Ejemplo n.º 1
0
 @Override
 protected void verifyPrecondition() throws CommandException, PreconditionException {
   if ((wfAction == null) || (wfAction.isComplete() && wfAction.isPending())) {
     if (wfJob.getStatus() != WorkflowJob.Status.RUNNING
         && wfJob.getStatus() != WorkflowJob.Status.PREP) {
       throw new PreconditionException(ErrorCode.E0813, wfJob.getStatusStr());
     }
   } else {
     throw new PreconditionException(
         ErrorCode.E0814, actionId, wfAction.getStatusStr(), wfAction.isPending());
   }
 }
Ejemplo n.º 2
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());
  }
Ejemplo n.º 3
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;
  }
Ejemplo n.º 4
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;
  }
Ejemplo n.º 5
0
  /**
   * Provides functionality to test non transient failures and coordinator action update
   *
   * @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 _testNonTransientWithCoordActionUpdate(
      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, false);

    final JPAService jpaService = Services.get().get(JPAService.class);
    final CoordinatorJobBean coordJob =
        addRecordToCoordJobTable(CoordinatorJob.Status.RUNNING, false, false);
    CoordinatorActionBean coordAction =
        addRecordToCoordActionTable(
            coordJob.getId(),
            1,
            CoordinatorAction.Status.RUNNING,
            "coord-action-get.xml",
            jobId,
            "RUNNING",
            0);

    engine.start(jobId);

    waitFor(
        5000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            return (engine.getJob(jobId).getStatus() == WorkflowJob.Status.SUSPENDED);
          }
        });

    assertNotNull(jpaService);
    WorkflowJobGetJPAExecutor wfGetCmd = new WorkflowJobGetJPAExecutor(jobId);
    WorkflowJobBean job = jpaService.execute(wfGetCmd);
    WorkflowActionsGetForJobJPAExecutor actionsGetExe =
        new WorkflowActionsGetForJobJPAExecutor(jobId);
    List<WorkflowActionBean> actionsList = jpaService.execute(actionsGetExe);

    int n = actionsList.size();
    WorkflowActionBean action = actionsList.get(n - 1);
    assertEquals("TEST_ERROR", action.getErrorCode());
    assertEquals(expErrorMsg, action.getErrorMessage());
    assertEquals(expStatus1, action.getStatus());
    assertFalse(action.isPending());

    assertEquals(WorkflowJob.Status.SUSPENDED, job.getStatus());

    waitFor(
        5000,
        new Predicate() {
          public boolean evaluate() throws Exception {
            CoordinatorActionBean coordAction2 =
                jpaService.execute(new CoordActionGetForExternalIdJPAExecutor(jobId));
            return coordAction2.getStatus().equals(CoordinatorAction.Status.SUSPENDED);
          }
        });

    coordAction = jpaService.execute(new CoordActionGetForExternalIdJPAExecutor(jobId));
    assertEquals(CoordinatorAction.Status.SUSPENDED, coordAction.getStatus());
  }