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

Example 1 with MapState

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

the class SimpleStateRequestHandler method getMapState.

private MapState<ByteArrayWrapper, byte[]> getMapState(BeamFnApi.StateRequest request) throws Exception {
    BeamFnApi.StateKey.MultimapSideInput mapUserState = request.getStateKey().getMultimapSideInput();
    byte[] data = Base64.getDecoder().decode(mapUserState.getSideInputId());
    FlinkFnApi.StateDescriptor stateDescriptor = FlinkFnApi.StateDescriptor.parseFrom(data);
    String stateName = PYTHON_STATE_PREFIX + stateDescriptor.getStateName();
    StateDescriptor cachedStateDescriptor = stateDescriptorCache.get(stateName);
    MapStateDescriptor<ByteArrayWrapper, byte[]> mapStateDescriptor;
    if (cachedStateDescriptor instanceof MapStateDescriptor) {
        mapStateDescriptor = (MapStateDescriptor<ByteArrayWrapper, byte[]>) cachedStateDescriptor;
    } else if (cachedStateDescriptor == null) {
        mapStateDescriptor = new MapStateDescriptor<>(stateName, ByteArrayWrapperSerializer.INSTANCE, valueSerializer);
        if (stateDescriptor.hasStateTtlConfig()) {
            FlinkFnApi.StateDescriptor.StateTTLConfig stateTtlConfigProto = stateDescriptor.getStateTtlConfig();
            StateTtlConfig stateTtlConfig = ProtoUtils.parseStateTtlConfigFromProto(stateTtlConfigProto);
            mapStateDescriptor.enableTimeToLive(stateTtlConfig);
        }
        stateDescriptorCache.put(stateName, mapStateDescriptor);
    } else {
        throw new RuntimeException(String.format("State name corrupt detected: " + "'%s' is used both as MAP state and '%s' state at the same time.", stateName, cachedStateDescriptor.getType()));
    }
    byte[] windowBytes = mapUserState.getWindow().toByteArray();
    if (windowBytes.length != 0) {
        bais.setBuffer(windowBytes, 0, windowBytes.length);
        Object namespace = namespaceSerializer.deserialize(baisWrapper);
        return (MapState<ByteArrayWrapper, byte[]>) keyedStateBackend.getPartitionedState(namespace, namespaceSerializer, mapStateDescriptor);
    } else {
        return (MapState<ByteArrayWrapper, byte[]>) keyedStateBackend.getPartitionedState(VoidNamespace.INSTANCE, VoidNamespaceSerializer.INSTANCE, mapStateDescriptor);
    }
}

Example 2 with MapState

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

the class SortingBoundedInputITCase method testBatchExecutionWithTimersOneInput.

@Test
public void testBatchExecutionWithTimersOneInput() {
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    // set parallelism to 1 to have consistent order of results
    env.setParallelism(1);
    Configuration config = new Configuration();
    config.set(ExecutionOptions.RUNTIME_MODE, RuntimeExecutionMode.BATCH);
    env.configure(config, this.getClass().getClassLoader());
    WatermarkStrategy<Tuple2<Integer, Integer>> watermarkStrategy = WatermarkStrategy.forGenerator(ctx -> GENERATE_WATERMARK_AFTER_4_14_TIMESTAMP).withTimestampAssigner((r, previousTimestamp) -> r.f1);
    SingleOutputStreamOperator<Tuple2<Integer, Integer>> elements = env.fromElements(Tuple2.of(1, 3), Tuple2.of(1, 1), Tuple2.of(2, 1), Tuple2.of(1, 4), // late element
    Tuple2.of(2, 3), // late element
    Tuple2.of(1, 2), Tuple2.of(1, 13), Tuple2.of(1, 11), Tuple2.of(2, 14), // late element
    Tuple2.of(1, 11)).assignTimestampsAndWatermarks(watermarkStrategy);
    OutputTag<Integer> lateElements = new OutputTag<>("late_elements", BasicTypeInfo.INT_TYPE_INFO);
    SingleOutputStreamOperator<Tuple3<Long, Integer, Integer>> sums = elements.map(element -> element.f0).keyBy(element -> element).process(new KeyedProcessFunction<Integer, Integer, Tuple3<Long, Integer, Integer>>() {

        private MapState<Long, Integer> countState;

        private ValueState<Long> previousTimestampState;

        @Override
        public void open(Configuration parameters) {
            countState = getRuntimeContext().getMapState(new MapStateDescriptor<>("sum", BasicTypeInfo.LONG_TYPE_INFO, BasicTypeInfo.INT_TYPE_INFO));
            previousTimestampState = getRuntimeContext().getState(new ValueStateDescriptor<>("previousTimestamp", BasicTypeInfo.LONG_TYPE_INFO));
        }

        @Override
        public void processElement(Integer value, Context ctx, Collector<Tuple3<Long, Integer, Integer>> out) throws Exception {
            Long elementTimestamp = ctx.timestamp();
            long nextTen = ((elementTimestamp + 10) / 10) * 10;
            ctx.timerService().registerEventTimeTimer(nextTen);
            if (elementTimestamp < ctx.timerService().currentWatermark()) {
                ctx.output(lateElements, value);
            } else {
                Long previousTimestamp = Optional.ofNullable(previousTimestampState.value()).orElse(0L);
                assertThat(elementTimestamp, greaterThanOrEqualTo(previousTimestamp));
                previousTimestampState.update(elementTimestamp);
                Integer currentCount = Optional.ofNullable(countState.get(nextTen)).orElse(0);
                countState.put(nextTen, currentCount + 1);
            }
        }

        @Override
        public void onTimer(long timestamp, OnTimerContext ctx, Collector<Tuple3<Long, Integer, Integer>> out) throws Exception {
            out.collect(Tuple3.of(timestamp, ctx.getCurrentKey(), countState.get(timestamp)));
            countState.remove(timestamp);
            // this would go in infinite loop if we did not quiesce the
            // timer service.
            ctx.timerService().registerEventTimeTimer(timestamp + 1);
        }
    });
    DataStream<Integer> lateStream = sums.getSideOutput(lateElements);
    List<Integer> lateRecordsCollected = CollectionUtil.iteratorToList(DataStreamUtils.collect(lateStream));
    List<Tuple3<Long, Integer, Integer>> sumsCollected = CollectionUtil.iteratorToList(DataStreamUtils.collect(sums));
    assertTrue(lateRecordsCollected.isEmpty());
    assertThat(sumsCollected, equalTo(Arrays.asList(Tuple3.of(10L, 1, 4), Tuple3.of(20L, 1, 3), Tuple3.of(10L, 2, 2), Tuple3.of(20L, 2, 1))));
}

Example 3 with MapState

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

the class SortingBoundedInputITCase method testBatchExecutionWithTimersTwoInput.

@Test
public void testBatchExecutionWithTimersTwoInput() {
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    // set parallelism to 1 to have consistent order of results
    env.setParallelism(1);
    Configuration config = new Configuration();
    config.set(ExecutionOptions.RUNTIME_MODE, RuntimeExecutionMode.BATCH);
    env.configure(config, this.getClass().getClassLoader());
    WatermarkStrategy<Tuple2<Integer, Integer>> watermarkStrategy = WatermarkStrategy.forGenerator(ctx -> GENERATE_WATERMARK_AFTER_4_14_TIMESTAMP).withTimestampAssigner((r, previousTimestamp) -> r.f1);
    SingleOutputStreamOperator<Integer> elements1 = env.fromElements(Tuple2.of(1, 3), Tuple2.of(1, 1), Tuple2.of(2, 1), Tuple2.of(1, 4), // late element
    Tuple2.of(2, 3), // late element
    Tuple2.of(1, 2), Tuple2.of(1, 13), Tuple2.of(1, 11), Tuple2.of(2, 14), // late element
    Tuple2.of(1, 11)).assignTimestampsAndWatermarks(watermarkStrategy).map(element -> element.f0);
    SingleOutputStreamOperator<Integer> elements2 = env.fromElements(Tuple2.of(1, 3), Tuple2.of(1, 1), Tuple2.of(2, 1), Tuple2.of(1, 4), // late element
    Tuple2.of(2, 3), // late element
    Tuple2.of(1, 2), Tuple2.of(1, 13), Tuple2.of(1, 11), Tuple2.of(2, 14), // late element
    Tuple2.of(1, 11)).assignTimestampsAndWatermarks(watermarkStrategy).map(element -> element.f0);
    OutputTag<Integer> lateElements = new OutputTag<>("late_elements", BasicTypeInfo.INT_TYPE_INFO);
    SingleOutputStreamOperator<Tuple3<Long, Integer, Integer>> sums = elements1.connect(elements2).keyBy(element -> element, element -> element).process(new KeyedCoProcessFunction<Integer, Integer, Integer, Tuple3<Long, Integer, Integer>>() {

        private MapState<Long, Integer> countState;

        private ValueState<Long> previousTimestampState;

        @Override
        public void open(Configuration parameters) {
            countState = getRuntimeContext().getMapState(new MapStateDescriptor<>("sum", BasicTypeInfo.LONG_TYPE_INFO, BasicTypeInfo.INT_TYPE_INFO));
            previousTimestampState = getRuntimeContext().getState(new ValueStateDescriptor<>("previousTimestamp", BasicTypeInfo.LONG_TYPE_INFO));
        }

        @Override
        public void processElement1(Integer value, Context ctx, Collector<Tuple3<Long, Integer, Integer>> out) throws Exception {
            processElement(value, ctx);
        }

        @Override
        public void processElement2(Integer value, Context ctx, Collector<Tuple3<Long, Integer, Integer>> out) throws Exception {
            processElement(value, ctx);
        }

        private void processElement(Integer value, Context ctx) throws Exception {
            Long elementTimestamp = ctx.timestamp();
            long nextTen = ((elementTimestamp + 10) / 10) * 10;
            ctx.timerService().registerEventTimeTimer(nextTen);
            if (elementTimestamp < ctx.timerService().currentWatermark()) {
                ctx.output(lateElements, value);
            } else {
                Long previousTimestamp = Optional.ofNullable(previousTimestampState.value()).orElse(0L);
                assertThat(elementTimestamp, greaterThanOrEqualTo(previousTimestamp));
                previousTimestampState.update(elementTimestamp);
                Integer currentCount = Optional.ofNullable(countState.get(nextTen)).orElse(0);
                countState.put(nextTen, currentCount + 1);
            }
        }

        @Override
        public void onTimer(long timestamp, OnTimerContext ctx, Collector<Tuple3<Long, Integer, Integer>> out) throws Exception {
            out.collect(Tuple3.of(timestamp, ctx.getCurrentKey(), countState.get(timestamp)));
            countState.remove(timestamp);
            // this would go in infinite loop if we did not quiesce the
            // timer service.
            ctx.timerService().registerEventTimeTimer(timestamp + 1);
        }
    });
    DataStream<Integer> lateStream = sums.getSideOutput(lateElements);
    List<Integer> lateRecordsCollected = CollectionUtil.iteratorToList(DataStreamUtils.collect(lateStream));
    List<Tuple3<Long, Integer, Integer>> sumsCollected = CollectionUtil.iteratorToList(DataStreamUtils.collect(sums));
    assertTrue(lateRecordsCollected.isEmpty());
    assertThat(sumsCollected, equalTo(Arrays.asList(Tuple3.of(10L, 1, 8), Tuple3.of(20L, 1, 6), Tuple3.of(10L, 2, 4), Tuple3.of(20L, 2, 2))));
}

Example 4 with MapState

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

the class AbstractKeyedStateBackend method getOrCreateKeyedState.

/**
	 * @see KeyedStateBackend
	 */
@Override
public <N, S extends State, V> S getOrCreateKeyedState(final TypeSerializer<N> namespaceSerializer, StateDescriptor<S, V> stateDescriptor) throws Exception {
    checkNotNull(namespaceSerializer, "Namespace serializer");
    if (keySerializer == null) {
        throw new UnsupportedOperationException("State key serializer has not been configured in the config. " + "This operation cannot use partitioned state.");
    }
    if (!stateDescriptor.isSerializerInitialized()) {
        throw new IllegalStateException("The serializer of the descriptor has not been initialized!");
    }
    InternalKvState<?> existing = keyValueStatesByName.get(stateDescriptor.getName());
    if (existing != null) {
        @SuppressWarnings("unchecked") S typedState = (S) existing;
        return typedState;
    }
    // create a new blank key/value state
    S state = stateDescriptor.bind(new StateBinder() {

        @Override
        public <T> ValueState<T> createValueState(ValueStateDescriptor<T> stateDesc) throws Exception {
            return AbstractKeyedStateBackend.this.createValueState(namespaceSerializer, stateDesc);
        }

        @Override
        public <T> ListState<T> createListState(ListStateDescriptor<T> stateDesc) throws Exception {
            return AbstractKeyedStateBackend.this.createListState(namespaceSerializer, stateDesc);
        }

        @Override
        public <T> ReducingState<T> createReducingState(ReducingStateDescriptor<T> stateDesc) throws Exception {
            return AbstractKeyedStateBackend.this.createReducingState(namespaceSerializer, stateDesc);
        }

        @Override
        public <T, ACC, R> AggregatingState<T, R> createAggregatingState(AggregatingStateDescriptor<T, ACC, R> stateDesc) throws Exception {
            return AbstractKeyedStateBackend.this.createAggregatingState(namespaceSerializer, stateDesc);
        }

        @Override
        public <T, ACC> FoldingState<T, ACC> createFoldingState(FoldingStateDescriptor<T, ACC> stateDesc) throws Exception {
            return AbstractKeyedStateBackend.this.createFoldingState(namespaceSerializer, stateDesc);
        }

        @Override
        public <UK, UV> MapState<UK, UV> createMapState(MapStateDescriptor<UK, UV> stateDesc) throws Exception {
            return AbstractKeyedStateBackend.this.createMapState(namespaceSerializer, stateDesc);
        }
    });
    @SuppressWarnings("unchecked") InternalKvState<N> kvState = (InternalKvState<N>) state;
    keyValueStatesByName.put(stateDescriptor.getName(), kvState);
    // Publish queryable state
    if (stateDescriptor.isQueryable()) {
        if (kvStateRegistry == null) {
            throw new IllegalStateException("State backend has not been initialized for job.");
        }
        String name = stateDescriptor.getQueryableStateName();
        kvStateRegistry.registerKvState(keyGroupRange, name, kvState);
    }
    return state;
}

Example 5 with MapState

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

the class AbstractQueryableStateTestBase method testMapState.

/**
 * Tests simple map state queryable state instance. Each source emits (subtaskIndex,
 * 0)..(subtaskIndex, numElements) tuples, which are then queried. The map state instance sums
 * the values up. The test succeeds after each subtask index is queried with result n*(n+1)/2.
 */
@Test
public void testMapState() 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 MapStateDescriptor<Integer, Tuple2<Integer, Long>> mapStateDescriptor = new MapStateDescriptor<>("timon", BasicTypeInfo.INT_TYPE_INFO, source.getType());
    mapStateDescriptor.setQueryable("timon-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;
        }
    }).process(new ProcessFunction<Tuple2<Integer, Long>, Object>() {

        private static final long serialVersionUID = -805125545438296619L;

        private transient MapState<Integer, Tuple2<Integer, Long>> mapState;

        @Override
        public void open(Configuration parameters) throws Exception {
            super.open(parameters);
            mapState = getRuntimeContext().getMapState(mapStateDescriptor);
        }

        @Override
        public void processElement(Tuple2<Integer, Long> value, Context ctx, Collector<Object> out) throws Exception {
            Tuple2<Integer, Long> v = mapState.get(value.f0);
            if (v == null) {
                v = new Tuple2<>(value.f0, 0L);
            }
            mapState.put(value.f0, new Tuple2<>(v.f0, v.f1 + value.f1));
        }
    });
    try (AutoCancellableJob autoCancellableJob = new AutoCancellableJob(deadline, clusterClient, env)) {
        final JobID jobId = autoCancellableJob.getJobId();
        final JobGraph jobGraph = autoCancellableJob.getJobGraph();
        clusterClient.submitJob(jobGraph).get();
        final long expected = numElements * (numElements + 1L) / 2L;
        for (int key = 0; key < maxParallelism; key++) {
            boolean success = false;
            while (deadline.hasTimeLeft() && !success) {
                CompletableFuture<MapState<Integer, Tuple2<Integer, Long>>> future = getKvState(deadline, client, jobId, "timon-queryable", key, BasicTypeInfo.INT_TYPE_INFO, mapStateDescriptor, false, executor);
                Tuple2<Integer, Long> value = future.get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS).get(key);
                if (value != null && value.f0 != null && expected == value.f1) {
                    assertEquals("Key mismatch", key, value.f0.intValue());
                    success = true;
                } else {
                    // Retry
                    Thread.sleep(RETRY_TIMEOUT);
                }
            }
            assertTrue("Did not succeed query", success);
        }
    }
}