Flink学习（三）状态机制于容错机制，State与CheckPoint

摘自Apache官网

一、State的基本概念

什么叫State？搜了一把叫做状态机制。可以用作以下用途。为了保证 at least once, exactly once，Flink引入了State和Checkpoint

• 某个task/operator某时刻的中间结果
• 快照（snapshot）
• 程序一旦crash，恢复用的
• 机器学习模型的参数

二、Flink中包含的State

Keyed State和Opreator State

1、Keyed State基于KeyedStream的状态。这个状态是跟特定的key绑定的。对KeyedStream流上的每一个key，可能都对应一个state。

2、Operator State。于Keyed State不同。Operator State根一个特定的Operator绑定。整个Operator对应一个State。相比较而言一个State上有多个Keyed State。举例子来说，在Flink中的Kafka Connector就使用了Operator State。会在每个connector实例中，保存该实例中消费topic的所有(partition, offset)映射。

一些原子State

• ValueState：即类型为T的单值状态。这个状态与对应的key绑定，是最简单的状态了。它可以通过update方法更新状态值，通过value()方法获取状态值。

• ListState：即key上的状态值为一个列表。可以通过add方法往列表中附加值；也可以通过get()方法返回一个Iterable<T>来遍历状态值。

• ReducingState：这种状态通过用户传入的reduceFunction，每次调用add方法添加值的时候，会调用reduceFunction，最后合并到一个单一的状态值。

• MapState：即状态值为一个map。用户通过put或putAll方法添加元素。

上述的State仅仅与状态进行交互。而真正的状态值，有可能是存在内存，磁盘、或者其它分布式系统中。相当于只是我们持有了这个句柄。那么我们如何得到这个状态的句柄呢？Flink通过StateDescriptor来定义一个状态。这是一个抽象类，内部定义了状态名称、类型、序列化器等基础信息。与上面的状态对应、从StateDescriptor派生了ValueStateDescriptor，ListStateDescriptor等descriptor。

3、研究下Keyed State内部的结构

在StreamingRuntimeContext这个类中可以看到各个State的get方法

/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package org.apache.flink.streaming.api.operators;

import org.apache.flink.annotation.PublicEvolving;
import org.apache.flink.api.common.accumulators.Accumulator;
import org.apache.flink.api.common.functions.BroadcastVariableInitializer;
import org.apache.flink.api.common.functions.util.AbstractRuntimeUDFContext;
import org.apache.flink.api.common.state.AggregatingState;
import org.apache.flink.api.common.state.AggregatingStateDescriptor;
import org.apache.flink.api.common.state.FoldingState;
import org.apache.flink.api.common.state.FoldingStateDescriptor;
import org.apache.flink.api.common.state.KeyedStateStore;
import org.apache.flink.api.common.state.ListState;
import org.apache.flink.api.common.state.ListStateDescriptor;
import org.apache.flink.api.common.state.MapState;
import org.apache.flink.api.common.state.MapStateDescriptor;
import org.apache.flink.api.common.state.ReducingState;
import org.apache.flink.api.common.state.ReducingStateDescriptor;
import org.apache.flink.api.common.state.StateDescriptor;
import org.apache.flink.api.common.state.ValueState;
import org.apache.flink.api.common.state.ValueStateDescriptor;
import org.apache.flink.runtime.execution.Environment;
import org.apache.flink.runtime.jobgraph.tasks.InputSplitProvider;
import org.apache.flink.streaming.api.CheckpointingMode;
import org.apache.flink.streaming.api.graph.StreamConfig;
import org.apache.flink.streaming.runtime.tasks.ProcessingTimeService;
import org.apache.flink.util.Preconditions;

import java.util.List;
import java.util.Map;

/**
 * Implementation of the {@link org.apache.flink.api.common.functions.RuntimeContext},
 * for streaming operators.
 */
@PublicEvolving
public class StreamingRuntimeContext extends AbstractRuntimeUDFContext {

    /** The operator to which this function belongs. */
    private final AbstractStreamOperator<?> operator;

    /** The task environment running the operator. */
    private final Environment taskEnvironment;

    private final StreamConfig streamConfig;

    public StreamingRuntimeContext(AbstractStreamOperator<?> operator,
                                    Environment env, Map<String, Accumulator<?, ?>> accumulators) {
        super(env.getTaskInfo(),
                env.getUserClassLoader(),
                operator.getExecutionConfig(),
                accumulators,
                env.getDistributedCacheEntries(),
                operator.getMetricGroup());

        this.operator = operator;
        this.taskEnvironment = env;
        this.streamConfig = new StreamConfig(env.getTaskConfiguration());
    }

    // ------------------------------------------------------------------------

    /**
     * Returns the input split provider associated with the operator.
     *
     * @return The input split provider.
     */
    public InputSplitProvider getInputSplitProvider() {
        return taskEnvironment.getInputSplitProvider();
    }

    public ProcessingTimeService getProcessingTimeService() {
        return operator.getProcessingTimeService();
    }

    // ------------------------------------------------------------------------
    //  broadcast variables
    // ------------------------------------------------------------------------

    @Override
    public boolean hasBroadcastVariable(String name) {
        throw new UnsupportedOperationException("Broadcast variables can only be used in DataSet programs");
    }

    @Override
    public <RT> List<RT> getBroadcastVariable(String name) {
        throw new UnsupportedOperationException("Broadcast variables can only be used in DataSet programs");
    }

    @Override
    public <T, C> C getBroadcastVariableWithInitializer(String name, BroadcastVariableInitializer<T, C> initializer) {
        throw new UnsupportedOperationException("Broadcast variables can only be used in DataSet programs");
    }

    // ------------------------------------------------------------------------
    //  key/value state
    // ------------------------------------------------------------------------

    @Override
    public <T> ValueState<T> getState(ValueStateDescriptor<T> stateProperties) {
        KeyedStateStore keyedStateStore = checkPreconditionsAndGetKeyedStateStore(stateProperties);
        stateProperties.initializeSerializerUnlessSet(getExecutionConfig());
        return keyedStateStore.getState(stateProperties);
    }

    @Override
    public <T> ListState<T> getListState(ListStateDescriptor<T> stateProperties) {
        KeyedStateStore keyedStateStore = checkPreconditionsAndGetKeyedStateStore(stateProperties);
        stateProperties.initializeSerializerUnlessSet(getExecutionConfig());
        return keyedStateStore.getListState(stateProperties);
    }

    @Override
    public <T> ReducingState<T> getReducingState(ReducingStateDescriptor<T> stateProperties) {
        KeyedStateStore keyedStateStore = checkPreconditionsAndGetKeyedStateStore(stateProperties);
        stateProperties.initializeSerializerUnlessSet(getExecutionConfig());
        return keyedStateStore.getReducingState(stateProperties);
    }

    @Override
    public <IN, ACC, OUT> AggregatingState<IN, OUT> getAggregatingState(AggregatingStateDescriptor<IN, ACC, OUT> stateProperties) {
        KeyedStateStore keyedStateStore = checkPreconditionsAndGetKeyedStateStore(stateProperties);
        stateProperties.initializeSerializerUnlessSet(getExecutionConfig());
        return keyedStateStore.getAggregatingState(stateProperties);
    }

    @Override
    public <T, ACC> FoldingState<T, ACC> getFoldingState(FoldingStateDescriptor<T, ACC> stateProperties) {
        KeyedStateStore keyedStateStore = checkPreconditionsAndGetKeyedStateStore(stateProperties);
        stateProperties.initializeSerializerUnlessSet(getExecutionConfig());
        return keyedStateStore.getFoldingState(stateProperties);
    }

    @Override
    public <UK, UV> MapState<UK, UV> getMapState(MapStateDescriptor<UK, UV> stateProperties) {
        KeyedStateStore keyedStateStore = checkPreconditionsAndGetKeyedStateStore(stateProperties);
        stateProperties.initializeSerializerUnlessSet(getExecutionConfig());
        return keyedStateStore.getMapState(stateProperties);
    }

    private KeyedStateStore checkPreconditionsAndGetKeyedStateStore(StateDescriptor<?, ?> stateDescriptor) {
        Preconditions.checkNotNull(stateDescriptor, "The state properties must not be null");
        KeyedStateStore keyedStateStore = operator.getKeyedStateStore();
        Preconditions.checkNotNull(keyedStateStore, "Keyed state can only be used on a 'keyed stream', i.e., after a 'keyBy()' operation.");
        return keyedStateStore;
    }

    // ------------------ expose (read only) relevant information from the stream config -------- //

    /**
     * Returns true if checkpointing is enabled for the running job.
     *
     * @return true if checkpointing is enabled.
     */
    public boolean isCheckpointingEnabled() {
        return streamConfig.isCheckpointingEnabled();
    }

    /**
     * Returns the checkpointing mode.
     *
     * @return checkpointing mode
     */
    public CheckpointingMode getCheckpointMode() {
        return streamConfig.getCheckpointMode();
    }

    /**
     * Returns the buffer timeout of the job.
     *
     * @return buffer timeout (in milliseconds)
     */
    public long getBufferTimeout() {
        return streamConfig.getBufferTimeout();
    }

}

所有的State都继承自StateDescpritor这个类。简单看一下构造函数。实际上包含了三个参数，

• 名称
• 类型--属于哪一类state
• 默认值

protected StateDescriptor(String name, Class<T> type, T defaultValue) {
        this.name = requireNonNull(name, "name must not be null");
        requireNonNull(type, "type class must not be null");

        try {
            this.typeInfo = TypeExtractor.createTypeInfo(type);
        } catch (Exception e) {
            throw new RuntimeException(
                    "Could not create the type information for '" + type.getName() + "'. " +
                    "The most common reason is failure to infer the generic type information, due to Java's type erasure. " +
                    "In that case, please pass a 'TypeHint' instead of a class to describe the type. " +
                    "For example, to describe 'Tuple2<String, String>' as a generic type, use " +
                    "'new PravegaDeserializationSchema<>(new TypeHint<Tuple2<String, String>>(){}, serializer);'", e);
        }

        this.defaultValue = defaultValue;
    }

再具体的可以自己研究研究，建议工作遇到或者对源码有兴趣可以读读。或者结合实际应用理解下会更快

三、容错机制

1、Flink的核心容错机制是不断的给数据流绘制Snapshots。当系统回滚的时候，这些snapshots就扮演了checkpoints的作用。快照机制受Chandy-Lamport 算法的启发。

读了下论文研究了下这个算法。英文的看不太懂。找了个中文的。

大概就是通过token 和marker来判断是哪里出了问题和需要恢复。

具体可以参考：

https://www.jianshu.com/p/938001e998f5