qi quan created FLINK-9668:
------------------------------

             Summary: In some case Trigger.onProcessingTime don't exectue
                 Key: FLINK-9668
                 URL: https://issues.apache.org/jira/browse/FLINK-9668
             Project: Flink
          Issue Type: Bug
            Reporter: qi quan


For example, I would like to achieve a statistical window of one day, and I 
want to output the result of the indicator every 1 minute.
So I implemented my Trigger like this.
onElement: check if valuestate has stored the nextfiretime, register the 
nextfiretime,
onProcessingTime: Registers the nextfiretime (time+1min),update valuestate, 
return FIRE_AND_PURGE.
(The amount of data in one day is too large. I don't want to store such a large 
window state.)
{code:java}
public class PayAmountTrigger extends Trigger<Tuple2<String, String>, 
TimeWindow> {
    private static final Logger LOGGER = 
LoggerFactory.getLogger(PayAmountTrigger.class);
    private static final Long PERIOD = 1000L * 5;
    ValueStateDescriptor<Long> stateDesc = new 
ValueStateDescriptor("fire-time", LongSerializer.INSTANCE);

    @Override
    public TriggerResult onElement(Tuple2<String, String> tuple2, long l, 
TimeWindow timeWindow, TriggerContext triggerContext) throws Exception {
        ValueState<Long> firstTimeState = 
triggerContext.getPartitionedState(stateDesc);
        long time = triggerContext.getCurrentProcessingTime();
        if (firstTimeState.value() == null) {
            long start = time - (time % PERIOD);
            long nextFireTimestamp = start + PERIOD;
            triggerContext.registerProcessingTimeTimer(nextFireTimestamp);
            firstTimeState.update(nextFireTimestamp);
            return TriggerResult.CONTINUE;
        }
        return TriggerResult.CONTINUE;
    }


    @Override
    public TriggerResult onProcessingTime(long l, TimeWindow timeWindow, 
TriggerContext triggerContext) throws Exception {
        ValueState<Long> state = triggerContext.getPartitionedState(stateDesc);
        if (state.value().equals(l)) {
            state.clear();
            state.update(l + PERIOD);
            triggerContext.registerProcessingTimeTimer(l + PERIOD);
            return TriggerResult.FIRE_AND_PURGE;
        }
        return TriggerResult.CONTINUE;
    }

    @Override
    public TriggerResult onEventTime(long l, TimeWindow timeWindow, 
TriggerContext triggerContext) throws Exception {
        return TriggerResult.CONTINUE;
    }

    @Override
    public void clear(TimeWindow timeWindow, TriggerContext triggerContext) 
throws Exception {
        System.out.println("PayAmountTrigger_clear");
        ValueState<Long> firstTimeState = 
triggerContext.getPartitionedState(stateDesc);
        long timestamp = firstTimeState.value();
        triggerContext.deleteProcessingTimeTimer(timestamp);
        firstTimeState.clear();
    }
}{code}
Then I found out that if there is no data in this minute, onProcessingTime will 
not be executed and you will miss the trigger time forever.
Then I dig through the code and find in the WindowOperator.onProcessingTime
{code:java}
ACC contents = null;
if (windowState != null) {
   contents = windowState.get();
}

if (contents != null) {
   TriggerResult triggerResult = 
triggerContext.onProcessingTime(timer.getTimestamp());
   if (triggerResult.isFire()) {
      emitWindowContents(triggerContext.window, contents);
   }
   if (triggerResult.isPurge()) {
      windowState.clear();
   }
}{code}
This means that if no data comes up for this minute,And I also purge the window 
data, triggerContext.onProcessingTime will never be executed.I think this is a 
bug in flink.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to