Examples with AggregatingStateDescriptor org.apache.flink.api.common.state.AggregatingStateDescriptor used on opensource projects

Example 1 with AggregatingStateDescriptor

use of org.apache.flink.api.common.state.AggregatingStateDescriptor in project flink by apache.

the class RocksDBAggregatingStateTest method testMerging.

@Test
public void testMerging() throws Exception {
    final AggregatingStateDescriptor<Long, MutableLong, Long> stateDescr = new AggregatingStateDescriptor<>("my-state", new AddingFunction(), MutableLong.class);
    stateDescr.initializeSerializerUnlessSet(new ExecutionConfig());
    final TimeWindow win1 = new TimeWindow(1000, 2000);
    final TimeWindow win2 = new TimeWindow(2000, 3000);
    final TimeWindow win3 = new TimeWindow(3000, 4000);
    final Long expectedResult = 165L;
    final RocksDBStateBackend backend = new RocksDBStateBackend(tmp.newFolder().toURI());
    backend.setDbStoragePath(tmp.newFolder().getAbsolutePath());
    final RocksDBKeyedStateBackend<String> keyedBackend = createKeyedBackend(backend);
    try {
        InternalAggregatingState<TimeWindow, Long, Long> state = keyedBackend.createAggregatingState(new TimeWindow.Serializer(), stateDescr);
        // populate the different namespaces
        //  - abc spreads the values over three namespaces
        //  - def spreads teh values over two namespaces (one empty)
        //  - ghi is empty
        //  - jkl has all elements already in the target namespace
        //  - mno has all elements already in one source namespace
        keyedBackend.setCurrentKey("abc");
        state.setCurrentNamespace(win1);
        state.add(33L);
        state.add(55L);
        state.setCurrentNamespace(win2);
        state.add(22L);
        state.add(11L);
        state.setCurrentNamespace(win3);
        state.add(44L);
        keyedBackend.setCurrentKey("def");
        state.setCurrentNamespace(win1);
        state.add(11L);
        state.add(44L);
        state.setCurrentNamespace(win3);
        state.add(22L);
        state.add(55L);
        state.add(33L);
        keyedBackend.setCurrentKey("jkl");
        state.setCurrentNamespace(win1);
        state.add(11L);
        state.add(22L);
        state.add(33L);
        state.add(44L);
        state.add(55L);
        keyedBackend.setCurrentKey("mno");
        state.setCurrentNamespace(win3);
        state.add(11L);
        state.add(22L);
        state.add(33L);
        state.add(44L);
        state.add(55L);
        keyedBackend.setCurrentKey("abc");
        state.mergeNamespaces(win1, asList(win2, win3));
        state.setCurrentNamespace(win1);
        assertEquals(expectedResult, state.get());
        keyedBackend.setCurrentKey("def");
        state.mergeNamespaces(win1, asList(win2, win3));
        state.setCurrentNamespace(win1);
        assertEquals(expectedResult, state.get());
        keyedBackend.setCurrentKey("ghi");
        state.mergeNamespaces(win1, asList(win2, win3));
        state.setCurrentNamespace(win1);
        assertNull(state.get());
        keyedBackend.setCurrentKey("jkl");
        state.mergeNamespaces(win1, asList(win2, win3));
        state.setCurrentNamespace(win1);
        assertEquals(expectedResult, state.get());
        keyedBackend.setCurrentKey("mno");
        state.mergeNamespaces(win1, asList(win2, win3));
        state.setCurrentNamespace(win1);
        assertEquals(expectedResult, state.get());
    } finally {
        keyedBackend.close();
        keyedBackend.dispose();
    }
}

Example 2 with AggregatingStateDescriptor

use of org.apache.flink.api.common.state.AggregatingStateDescriptor in project flink by apache.

the class HeapAggregatingStateTest method testAddAndGet.

@Test
public void testAddAndGet() throws Exception {
    final AggregatingStateDescriptor<Long, MutableLong, Long> stateDescr = new AggregatingStateDescriptor<>("my-state", new AddingFunction(), MutableLong.class);
    stateDescr.initializeSerializerUnlessSet(new ExecutionConfig());
    final HeapKeyedStateBackend<String> keyedBackend = createKeyedBackend();
    try {
        InternalAggregatingState<VoidNamespace, Long, Long> state = keyedBackend.createAggregatingState(VoidNamespaceSerializer.INSTANCE, stateDescr);
        state.setCurrentNamespace(VoidNamespace.INSTANCE);
        keyedBackend.setCurrentKey("abc");
        assertNull(state.get());
        keyedBackend.setCurrentKey("def");
        assertNull(state.get());
        state.add(17L);
        state.add(11L);
        assertEquals(28L, state.get().longValue());
        keyedBackend.setCurrentKey("abc");
        assertNull(state.get());
        keyedBackend.setCurrentKey("g");
        assertNull(state.get());
        state.add(1L);
        state.add(2L);
        keyedBackend.setCurrentKey("def");
        assertEquals(28L, state.get().longValue());
        state.clear();
        assertNull(state.get());
        keyedBackend.setCurrentKey("g");
        state.add(3L);
        state.add(2L);
        state.add(1L);
        keyedBackend.setCurrentKey("def");
        assertNull(state.get());
        keyedBackend.setCurrentKey("g");
        assertEquals(9L, state.get().longValue());
        state.clear();
        // make sure all lists / maps are cleared
        StateTable<String, VoidNamespace, MutableLong> stateTable = ((HeapAggregatingState<String, VoidNamespace, Long, MutableLong, Long>) state).stateTable;
        assertTrue(stateTable.isEmpty());
    } finally {
        keyedBackend.close();
        keyedBackend.dispose();
    }
}

Example 3 with AggregatingStateDescriptor

use of org.apache.flink.api.common.state.AggregatingStateDescriptor in project flink by apache.

the class AllWindowedStream method aggregate.

/**
	 * Applies the given window function to each window. The window function is called for each
	 * evaluation of the window for each key individually. The output of the window function is
	 * interpreted as a regular non-windowed stream.
	 *
	 * <p>Arriving data is incrementally aggregated using the given aggregate function. This means
	 * that the window function typically has only a single value to process when called.
	 *
	 * @param aggregateFunction The aggregation function that is used for incremental aggregation.
	 * @param windowFunction The window function.
	 * @param accumulatorType Type information for the internal accumulator type of the aggregation function
	 * @param resultType Type information for the result type of the window function
	 *
	 * @return The data stream that is the result of applying the window function to the window.
	 *
	 * @param <ACC> The type of the AggregateFunction's accumulator
	 * @param <V> The type of AggregateFunction's result, and the WindowFunction's input  
	 * @param <R> The type of the elements in the resulting stream, equal to the
	 *            WindowFunction's result type
	 */
@PublicEvolving
public <ACC, V, R> SingleOutputStreamOperator<R> aggregate(AggregateFunction<T, ACC, V> aggregateFunction, AllWindowFunction<V, R, W> windowFunction, TypeInformation<ACC> accumulatorType, TypeInformation<V> aggregateResultType, TypeInformation<R> resultType) {
    checkNotNull(aggregateFunction, "aggregateFunction");
    checkNotNull(windowFunction, "windowFunction");
    checkNotNull(accumulatorType, "accumulatorType");
    checkNotNull(aggregateResultType, "aggregateResultType");
    checkNotNull(resultType, "resultType");
    if (aggregateFunction instanceof RichFunction) {
        throw new UnsupportedOperationException("This aggregate function cannot be a RichFunction.");
    }
    //clean the closures
    windowFunction = input.getExecutionEnvironment().clean(windowFunction);
    aggregateFunction = input.getExecutionEnvironment().clean(aggregateFunction);
    final String callLocation = Utils.getCallLocationName();
    final String udfName = "AllWindowedStream." + callLocation;
    final String opName;
    final KeySelector<T, Byte> keySel = input.getKeySelector();
    OneInputStreamOperator<T, R> operator;
    if (evictor != null) {
        @SuppressWarnings({ "unchecked", "rawtypes" }) TypeSerializer<StreamRecord<T>> streamRecordSerializer = (TypeSerializer<StreamRecord<T>>) new StreamElementSerializer(input.getType().createSerializer(getExecutionEnvironment().getConfig()));
        ListStateDescriptor<StreamRecord<T>> stateDesc = new ListStateDescriptor<>("window-contents", streamRecordSerializer);
        opName = "TriggerWindow(" + windowAssigner + ", " + stateDesc + ", " + trigger + ", " + evictor + ", " + udfName + ")";
        operator = new EvictingWindowOperator<>(windowAssigner, windowAssigner.getWindowSerializer(getExecutionEnvironment().getConfig()), keySel, input.getKeyType().createSerializer(getExecutionEnvironment().getConfig()), stateDesc, new InternalIterableAllWindowFunction<>(new AggregateApplyAllWindowFunction<>(aggregateFunction, windowFunction)), trigger, evictor, allowedLateness, lateDataOutputTag);
    } else {
        AggregatingStateDescriptor<T, ACC, V> stateDesc = new AggregatingStateDescriptor<>("window-contents", aggregateFunction, accumulatorType.createSerializer(getExecutionEnvironment().getConfig()));
        opName = "TriggerWindow(" + windowAssigner + ", " + stateDesc + ", " + trigger + ", " + udfName + ")";
        operator = new WindowOperator<>(windowAssigner, windowAssigner.getWindowSerializer(getExecutionEnvironment().getConfig()), keySel, input.getKeyType().createSerializer(getExecutionEnvironment().getConfig()), stateDesc, new InternalSingleValueAllWindowFunction<>(windowFunction), trigger, allowedLateness, lateDataOutputTag);
    }
    return input.transform(opName, resultType, operator).forceNonParallel();
}

Example 4 with AggregatingStateDescriptor

use of org.apache.flink.api.common.state.AggregatingStateDescriptor in project flink by apache.

the class WindowedStream method aggregate.

/**
	 * Applies the given window function to each window. The window function is called for each
	 * evaluation of the window for each key individually. The output of the window function is
	 * interpreted as a regular non-windowed stream.
	 *
	 * <p>Arriving data is incrementally aggregated using the given aggregate function. This means
	 * that the window function typically has only a single value to process when called.
	 *
	 * @param aggregateFunction The aggregation function that is used for incremental aggregation.
	 * @param windowFunction The window function.
	 * @param accumulatorType Type information for the internal accumulator type of the aggregation function
	 * @param resultType Type information for the result type of the window function
	 *
	 * @return The data stream that is the result of applying the window function to the window.
	 *
	 * @param <ACC> The type of the AggregateFunction's accumulator
	 * @param <V> The type of AggregateFunction's result, and the WindowFunction's input
	 * @param <R> The type of the elements in the resulting stream, equal to the
	 *            WindowFunction's result type
	 */
@PublicEvolving
public <ACC, V, R> SingleOutputStreamOperator<R> aggregate(AggregateFunction<T, ACC, V> aggregateFunction, ProcessWindowFunction<V, R, K, W> windowFunction, TypeInformation<ACC> accumulatorType, TypeInformation<V> aggregateResultType, TypeInformation<R> resultType) {
    checkNotNull(aggregateFunction, "aggregateFunction");
    checkNotNull(windowFunction, "windowFunction");
    checkNotNull(accumulatorType, "accumulatorType");
    checkNotNull(aggregateResultType, "aggregateResultType");
    checkNotNull(resultType, "resultType");
    if (aggregateFunction instanceof RichFunction) {
        throw new UnsupportedOperationException("This aggregate function cannot be a RichFunction.");
    }
    //clean the closures
    windowFunction = input.getExecutionEnvironment().clean(windowFunction);
    aggregateFunction = input.getExecutionEnvironment().clean(aggregateFunction);
    String callLocation = Utils.getCallLocationName();
    String udfName = "WindowedStream." + callLocation;
    String opName;
    KeySelector<T, K> keySel = input.getKeySelector();
    OneInputStreamOperator<T, R> operator;
    if (evictor != null) {
        @SuppressWarnings({ "unchecked", "rawtypes" }) TypeSerializer<StreamRecord<T>> streamRecordSerializer = (TypeSerializer<StreamRecord<T>>) new StreamElementSerializer(input.getType().createSerializer(getExecutionEnvironment().getConfig()));
        ListStateDescriptor<StreamRecord<T>> stateDesc = new ListStateDescriptor<>("window-contents", streamRecordSerializer);
        opName = "TriggerWindow(" + windowAssigner + ", " + stateDesc + ", " + trigger + ", " + evictor + ", " + udfName + ")";
        operator = new EvictingWindowOperator<>(windowAssigner, windowAssigner.getWindowSerializer(getExecutionEnvironment().getConfig()), keySel, input.getKeyType().createSerializer(getExecutionEnvironment().getConfig()), stateDesc, new InternalAggregateProcessWindowFunction<>(aggregateFunction, windowFunction), trigger, evictor, allowedLateness, lateDataOutputTag);
    } else {
        AggregatingStateDescriptor<T, ACC, V> stateDesc = new AggregatingStateDescriptor<>("window-contents", aggregateFunction, accumulatorType.createSerializer(getExecutionEnvironment().getConfig()));
        opName = "TriggerWindow(" + windowAssigner + ", " + stateDesc + ", " + trigger + ", " + udfName + ")";
        operator = new WindowOperator<>(windowAssigner, windowAssigner.getWindowSerializer(getExecutionEnvironment().getConfig()), keySel, input.getKeyType().createSerializer(getExecutionEnvironment().getConfig()), stateDesc, new InternalSingleValueProcessWindowFunction<>(windowFunction), trigger, allowedLateness, lateDataOutputTag);
    }
    return input.transform(opName, resultType, operator);
}

Example 5 with AggregatingStateDescriptor

use of org.apache.flink.api.common.state.AggregatingStateDescriptor in project flink by apache.

the class AbstractQueryableStateTestBase method testAggregatingState.

@Test
public void testAggregatingState() throws Exception {
    final Deadline deadline = Deadline.now().plus(TEST_TIMEOUT);
    final long numElements = 1024L;
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    env.setStateBackend(stateBackend);
    env.setParallelism(maxParallelism);
    // Very important, because cluster is shared between tests and we
    // don't explicitly check that all slots are available before
    // submitting.
    env.setRestartStrategy(RestartStrategies.fixedDelayRestart(Integer.MAX_VALUE, 1000L));
    DataStream<Tuple2<Integer, Long>> source = env.addSource(new TestAscendingValueSource(numElements));
    final AggregatingStateDescriptor<Tuple2<Integer, Long>, String, String> aggrStateDescriptor = new AggregatingStateDescriptor<>("aggregates", new SumAggr(), String.class);
    aggrStateDescriptor.setQueryable("aggr-queryable");
    source.keyBy(new KeySelector<Tuple2<Integer, Long>, Integer>() {

        private static final long serialVersionUID = 8470749712274833552L;

        @Override
        public Integer getKey(Tuple2<Integer, Long> value) {
            return value.f0;
        }
    }).transform("TestAggregatingOperator", BasicTypeInfo.STRING_TYPE_INFO, new AggregatingTestOperator(aggrStateDescriptor));
    try (AutoCancellableJob autoCancellableJob = new AutoCancellableJob(deadline, clusterClient, env)) {
        final JobID jobId = autoCancellableJob.getJobId();
        final JobGraph jobGraph = autoCancellableJob.getJobGraph();
        clusterClient.submitJob(jobGraph).get();
        for (int key = 0; key < maxParallelism; key++) {
            boolean success = false;
            while (deadline.hasTimeLeft() && !success) {
                CompletableFuture<AggregatingState<Tuple2<Integer, Long>, String>> future = getKvState(deadline, client, jobId, "aggr-queryable", key, BasicTypeInfo.INT_TYPE_INFO, aggrStateDescriptor, false, executor);
                String value = future.get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS).get();
                if (Long.parseLong(value) == numElements * (numElements + 1L) / 2L) {
                    success = true;
                } else {
                    // Retry
                    Thread.sleep(RETRY_TIMEOUT);
                }
            }
            assertTrue("Did not succeed query", success);
        }
    }
}