
              jpaService.execute(new WorkflowActionGetJPAExecutor(actionToFailId));
          actionToFail.resetPending();
          actionToFail.setStatus(WorkflowActionBean.Status.FAILED);
          SLADbXOperations.writeStausEvent(
              wfAction.getSlaXml(), wfAction.getId(), Status.FAILED, SlaAppType.WORKFLOW_ACTION);
          jpaService.execute(new WorkflowActionUpdateJPAExecutor(actionToFail));
        }
      } catch (JPAExecutorException je) {
        throw new CommandException(je);
      }

      wfJob.setStatus(WorkflowJob.Status.valueOf(workflowInstance.getStatus().toString()));
      wfJob.setEndTime(new Date());
      wfJob.setWorkflowInstance(workflowInstance);
      Status slaStatus = Status.SUCCEEDED;
      switch (wfJob.getStatus()) {
        case SUCCEEDED:
          slaStatus = Status.SUCCEEDED;
          break;
        case KILLED:
          slaStatus = Status.KILLED;
          break;
        case FAILED:
          slaStatus = Status.FAILED;
          break;
        default: // TODO SUSPENDED
          break;
      }
      SLADbXOperations.writeStausEvent(
          wfJob.getSlaXml(), jobId, slaStatus, SlaAppType.WORKFLOW_JOB);
      queue(new NotificationXCommand(wfJob));
      if (wfJob.getStatus() == WorkflowJob.Status.SUCCEEDED) {
        InstrumentUtils.incrJobCounter(INSTR_SUCCEEDED_JOBS_COUNTER_NAME, 1, getInstrumentation());
      }
    } else {
      for (WorkflowActionBean newAction :
          WorkflowStoreService.getStartedActions(workflowInstance)) {
        String skipVar =
            workflowInstance.getVar(
                newAction.getName() + WorkflowInstance.NODE_VAR_SEPARATOR + ReRunCommand.TO_SKIP);
        boolean skipNewAction = false;
        if (skipVar != null) {
          skipNewAction = skipVar.equals("true");
        }
        try {
          if (skipNewAction) {
            WorkflowActionBean oldAction;

            oldAction = jpaService.execute(new WorkflowActionGetJPAExecutor(newAction.getId()));

            oldAction.setPending();
            jpaService.execute(new WorkflowActionUpdateJPAExecutor(oldAction));

            queue(new SignalXCommand(jobId, oldAction.getId()));
          } else {
            newAction.setPending();
            String actionSlaXml =
                getActionSLAXml(
                    newAction.getName(),
                    workflowInstance.getApp().getDefinition(),
                    wfJob.getConf());
            newAction.setSlaXml(actionSlaXml);
            jpaService.execute(new WorkflowActionInsertJPAExecutor(newAction));
            LOG.debug(
                "SignalXCommand: Name: "
                    + newAction.getName()
                    + ", Id: "
                    + newAction.getId()
                    + ", Authcode:"
                    + newAction.getCred());
            queue(new ActionStartXCommand(newAction.getId(), newAction.getType()));
          }
        } catch (JPAExecutorException je) {
          throw new CommandException(je);
        }
      }
    }

    try {
      jpaService.execute(new WorkflowJobUpdateJPAExecutor(wfJob));
    } catch (JPAExecutorException je) {
      throw new CommandException(je);
    }
    XLog.getLog(getClass())
        .debug(
            "Updated the workflow status to "
                + wfJob.getId()
                + "  status ="
                + wfJob.getStatusStr());
    if (wfJob.getStatus() != WorkflowJob.Status.RUNNING
        && wfJob.getStatus() != WorkflowJob.Status.SUSPENDED) {
      // update coordinator action
      new CoordActionUpdateXCommand(wfJob).call();
      new WfEndXCommand(wfJob).call(); // To delete the WF temp dir
    }
    LOG.debug("ENDED SignalCommand for jobid=" + jobId + ", actionId=" + actionId);
    return null;
  }
Example #9
0
  /**
   * Provides functionality to test transient failures.
   *
   * @param errorType the error type. (start.transient, end.transient)
   * @param expStatus1 expected status after the first step (START_RETRY, END_RETRY)
   * @param expStatus2 expected status after the second step (START_MANUAL, END_MANUAL)
   * @param expErrorMsg the expected error message.
   * @throws Exception
   */
  private void _testTransient(
      String errorType,
      WorkflowActionBean.Status expStatus1,
      final WorkflowActionBean.Status expStatus2,
      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 int maxRetries = 2;
    final int retryInterval = 10;

    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);
    conf.setInt(OozieClient.ACTION_MAX_RETRIES, maxRetries);
    conf.setInt(OozieClient.ACTION_RETRY_INTERVAL, retryInterval);

    final String jobId = engine.submitJob(conf, true);

    int retryCount = 1;
    WorkflowActionBean.Status expectedStatus = expStatus1;
    int expectedRetryCount = 2;

    Thread.sleep(20000);
    String aId = null;
    final WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
    store.beginTrx();
    while (retryCount <= maxRetries) {
      List<WorkflowActionBean> actions = store.getActionsForWorkflow(jobId, false);
      WorkflowActionBean action = null;
      for (WorkflowActionBean bean : actions) {
        if (bean.getType().equals("test")) {
          action = bean;
          break;
        }
      }
      assertNotNull(action);
      aId = action.getId();
      assertEquals(expectedStatus, action.getStatus());
      assertEquals(expectedRetryCount, action.getRetries());
      assertEquals("TEST_ERROR", action.getErrorCode());
      assertEquals(expErrorMsg, action.getErrorMessage());
      if (action.getRetries() == maxRetries) {
        expectedRetryCount = 0;
        expectedStatus = expStatus2;
        break;
      } else {
        expectedRetryCount++;
      }
      Thread.sleep(retryInterval * 1000);
      retryCount++;
    }
    store.commitTrx();
    store.closeTrx();
    Thread.sleep(5000);

    final String actionId = aId;

    waitFor(
        5000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            return (engine.getWorkflowAction(actionId).getStatus() == expStatus2);
          }
        });

    final WorkflowStore store2 = Services.get().get(WorkflowStoreService.class).create();
    store2.beginTrx();
    WorkflowActionBean action = engine.getWorkflowAction(actionId);
    assertEquals("TEST_ERROR", action.getErrorCode());
    assertEquals(expErrorMsg, action.getErrorMessage());
    assertEquals(expStatus2, action.getStatus());
    assertTrue(action.isPending() == false);
    assertEquals(WorkflowJob.Status.SUSPENDED, engine.getJob(jobId).getStatus());
    store2.commitTrx();
    store2.closeTrx();
  }