@Override
public void rerunChildren() throws CommandException {
boolean isError = false;
try {
CoordinatorActionInfo coordInfo = null;
InstrumentUtils.incrJobCounter(getName(), 1, getInstrumentation());
List<CoordinatorActionBean> coordActions =
CoordUtils.getCoordActions(rerunType, jobId, scope, false);
if (checkAllActionsRunnable(coordActions)) {
for (CoordinatorActionBean coordAction : coordActions) {
String actionXml = coordAction.getActionXml();
if (!noCleanup && !failed) {
Element eAction = XmlUtils.parseXml(actionXml);
cleanupOutputEvents(eAction);
}
if (refresh) {
refreshAction(coordJob, coordAction);
}
updateAction(coordJob, coordAction);
if (SLAService.isEnabled()) {
SLAOperations.updateRegistrationEvent(coordAction.getId());
}
queue(new CoordActionNotificationXCommand(coordAction), 100);
queue(
new CoordActionInputCheckXCommand(coordAction.getId(), coordAction.getJobId()), 100);
}
} else {
isError = true;
throw new CommandException(
ErrorCode.E1018, "part or all actions are not eligible to rerun!");
}
coordInfo = new CoordinatorActionInfo(coordActions);
ret = coordInfo;
} catch (XException xex) {
isError = true;
throw new CommandException(xex);
} catch (JDOMException jex) {
isError = true;
throw new CommandException(ErrorCode.E0700, jex.getMessage(), jex);
} catch (Exception ex) {
isError = true;
throw new CommandException(ErrorCode.E1018, ex.getMessage(), ex);
} finally {
if (isError) {
transitToPrevious();
}
}
}
/* (non-Javadoc)
* @see org.apache.oozie.command.TransitionXCommand#updateJob()
*/
@Override
public void updateJob() {
InstrumentUtils.incrJobCounter("bundle_suspend", 1, null);
bundleJob.setSuspendedTime(new Date());
bundleJob.setLastModifiedTime(new Date());
LOG.debug(
"Suspend bundle job id = "
+ jobId
+ ", status = "
+ bundleJob.getStatus()
+ ", pending = "
+ bundleJob.isPending());
updateList.add(bundleJob);
}
/* (non-Javadoc)
* @see org.apache.oozie.command.XCommand#execute()
*/
@Override
protected String submit() throws CommandException {
String jobId = null;
LOG.info("STARTED Coordinator Submit");
InstrumentUtils.incrJobCounter(getName(), 1, getInstrumentation());
boolean exceptionOccured = false;
try {
mergeDefaultConfig();
String appXml = readAndValidateXml();
coordJob.setOrigJobXml(appXml);
LOG.debug("jobXml after initial validation " + XmlUtils.prettyPrint(appXml).toString());
String appNamespace = readAppNamespace(appXml);
coordJob.setAppNamespace(appNamespace);
appXml = XmlUtils.removeComments(appXml);
initEvaluators();
Element eJob = basicResolveAndIncludeDS(appXml, conf, coordJob);
LOG.debug("jobXml after all validation " + XmlUtils.prettyPrint(eJob).toString());
jobId = storeToDB(eJob, coordJob);
// log job info for coordinator job
LogUtils.setLogInfo(coordJob, logInfo);
LOG = XLog.resetPrefix(LOG);
if (!dryrun) {
// submit a command to materialize jobs for the next 1 hour (3600 secs)
// so we don't wait 10 mins for the Service to run.
queue(new CoordMaterializeTransitionXCommand(jobId, 3600), 100);
} else {
Date startTime = coordJob.getStartTime();
long startTimeMilli = startTime.getTime();
long endTimeMilli = startTimeMilli + (3600 * 1000);
Date jobEndTime = coordJob.getEndTime();
Date endTime = new Date(endTimeMilli);
if (endTime.compareTo(jobEndTime) > 0) {
endTime = jobEndTime;
}
jobId = coordJob.getId();
LOG.info("[" + jobId + "]: Update status to RUNNING");
coordJob.setStatus(Job.Status.RUNNING);
coordJob.setPending();
CoordActionMaterializeCommand coordActionMatCom =
new CoordActionMaterializeCommand(jobId, startTime, endTime);
Configuration jobConf = null;
try {
jobConf = new XConfiguration(new StringReader(coordJob.getConf()));
} catch (IOException e1) {
LOG.warn("Configuration parse error. read from DB :" + coordJob.getConf(), e1);
}
String action = coordActionMatCom.materializeJobs(true, coordJob, jobConf, null);
String output =
coordJob.getJobXml()
+ System.getProperty("line.separator")
+ "***actions for instance***"
+ action;
return output;
}
} catch (CoordinatorJobException cex) {
exceptionOccured = true;
LOG.warn("ERROR: ", cex);
throw new CommandException(cex);
} catch (IllegalArgumentException iex) {
exceptionOccured = true;
LOG.warn("ERROR: ", iex);
throw new CommandException(ErrorCode.E1003, iex);
} catch (Exception ex) {
exceptionOccured = true;
LOG.warn("ERROR: ", ex);
throw new CommandException(ErrorCode.E0803, ex);
} finally {
if (exceptionOccured) {
if (coordJob.getId() == null || coordJob.getId().equalsIgnoreCase("")) {
coordJob.setStatus(CoordinatorJob.Status.FAILED);
coordJob.resetPending();
}
}
}
LOG.info("ENDED Coordinator Submit jobId=" + jobId);
return jobId;
}
@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;
}
