为什么flink无法从保存点恢复

Question

版本flink 1.7

我正在尝试从保存点（或检查点）还原flink作业，该作业所做的是读取kafka的内容-

我使用rocksdb和启用的检查点。

现在我尝试手动触发一个保存点。每个聚合的预期值为30（1个数据/每分钟）。但是当我从保存点（flink run-d-s{url}）恢复时，聚合值不是30（小于30，取决于我取消flink作业和恢复的时间）。当作业正常运行时，它得到30。

我不知道为什么有些数据会丢失？

日志显示“FlinkKafkaConsumer没有恢复状态”

主代码：

        source.flatMap(new FlatMapFunction<String, Model>() {
        private static final long serialVersionUID = 5814342517597371470L;

        @Override
        public void flatMap(String value, Collector<Model> out) throws Exception {
            LOGGER.info("----> catch value: " + value);
            Model model =  JSONObject.parseObject(value, Model.class);
            out.collect(model);
        }
    }).uid("flatmap-1").name("flatmap-1").assignTimestampsAndWatermarks(new AssignerWithPunctuatedWatermarks<Model>() {

        private static final long serialVersionUID = -1742071229344039681L;

        @Override
        public long extractTimestamp(Model element, long previousElementTimestamp) {
            return element.getTime();
        }

        @Nullable
        @Override
        public Watermark checkAndGetNextWatermark(Model lastElement, long extractedTimestamp) {
            return new Watermark(extractedTimestamp);
        }
    }).setParallelism(1).keyBy(Model::getDim).window(new DynamicWindowAssigner()).aggregate(new AggregateFunction<Model, Model, Model>() {
        @Override
        public Model createAccumulator() {
            return new Model();
        }

        @Override
        public Model add(Model value, Model accumulator) {
            init(value, accumulator);
            accumulator.setValue(accumulator.getValue() + 1);
            return accumulator;
        }

        @Override
        public Model getResult(Model accumulator) {
            return accumulator;
        }

        @Override
        public Model merge(Model a, Model b) {
            return null;
        }

        private void init(Model value, Model accumulator){
            if(accumulator.getTime() == 0L){
                accumulator.setValue(0);
                accumulator.setDim(value.getDim());
                accumulator.setTime(value.getTime());
            }
        }
    }).uid("agg-1").name("agg-1").map(new MapFunction<Model, String>() {
        private static final long serialVersionUID = -1742071229344039681L;

        @Override
        public String map(Model value) throws Exception {
            value.setTime(TimeWindow.getWindowStartWithOffset(value.getTime(), 0, TimeUnit.MINUTES.toMillis(30)));
            return JSONObject.toJSONString(value);
        }
    }).uid("flatmap-2").name("flatmap-2").setParallelism(4).addSink(metricProducer).uid("sink").name("sink").setParallelism(2);

检查点设置：

    final StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime);
    env.enableCheckpointing(60000);
    env.getCheckpointConfig().setCheckpointingMode(CheckpointingMode.EXACTLY_ONCE);
    env.getCheckpointConfig().setCheckpointTimeout(120000);
    env.getCheckpointConfig().setMaxConcurrentCheckpoints(1);
    env.getCheckpointConfig().setMinPauseBetweenCheckpoints(50000);
    StateBackend stateBackend = new RocksDBStateBackend(${path}, true);
    env.setStateBackend(stateBackend);
    env.getCheckpointConfig().enableExternalizedCheckpoints(CheckpointConfig.ExternalizedCheckpointCleanup.RETAIN_ON_CANCELLATION);
    env.getConfig().disableSysoutLogging();

Answer 1

最后我发现我应该使用flink run-s{savepoint}-dxxx.jarinstesad的flink run-dxxx.jar-s{savepoint}，如果-d标志在-s标志前面，那么flink忽略-s