Search in sources :

Example 11 with SLAEventBean

use of org.apache.oozie.SLAEventBean in project oozie by apache.

the class SignalXCommand method createSLARegistrationForAllActions.

@SuppressWarnings("unchecked")
private void createSLARegistrationForAllActions(String wfXml, String user, String group, String strConf) throws CommandException {
    try {
        Element eWfJob = XmlUtils.parseXml(wfXml);
        Configuration conf = new XConfiguration(new StringReader(strConf));
        for (Element action : (List<Element>) eWfJob.getChildren("action", eWfJob.getNamespace())) {
            Element eSla = XmlUtils.getSLAElement(action);
            if (eSla != null) {
                String slaXml = resolveSla(eSla, conf);
                eSla = XmlUtils.parseXml(slaXml);
                String actionId = Services.get().get(UUIDService.class).generateChildId(jobId, action.getAttributeValue("name") + "");
                SLAEventBean slaEvent = SLADbXOperations.createSlaRegistrationEvent(eSla, actionId, SlaAppType.WORKFLOW_ACTION, user, group);
                if (slaEvent != null) {
                    insertList.add(slaEvent);
                }
            }
        }
    } catch (Exception e) {
        throw new CommandException(ErrorCode.E1007, "workflow:Actions " + jobId, e.getMessage(), e);
    }
}
Also used : XConfiguration(org.apache.oozie.util.XConfiguration) Configuration(org.apache.hadoop.conf.Configuration) XConfiguration(org.apache.oozie.util.XConfiguration) Element(org.jdom.Element) StringReader(java.io.StringReader) List(java.util.List) ArrayList(java.util.ArrayList) UUIDService(org.apache.oozie.service.UUIDService) CommandException(org.apache.oozie.command.CommandException) SLAEventBean(org.apache.oozie.SLAEventBean) JPAExecutorException(org.apache.oozie.executor.jpa.JPAExecutorException) CommandException(org.apache.oozie.command.CommandException) PreconditionException(org.apache.oozie.command.PreconditionException) IOException(java.io.IOException) XException(org.apache.oozie.XException) WorkflowException(org.apache.oozie.workflow.WorkflowException)

Example 12 with SLAEventBean

use of org.apache.oozie.SLAEventBean in project oozie by apache.

the class SignalXCommand method execute.

@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);
    WorkflowJob.Status prevStatus = wfJob.getStatus();
    boolean completed = false, skipAction = false;
    WorkflowActionBean syncAction = null;
    List<WorkflowActionBean> workflowActionBeanListForForked = new ArrayList<WorkflowActionBean>();
    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);
            generateEvent = true;
            // 1. Add SLA status event for WF-JOB with status STARTED
            SLAEventBean slaEvent = SLADbXOperations.createStatusEvent(wfJob.getSlaXml(), jobId, Status.STARTED, SlaAppType.WORKFLOW_JOB);
            if (slaEvent != null) {
                insertList.add(slaEvent);
            }
            // 2. Add SLA registration events for all WF_ACTIONS
            createSLARegistrationForAllActions(workflowInstance.getApp().getDefinition(), wfJob.getUser(), wfJob.getGroup(), wfJob.getConf());
            queue(new WorkflowNotificationXCommand(wfJob));
        } else {
            throw new CommandException(ErrorCode.E0801, wfJob.getId());
        }
    } else {
        WorkflowInstance.Status initialStatus = workflowInstance.getStatus();
        String skipVar = workflowInstance.getVar(wfAction.getName() + WorkflowInstance.NODE_VAR_SEPARATOR + ReRunXCommand.TO_SKIP);
        if (skipVar != null) {
            skipAction = skipVar.equals("true");
        }
        try {
            completed = workflowInstance.signal(wfAction.getExecutionPath(), wfAction.getSignalValue());
        } catch (WorkflowException e) {
            LOG.error("Workflow action failed : " + e.getMessage(), e);
            wfJob.setStatus(WorkflowJob.Status.valueOf(workflowInstance.getStatus().toString()));
            completed = true;
        }
        wfJob.setWorkflowInstance(workflowInstance);
        wfAction.resetPending();
        if (!skipAction) {
            wfAction.setTransition(workflowInstance.getTransition(wfAction.getName()));
            queue(new WorkflowNotificationXCommand(wfJob, wfAction));
        }
        updateList.add(new UpdateEntry<WorkflowActionQuery>(WorkflowActionQuery.UPDATE_ACTION_PENDING_TRANS, wfAction));
        WorkflowInstance.Status endStatus = workflowInstance.getStatus();
        if (endStatus != initialStatus) {
            generateEvent = true;
        }
    }
    if (completed) {
        try {
            for (String actionToKillId : WorkflowStoreService.getActionsToKill(workflowInstance)) {
                WorkflowActionBean actionToKill;
                actionToKill = WorkflowActionQueryExecutor.getInstance().get(WorkflowActionQuery.GET_ACTION_ID_TYPE_LASTCHECK, actionToKillId);
                actionToKill.setPending();
                actionToKill.setStatus(WorkflowActionBean.Status.KILLED);
                updateList.add(new UpdateEntry<WorkflowActionQuery>(WorkflowActionQuery.UPDATE_ACTION_STATUS_PENDING, actionToKill));
                queue(new ActionKillXCommand(actionToKill.getId(), actionToKill.getType()));
            }
            for (String actionToFailId : WorkflowStoreService.getActionsToFail(workflowInstance)) {
                WorkflowActionBean actionToFail = WorkflowActionQueryExecutor.getInstance().get(WorkflowActionQuery.GET_ACTION_FAIL, actionToFailId);
                actionToFail.resetPending();
                actionToFail.setStatus(WorkflowActionBean.Status.FAILED);
                if (wfJobErrorCode != null) {
                    wfJobErrorCode = actionToFail.getErrorCode();
                    wfJobErrorMsg = actionToFail.getErrorMessage();
                }
                queue(new WorkflowNotificationXCommand(wfJob, actionToFail));
                SLAEventBean slaEvent = SLADbXOperations.createStatusEvent(wfAction.getSlaXml(), wfAction.getId(), Status.FAILED, SlaAppType.WORKFLOW_ACTION);
                if (slaEvent != null) {
                    insertList.add(slaEvent);
                }
                updateList.add(new UpdateEntry<WorkflowActionQuery>(WorkflowActionQuery.UPDATE_ACTION_STATUS_PENDING, 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;
        }
        SLAEventBean slaEvent = SLADbXOperations.createStatusEvent(wfJob.getSlaXml(), jobId, slaStatus, SlaAppType.WORKFLOW_JOB);
        if (slaEvent != null) {
            insertList.add(slaEvent);
        }
        queue(new WorkflowNotificationXCommand(wfJob));
        if (wfJob.getStatus() == WorkflowJob.Status.SUCCEEDED) {
            InstrumentUtils.incrJobCounter(INSTR_SUCCEEDED_JOBS_COUNTER_NAME, 1, getInstrumentation());
        }
        // output message for Kill node
        if (wfAction != null) {
            // wfAction could be a no-op job
            NodeDef nodeDef = workflowInstance.getNodeDef(wfAction.getExecutionPath());
            if (nodeDef != null && nodeDef instanceof KillNodeDef) {
                boolean isRetry = false;
                boolean isUserRetry = false;
                ActionExecutorContext context = new ActionXCommand.ActionExecutorContext(wfJob, wfAction, isRetry, isUserRetry);
                InstrumentUtils.incrJobCounter(INSTR_KILLED_JOBS_COUNTER_NAME, 1, getInstrumentation());
                try {
                    String tmpNodeConf = nodeDef.getConf();
                    String message = context.getELEvaluator().evaluate(tmpNodeConf, String.class);
                    LOG.debug("Try to resolve KillNode message for jobid [{0}], actionId [{1}], before resolve [{2}], " + "after resolve [{3}]", jobId, actionId, tmpNodeConf, message);
                    if (wfAction.getErrorCode() != null) {
                        wfAction.setErrorInfo(wfAction.getErrorCode(), message);
                    } else {
                        wfAction.setErrorInfo(ErrorCode.E0729.toString(), message);
                    }
                } catch (Exception ex) {
                    LOG.warn("Exception in SignalXCommand when processing Kill node message: {0}", ex.getMessage(), ex);
                    wfAction.setErrorInfo(ErrorCode.E0756.toString(), ErrorCode.E0756.format(ex.getMessage()));
                    wfAction.setStatus(WorkflowAction.Status.ERROR);
                }
                updateList.add(new UpdateEntry<WorkflowActionQuery>(WorkflowActionQuery.UPDATE_ACTION_PENDING_TRANS_ERROR, wfAction));
            }
        }
    } else {
        for (WorkflowActionBean newAction : WorkflowStoreService.getActionsToStart(workflowInstance)) {
            boolean isOldWFAction = false;
            // old action. To avoid twice entry for same action, Checking in Db if the workflow action already exist.
            if (SubWorkflowActionExecutor.ACTION_TYPE.equals(newAction.getType())) {
                try {
                    WorkflowActionBean oldAction = WorkflowActionQueryExecutor.getInstance().get(WorkflowActionQuery.GET_ACTION_CHECK, newAction.getId());
                    newAction.setExternalId(oldAction.getExternalId());
                    newAction.setCreatedTime(oldAction.getCreatedTime());
                    isOldWFAction = true;
                } catch (JPAExecutorException e) {
                    if (e.getErrorCode() != ErrorCode.E0605) {
                        throw new CommandException(e);
                    }
                }
            }
            String skipVar = workflowInstance.getVar(newAction.getName() + WorkflowInstance.NODE_VAR_SEPARATOR + ReRunXCommand.TO_SKIP);
            boolean skipNewAction = false, suspendNewAction = false;
            if (skipVar != null) {
                skipNewAction = skipVar.equals("true");
            }
            if (skipNewAction) {
                WorkflowActionBean oldAction = new WorkflowActionBean();
                oldAction.setId(newAction.getId());
                oldAction.setPending();
                oldAction.setExecutionPath(newAction.getExecutionPath());
                updateList.add(new UpdateEntry<WorkflowActionQuery>(WorkflowActionQuery.UPDATE_ACTION_PENDING, oldAction));
                queue(new SignalXCommand(jobId, oldAction.getId()));
            } else {
                if (!skipAction) {
                    try {
                        // Make sure that transition node for a forked action
                        // is inserted only once
                        WorkflowActionQueryExecutor.getInstance().get(WorkflowActionQuery.GET_ACTION_ID_TYPE_LASTCHECK, newAction.getId());
                        continue;
                    } catch (JPAExecutorException jee) {
                    }
                }
                suspendNewAction = checkForSuspendNode(newAction);
                newAction.setPending();
                String actionSlaXml = getActionSLAXml(newAction.getName(), workflowInstance.getApp().getDefinition(), wfJob.getConf());
                newAction.setSlaXml(actionSlaXml);
                if (!isOldWFAction) {
                    newAction.setCreatedTime(new Date());
                    insertList.add(newAction);
                } else {
                    updateList.add(new UpdateEntry<WorkflowActionQuery>(WorkflowActionQuery.UPDATE_ACTION_START, newAction));
                }
                LOG.debug("SignalXCommand: Name: " + newAction.getName() + ", Id: " + newAction.getId() + ", Authcode:" + newAction.getCred());
                if (wfAction != null) {
                    // null during wf job submit
                    ActionService as = Services.get().get(ActionService.class);
                    ActionExecutor current = as.getExecutor(wfAction.getType());
                    LOG.trace("Current Action Type:" + current.getClass());
                    if (!suspendNewAction) {
                        if (current instanceof StartActionExecutor) {
                            // Excluding :start: here from executing first action synchronously since it
                            // blocks the consumer thread till the action is submitted to Hadoop,
                            // in turn reducing the number of new submissions the threads can accept.
                            // Would also be susceptible to longer delays in case Hadoop cluster is busy.
                            queue(new ActionStartXCommand(newAction.getId(), newAction.getType()));
                        } else if (current instanceof ForkActionExecutor) {
                            if (ConfigurationService.getBoolean(SignalXCommand.FORK_PARALLEL_JOBSUBMISSION)) {
                                workflowActionBeanListForForked.add(newAction);
                            } else {
                                queue(new ActionStartXCommand(newAction.getId(), newAction.getType()));
                            }
                        } else {
                            syncAction = newAction;
                        }
                    } else {
                        // will be ignored.
                        if (ConfigurationService.getBoolean(SignalXCommand.FORK_PARALLEL_JOBSUBMISSION)) {
                            workflowActionBeanListForForked.add(newAction);
                        }
                    }
                } else {
                    // first action after wf submit should always be sync
                    syncAction = newAction;
                }
            }
        }
    }
    try {
        wfJob.setLastModifiedTime(new Date());
        updateList.add(new UpdateEntry<WorkflowJobQuery>(WorkflowJobQuery.UPDATE_WORKFLOW_STATUS_INSTANCE_MOD_START_END, wfJob));
        // call JPAExecutor to do the bulk writes
        BatchQueryExecutor.getInstance().executeBatchInsertUpdateDelete(insertList, updateList, null);
        if (prevStatus != wfJob.getStatus()) {
            LOG.debug("Updated the workflow status to " + wfJob.getId() + "  status =" + wfJob.getStatusStr());
        }
        if (generateEvent && EventHandlerService.isEnabled()) {
            generateEvent(wfJob, wfJobErrorCode, wfJobErrorMsg);
        }
    } catch (JPAExecutorException je) {
        throw new CommandException(je);
    }
    // undue delay from between end of previous and start of next action
    if (wfJob.getStatus() != WorkflowJob.Status.RUNNING && wfJob.getStatus() != WorkflowJob.Status.SUSPENDED) {
        // only for asynchronous actions, parent coord action's external id will
        // persisted and following update will succeed.
        updateParentIfNecessary(wfJob);
        // To delete the WF temp dir
        new WfEndXCommand(wfJob).call();
    } else if (syncAction != null) {
        new ActionStartXCommand(wfJob, syncAction.getId(), syncAction.getType()).call();
    } else if (!workflowActionBeanListForForked.isEmpty() && !checkForSuspendNode(workflowActionBeanListForForked)) {
        startForkedActions(workflowActionBeanListForForked);
    }
    LOG.debug("ENDED SignalCommand for jobid=" + jobId + ", actionId=" + actionId);
    return null;
}
Also used : NodeDef(org.apache.oozie.workflow.lite.NodeDef) KillNodeDef(org.apache.oozie.workflow.lite.KillNodeDef) ArrayList(java.util.ArrayList) WorkflowInstance(org.apache.oozie.workflow.WorkflowInstance) WorkflowActionBean(org.apache.oozie.WorkflowActionBean) JPAExecutorException(org.apache.oozie.executor.jpa.JPAExecutorException) WorkflowJob(org.apache.oozie.client.WorkflowJob) ForkedActionExecutorContext(org.apache.oozie.command.wf.ActionXCommand.ForkedActionExecutorContext) ActionExecutorContext(org.apache.oozie.command.wf.ActionXCommand.ActionExecutorContext) ActionService(org.apache.oozie.service.ActionService) Status(org.apache.oozie.client.SLAEvent.Status) SubWorkflowActionExecutor(org.apache.oozie.action.oozie.SubWorkflowActionExecutor) ForkActionExecutor(org.apache.oozie.action.control.ForkActionExecutor) StartActionExecutor(org.apache.oozie.action.control.StartActionExecutor) ActionExecutor(org.apache.oozie.action.ActionExecutor) WorkflowActionQuery(org.apache.oozie.executor.jpa.WorkflowActionQueryExecutor.WorkflowActionQuery) WorkflowException(org.apache.oozie.workflow.WorkflowException) CommandException(org.apache.oozie.command.CommandException) Date(java.util.Date) SLAEventBean(org.apache.oozie.SLAEventBean) JPAExecutorException(org.apache.oozie.executor.jpa.JPAExecutorException) CommandException(org.apache.oozie.command.CommandException) PreconditionException(org.apache.oozie.command.PreconditionException) IOException(java.io.IOException) XException(org.apache.oozie.XException) WorkflowException(org.apache.oozie.workflow.WorkflowException) WorkflowJobQuery(org.apache.oozie.executor.jpa.WorkflowJobQueryExecutor.WorkflowJobQuery) StartActionExecutor(org.apache.oozie.action.control.StartActionExecutor) ForkActionExecutor(org.apache.oozie.action.control.ForkActionExecutor) KillNodeDef(org.apache.oozie.workflow.lite.KillNodeDef)

Example 13 with SLAEventBean

use of org.apache.oozie.SLAEventBean in project oozie by apache.

the class CoordRerunXCommand method writeActionRegistration.

/**
 * Create SLA RegistrationEvent
 *
 * @param actionXml action xml
 * @param actionBean coordinator action bean
 * @param user user name
 * @param group group name
 * @throws Exception thrown if unable to write sla registration event
 */
private void writeActionRegistration(String actionXml, CoordinatorActionBean actionBean, String user, String group) throws Exception {
    Element eAction = XmlUtils.parseXml(actionXml);
    Element eSla = eAction.getChild("action", eAction.getNamespace()).getChild("info", eAction.getNamespace("sla"));
    SLAEventBean slaEvent = SLADbOperations.createSlaRegistrationEvent(eSla, actionBean.getId(), SlaAppType.COORDINATOR_ACTION, user, group, LOG);
    if (slaEvent != null) {
        insertList.add(slaEvent);
    }
}
Also used : Element(org.jdom.Element) SLAEventBean(org.apache.oozie.SLAEventBean)

Example 14 with SLAEventBean

use of org.apache.oozie.SLAEventBean in project oozie by apache.

the class ActionStartXCommand method endWF.

protected void endWF() throws CommandException {
    updateParentIfNecessary(wfJob, 3);
    // To delete the WF temp dir
    new WfEndXCommand(wfJob).call();
    SLAEventBean slaEvent2 = SLADbXOperations.createStatusEvent(wfJob.getSlaXml(), wfJob.getId(), Status.FAILED, SlaAppType.WORKFLOW_JOB);
    if (slaEvent2 != null) {
        insertList.add(slaEvent2);
    }
}
Also used : SLAEventBean(org.apache.oozie.SLAEventBean)

Example 15 with SLAEventBean

use of org.apache.oozie.SLAEventBean in project oozie by apache.

the class ActionStartXCommand method handleError.

protected void handleError(ActionExecutorContext context, WorkflowJobBean workflow, WorkflowActionBean action) throws CommandException {
    failJob(context);
    updateList.add(new UpdateEntry<WorkflowActionQuery>(WorkflowActionQuery.UPDATE_ACTION_START, wfAction));
    updateJobLastModified();
    SLAEventBean slaEvent1 = SLADbXOperations.createStatusEvent(action.getSlaXml(), action.getId(), Status.FAILED, SlaAppType.WORKFLOW_ACTION);
    if (slaEvent1 != null) {
        insertList.add(slaEvent1);
    }
    endWF();
    return;
}
Also used : WorkflowActionQuery(org.apache.oozie.executor.jpa.WorkflowActionQueryExecutor.WorkflowActionQuery) SLAEventBean(org.apache.oozie.SLAEventBean)

Aggregations

SLAEventBean (org.apache.oozie.SLAEventBean)42 Date (java.util.Date)15 JPAService (org.apache.oozie.service.JPAService)14 CommandException (org.apache.oozie.command.CommandException)13 List (java.util.List)12 JPAExecutorException (org.apache.oozie.executor.jpa.JPAExecutorException)10 Element (org.jdom.Element)7 XException (org.apache.oozie.XException)6 WorkflowActionQuery (org.apache.oozie.executor.jpa.WorkflowActionQueryExecutor.WorkflowActionQuery)6 ArrayList (java.util.ArrayList)5 UpdateEntry (org.apache.oozie.executor.jpa.BatchQueryExecutor.UpdateEntry)5 WorkflowJobQuery (org.apache.oozie.executor.jpa.WorkflowJobQueryExecutor.WorkflowJobQuery)5 IOException (java.io.IOException)4 Configuration (org.apache.hadoop.conf.Configuration)4 ControlNodeActionExecutor (org.apache.oozie.action.control.ControlNodeActionExecutor)4 Status (org.apache.oozie.client.SLAEvent.Status)4 PreconditionException (org.apache.oozie.command.PreconditionException)4 Query (javax.persistence.Query)3 ActionExecutorException (org.apache.oozie.action.ActionExecutorException)3 CoordActionQuery (org.apache.oozie.executor.jpa.CoordActionQueryExecutor.CoordActionQuery)3