flink 在 k8s 上缺少状态值 - 在 jobmanager/taskmanager 崩溃时恢复作业

Question

在kubernetes上的flink job cluster(deployment/pod) 运行，我们删除了jobmanager和taskmanager(kubectl delete pod XXX)。我们发现之前的 pod 中缺少状态，在 pod 为运行并且工作正常后，rocksDB 和检查点文件路径从 PVC 挂载。在 pod 为运行后有什么恢复状态的建议吗？我仔细检查了代码。我发现检查点未启用。这是工作无法恢复的根本原因吗？

环境设置如下

RocksDBStateBackend backend = new RocksDBStateBackend(checkPointDataUri + "/checkpoint",true);
        backend.setDbStoragePath(checkPointDataUri + "/RocksDB");
        backend.setNumberOfTransferingThreads(1);

        // add state backend
        env.setStateBackend((StateBackend)backend);

我们可以像下面那样启用检查点吗？

    env.enableCheckpointing(1000);
        env.getCheckpointConfig().setCheckpointingMode(CheckpointingMode.EXACTLY_ONCE);
        env.getCheckpointConfig().setMinPauseBetweenCheckpoints(500);
        env.getCheckpointConfig().setCheckpointTimeout(60000);
        env.getCheckpointConfig().setMaxConcurrentCheckpoints(1);
        env.getCheckpointConfig().enableExternalizedCheckpoints(CheckpointConfig.ExternalizedCheckpointCleanup.RETAIN_ON_CANCELLATION);

下面是重启日志

2020-06-09 06:48:11,921 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,921 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,962 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,941 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,962 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,941 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,963 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,921 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,963 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,963 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,963 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,942 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,965 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,961 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,965 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,942 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}
2020-06-09 06:48:11,981 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Configuring application-defined state backend with job/cluster config
2020-06-09 06:48:11,944 INFO  org.apache.flink.streaming.runtime.tasks.StreamTask           - Using application-defined state backend: RocksDBStateBackend{checkpointStreamBackend=File State Backend (checkpoints: 'file:/opt/flink/data/ss-kpi-ewfj8/checkpoint', savepoints: 'null', asynchronous: UNDEFINED, fileStateThreshold: -1), localRocksDbDirectories=[/opt/flink/data/ss-kpi-ewfj8/RocksDB], enableIncrementalCheckpointing=TRUE, numberOfTransferingThreads=1}

Answer 1

RocksDB和checkpoints放在同一个文件系统中没有意义。 RocksDB 应该使用最快的可用本地文件系统——kubernetes 临时存储就可以了。检查点必须存储在某种持久的分布式文件系统中。

flink 在 k8s 上缺少状态值 - 在 jobmanager/taskmanager 崩溃时恢复作业

flink missing state value on k8s - Resume job while jobmanager/taskmanager crashes

apache-flink

flink-streaming

flink-cep