Flink源码分析: 重启策略机制RestartStrategy

Flink源码专辑:
Flink源码分析: 广播状态流实现规格更新或字段参数变更

版权说明：本专栏是作者在日常工作期间对技术的不断深入研究后的沉淀，辛苦码字总结而成。严禁转载，严禁作为商业用途转发。欢迎同行加入大数据技术群QQ：427560730

版本号：Flink1.10.0

重启策略概览：

public static RestartStrategies.RestartStrategyConfiguration noRestart() {
        return new RestartStrategies.NoRestartStrategyConfiguration();
    }

    public static RestartStrategies.RestartStrategyConfiguration fallBackRestart() {
        return new RestartStrategies.FallbackRestartStrategyConfiguration();
    }

    public static RestartStrategies.RestartStrategyConfiguration fixedDelayRestart(int restartAttempts, long delayBetweenAttempts) {
        return fixedDelayRestart(restartAttempts, Time.of(delayBetweenAttempts, TimeUnit.MILLISECONDS));
    }

    public static RestartStrategies.RestartStrategyConfiguration fixedDelayRestart(int restartAttempts, Time delayInterval) {
        return new RestartStrategies.FixedDelayRestartStrategyConfiguration(restartAttempts, delayInterval);
    }

    public static RestartStrategies.FailureRateRestartStrategyConfiguration failureRateRestart(int failureRate, Time failureInterval, Time delayInterval) {
        return new RestartStrategies.FailureRateRestartStrategyConfiguration(failureRate, failureInterval, delayInterval);
    }

Flink支持多种重启策略，可以在程序中发生故障时选择不同的重启模式进行重启。常用的重启策略可以从源码中看书：

• noRestart
• fallBackRestart
• fixedDelayRestart
• failureRateRestart

parseConfiguration方法

parseConfiguration方法输入两个参数：String类型的restartstrategyKind和ReadableConfig类型的configuration，从源码中可以看出根据var2=restartstrategyKind.toLowerCase()的hashCode值进行匹配给变量var3进行赋值，这样可以有效的防止hash冲突的问题。

• 当restartstrategyKind为"none",“off”,“disable"时var3等于"0”,“1”,“2”，此时重启策略为noRestart。
• 当restartstrategyKind为"fixed-delay",“fixeddelay"时，var3等于"3”,“4”。

case 3:
case 4:
     int attempts = (Integer)configuration.get(RestartStrategyOptions.RESTART_STRATEGY_FIXED_DELAY_ATTEMPTS);
     Duration delay = (Duration)configuration.get(RestartStrategyOptions.RESTART_STRATEGY_FIXED_DELAY_DELAY);
     return fixedDelayRestart(attempts, delay.toMillis());

int类型的attempts表示flink作业失败前的重启次数。Duration类型的delay表示两次重启之间的延迟时间
这里都调用的是RestartStrategyOptions这个类。

• 当restartstrategyKind为"failure-rate",“failurerate"时，var3等于"5”,“6”。

case 5:
case 6:
     int maxFailures = (Integer)configuration.get(RestartStrategyOptions.RESTART_STRATEGY_FAILURE_RATE_MAX_FAILURES_PER_INTERVAL);
     Duration failureRateInterval = (Duration)configuration.get(RestartStrategyOptions.RESTART_STRATEGY_FAILURE_RATE_FAILURE_RATE_INTERVAL);
     Duration failureRateDelay = (Duration)configuration.get(RestartStrategyOptions.RESTART_STRATEGY_FAILURE_RATE_DELAY);
     return failureRateRestart(maxFailures, Time.milliseconds(failureRateInterval.toMillis()), Time.milliseconds(failureRateDelay.toMillis()));

int类型的maxFailures表示flink作业最大失败次数，failureRateInterval表示flink故障率的时间间隔，failureRateDelay表示两个重启之间的延迟时间。

代码测试

// set up the execution environment
    val env = StreamExecutionEnvironment.getExecutionEnvironment
    // operate in Event-time
    env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime)
    // create a checkpoint every 5 seconds
    env.enableCheckpointing(5000)
    // try to restart 60 times with 10 seconds delay (10 Minutes)
    env.setRestartStrategy(RestartStrategies.fixedDelayRestart(60, Time.of(10, TimeUnit.SECONDS)))

欢迎加群，一起学习，群号：427560730