使用Kafka翻滚窗口查询时返回空数据

Question

I'm trying to query the state store to get the data in a window of 5 mins. For that I'm using tumbling window . Have added REST to query the data. I've stream A which consumes data from topic1 and performs some transformations and output a key value to topic2 . Now in stream B I'm doing tumbling window operation on topic2 data. When I run the code and queried using REST, I'm seeing empty data on my browser. I can see the data in the state store flowing.

What I've observed is, instead of topic2 getting data from stream A , I used a producer class to inject the data to topic2 and able to query the data from browser. But when the topic2 is getting data from stream A , I'm getting empty data.

Here is my stream A code :

public static void main(String[] args) {    

                final StreamsBuilder builder = new StreamsBuilder();
                KStream<String, String> source = builder.stream("topic1");
                KStream<String, String> output = source
                        .map((k,v)->
                        {                                                                       
                            Map<String, Object> Fields = new LinkedHashMap<>();

                            Fields.put("FNAME","ABC");
                            Fields.put("LNAME","XYZ");

                            Map<String, Object> nFields = new LinkedHashMap<>();
                            nFields.put("ADDRESS1","HY");
                            nFields.put("ADDRESS2","BA");               
                            nFields.put("addF",Fields);

                            Map<String, Object> eve = new LinkedHashMap<>();                            
                            eve.put("nFields", nFields);

                            Map<String, Object> fevent = new LinkedHashMap<>();
                            fevent.put("eve", eve);             
                            LinkedHashMap<String, Object> newMap = new LinkedHashMap<>(fevent);                                 

                            return new KeyValue<>("JAY1234",newMap.toString());  
                        });

                output.to("topic2");        

}

Here is my stream B code (where tumbling window operation happening):

public static void main(String[] args) {

    final StreamsBuilder builder = new StreamsBuilder();
    KStream<String, String> eventStream = builder.stream("topic2");

    eventStream.groupByKey()
        .windowedBy(TimeWindows.of(300000))         
        .reduce((v1, v2) -> v1 + ";" + v2, Materialized.as("TumblingWindowPoc"));

    final Topology topology = builder.build();      
    KafkaStreams streams = new KafkaStreams(topology, props);   
    streams.start();      
}

REST code :

@GET()
    @Path("/{storeName}/{key}")
    @Produces(MediaType.APPLICATION_JSON)
    public List<KeyValue<String, String>> windowedByKey(@PathParam("storeName") final String storeName,
            @PathParam("key") final String key) {

        final ReadOnlyWindowStore<String, String> store = streams.store(storeName,
                QueryableStoreTypes.<String, String>windowStore());
        if (store == null) {
            throw new NotFoundException();      }

        long timeTo = System.currentTimeMillis(); 
        long timeFrom = timeTo - 30000;         
        final WindowStoreIterator<String> results = store.fetch(key, timeFrom, timeTo);

        final List<KeyValue<String,String>> windowResults = new ArrayList<>();
        while (results.hasNext()) {
            final KeyValue<Long, String> next = results.next();     
            windowResults.add(new KeyValue<String,String>(key + "@" + next.key, next.value));
        }
        return windowResults;
    }

And this is how my key value data looks like :

JAY1234 {eve = {nFields = {ADDRESS1 = HY,ADDRESS2 = BA,Fields = {FNAME = ABC,LNAME = XYZ,}}}}

I should be able to get the data when querying using REST. Any help is greatly appreciated. Thanks!

Answer 1

to fetch the window timeFrom should be before window start. So if you want the data for last 30 seconds, you can substract window duration for fetching, like timeTo - 30000 - 300000, and then filter out events required events from whole window data