/**
* 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);
}
}
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;
}