To amplify Sandesh's answer a bit, the main operator thread invokes user callbacks like beginWindow(), endWindow(), process() method of input ports, and emitTuples() in input operators.
Additionally, if the operator implements the IdleTimeHandler, and if the operator is deemed to be idle, the handleIdleTime() callback will be called. All tuple output must be done in one of these callbacks. So you can check the thread-safe queue in any of these callbacks and emit output tuples as needed. Ram On Tue, Feb 21, 2017 at 11:42 AM, Sandesh Hegde <[email protected]> wrote: > -Removed dev@ > > Operators can implement idle Time Handler. > https://www.datatorrent.com/docs/apidocs/com/datatorrent/ > api/Operator.IdleTimeHandler.html > > On Tue, Feb 21, 2017 at 11:33 AM Sunil Parmar <[email protected]> > wrote: > >> Ram, >> Thanks for the prompt response. If we use the approach you suggested >> we’re dependent on main thread’s process call I.e. Tuples in the thread >> safe queue gets only processed when main thread is processing incoming >> tuples. How can we explicitly call the process from polling of delay queue ? >> >> Just for reference here’s the sample code snippet for our operator. >> >> public class MyOperator extends BaseOperator implements >> >> Operator.ActivationListener<Context.OperatorContext> { >> ….. >> >> @InputPortFieldAnnotation >> >> public transient DefaultInputPort<String> kafkaStreamInput = >> >> new DefaultInputPort<String>() { >> >> List<String> errors = new ArrayList<String>(); >> >> @Override >> >> public void process(String consumerRecord) { >> >> //Code for normal tuple process >> >> //Code to poll thread safe queue >> >> } >> >> *————————————**—————————————————————————* >> *From: *Munagala Ramanath <[email protected]> >> *To: *[email protected] >> *CC: *"[email protected]" <[email protected]>, Allan De Leon < >> [email protected]>, Tim Zhu <[email protected]> >> *Subject: *Re: Occasional Out of order tuples when emitting from a thread >> *Date: *2017-02-21 10:08 (-0800) >> *List: *[email protected] >> <https://lists.apache.org/[email protected]> >> >> Please note that tuples should not be emitted by any thread other than the >> main operator thread. >> >> A common pattern is to use a thread-safe queue and have worker threads >> enqueue >> tuples there; the main operator thread then pulls tuples from the queue and >> emits them. >> >> Ram >> >> _______________________________________________________ >> >> Munagala V. Ramanath >> >> Software Engineer >> >> E: [email protected] | M: (408) 331-5034 | Twitter: @UnknownRam >> www.datatorrent.com | apex.apache.org >> >> >> From: Sunil Parmar <[email protected]> >> Date: Tuesday, February 21, 2017 at 10:05 AM >> To: "[email protected]" <[email protected]>, "[email protected]" >> <[email protected]> >> Cc: Allan De Leon <[email protected]>, Tim Zhu < >> [email protected]> >> Subject: Occasional Out of order tuples when emitting from a thread >> >> Hi there, >> We have the following setup: >> >> - we have a generic operator that’s processing tuples in its input >> port >> - in the input port’s process method, we check for a condition, and: >> - if the condition is met, the tuple is emitted to the next >> operator right away (in the process method) >> - Otherwise, if the condition is not met, we store the tuple in >> some cache and we use some threads that periodically check the >> condition to >> become true. Once the condition is true, the threads call the emit >> method >> on the stored tuples. >> >> With this setup, we occasionally encounter the following error: >> 2017-02-15 17:29:09,364 ERROR com.datatorrent.stram.engine.GenericNode: >> Catastrophic Error: Out of sequence BEGIN_WINDOW tuple 58a4046100003b7f on >> port transformedJSON while expecting 58a4046100003b7e >> >> Is there a way to make the above work correctly? >> If not, can you recommend a better way of doing this? >> How can we ensure window assignment is done synchronously before emitting >> tuples ? >> >> Thanks very much in advance… >> -allan >> > -- > *Join us at Apex Big Data World-San Jose > <http://www.apexbigdata.com/san-jose.html>, April 4, 2017!* > [image: http://www.apexbigdata.com/san-jose-register.html] > <http://www.apexbigdata.com/san-jose-register.html> > -- _______________________________________________________ Munagala V. Ramanath Software Engineer E: [email protected] | M: (408) 331-5034 | Twitter: @UnknownRam www.datatorrent.com | apex.apache.org
