• Flink如何设置RocksDB日志:How to Configure RocksDB Logging for Advanced Troubleshooting


    Question

    RocksDB's LOG file comes in handy when troubleshooting Flink with RocksDB. How can I configure RocksDB logging?

    Answer

    Note: This section applies to Flink 1.10 - 1.14

    By default, Flink uses the log level HEADER_LEVEL for RocksDB. This essentially disabled RocksDB logging and leave only RocksDB configuration printed in the RocksDB log file. The main reason is that the RocksDB log file is not controllable in size prior to Flink 1.14. The way to configure RocksDB logging depends on the version of Flink you are using. Flink 1.13 or later supports changing RocksDB log level via configuration. Flink 1.14 additionally supports specifying the logging directory which can be located on a (separate) volume that is retained after container shutdown for debugging purposes. Thanks to the new RocksDB version, you can also configure log rotation in Flink 1.14 or later. For other scenarios, you will need to define your own custom Options Factory.

    Via Configuration in Flink 1.14 or later

    For example:

    state.backend.rocksdb.log.level: INFO_LEVEL
    state.backend.rocksdb.log.max-file-size: 10MB
    state.backend.rocksdb.log.file-num: 10
    state.backend.rocksdb.log.dir: /rocksdb/logs

    Refer to Flink's documentation for more details of these configuration options.

    Via Configuration in Flink 1.13

    For example:

    state.backend.rocksdb.log.level: INFO_LEVEL

    Custom Options Factory

    If you are using older versions of Flink (<1.13), you want to configure logging directory in Flink 1.13, or you want to dump RocksDB statistics in RocksDB's LOG file, you can create a custom Options Factory by extending Flink's DefaultConfigurableOptionsFactory. This mechanism gives you options to configure RocksDB logging while still allows your jobs to continue using any other RocksDB tuning options the way you used them before.

    Important: DefaultConfigurableOptionsFactory was not really meant for being extended and may change among releases. If you plan to take this into production, you should write your own complete ConfigurableRocksDBOptionsFactory and set all the options you need in there.

    Extending DefaultConfigurableOptionsFactory

    First, if you have not done so yet, add a dependency to Flink's RocksDB state backend. For example, add this to your Maven project's pom.xml:

    <dependency>
    <groupId>org.apache.flink</groupId>
    <artifactId>flink-statebackend-rocksdb_${scala.binary.version}</artifactId>
    <version>${flink.version}</version>
    <scope>provided</scope>
    </dependency>

    Then create your own Options Factory:

    package com.ververica.troubleshooting;

    import org.apache.flink.contrib.streaming.state.DefaultConfigurableOptionsFactory;
    import org.rocksdb.DBOptions;
    import org.rocksdb.InfoLogLevel; public class MyCustomRocksDBOptionsFactory extends DefaultConfigurableOptionsFactory { private static final long serialVersionUID = 1L; private String dbLogDir = ""; @Override public DBOptions createDBOptions(DBOptions currentOptions, Collection<AutoCloseable> handlesToClose) { currentOptions = super.createDBOptions(currentOptions, handlesToClose); currentOptions.setInfoLogLevel(InfoLogLevel.INFO_LEVEL); currentOptions.setStatsDumpPeriodSec(60); currentOptions.setDbLogDir(dbLogDir); return currentOptions; } @Override public String toString() { return this.getClass().toString() + "{" + super.toString() + '}'; } /** * Set directory where RocksDB writes its info LOG file (empty = data dir, otherwise the * data directory's absolute path will be used as the log file prefix). */ public void setDbLogDir(String dbLogDir) { this.dbLogDir = dbLogDir; } }

    Three points in createDBOptions are important here:

    • setInfoLogLevel(InfoLogLevel.INFO_LEVEL) sets the logging level to INFO from which you would get a decent amount of logging data (increase if needed)
    • setStatsDumpPeriodSec(60) dumps various RocksDB statistics every this many seconds: this includes compaction statistics.
    • setDbLogDir(dbLogDir) specifies the path where to put the LOG file: depending on what you are trying to troubleshoot, you can just use a local directory or you may need to put this onto a distributed file system (or persistent volume) to survive node/pod/job restarts

    Configuring Flink

    With the custom Options Factory, you can configure Flink either programmatically or through its flink-conf.yaml.

    (1) Programmatically

    Note: The state backend interface is changed since Flink 1.13; we provide both versions below.

    // Flink 1.13 or later
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    EmbeddedRocksDBStateBackend stateBackend = new EmbeddedRocksDBStateBackend()
    env.setStateBackend(stateBackend);
    env.getCheckpointConfig().setCheckpointStorage("file:///path/to/checkpoints");

    MyCustomRocksDBOptionsFactory options = new MyCustomRocksDBOptionsFactory();
    options.setDbLogDir("/path/to/rocksdb/logging/");
    stateBackend.setRocksDBOptions(options);
    // Flink <1.13
    RocksDBStateBackend stateBackend = new RocksDBStateBackend("file:///path/to/checkpoints");
    MyCustomRocksDBOptionsFactory options = new MyCustomRocksDBOptionsFactory();
    options.setDbLogDir("/path/to/rocksdb/logging/");
    stateBackend.setRocksDBOptions(options);

    (2) Via Flink configuration (flink-conf.yaml)

    If you want to configure your options factory completely via flink-conf.yaml, we may extend the code above to update its settings from the configuration. The code below shows how to do this for the log directory but you can extend it to make further settings configurable.

    Note: The interface slightly changed since Flink 1.11; we provide both versions below.

    // Flink 1.11 or later
    public static final ConfigOption<String> LOG_DIR = key("state.backend.rocksdb.log.dir") .stringType() .noDefaultValue() .withDescription("Location of RocksDB's info LOG file (empty = data dir, otherwise the " + "data directory's absolute path will be used as the log file prefix)"); @Override public DefaultConfigurableOptionsFactory configure(ReadableConfig configuration) { DefaultConfigurableOptionsFactory optionsFactory = super.configure(configuration); this.dbLogDir = configuration.getOptional(LOG_DIR).orElse(this.dbLogDir); return optionsFactory; }
    // Flink 1.10
    public static final ConfigOption<String> LOG_DIR = key("state.backend.rocksdb.log.dir") .stringType() .noDefaultValue() .withDescription("Location of RocksDB's info LOG file (empty = data dir, otherwise the " + "data directory's absolute path will be used as the log file prefix)"); @Override public DefaultConfigurableOptionsFactory configure(Configuration configuration) { DefaultConfigurableOptionsFactory optionsFactory = super.configure(configuration); this.dbLogDir = configuration.getOptional(LOG_DIR).orElse(this.dbLogDir); return optionsFactory; }

    With the code additions from above, you can simply adapt your flink-conf.yaml and configure it like this:

    state.backend.rocksdb.log.dir: /path/to/rocksdb/logging/
    state.backend.rocksdb.options-factory: com.ververica.troubleshooting.MyCustomRocksDBOptionsFactory

    Note: Configuring the options factory via flink-conf.yaml will apply the options factory to all jobs started in the Flink cluster. Make sure that this class is available cluster-wide or in all jobs started on this cluster!
    (For Ververica Platform deployments, this will not be a problem since each deployment spawns its own Flink cluster.)

    Related Information

  • 相关阅读:
    tomcat端口号被占用问题
    Idea debug时报错:Command line is too long
    IDEA报错: Clone failed: Authentication failed for 'http://10.70.XXXXXXXXXXXXXXXXX'
    LiquiBase实战总结
    Iedis
    IntelliJ IDEA热部署插件JRebel免费激活图文教程(持续更新)转载
    MySql 8.0.11 客户端连接失败:2059
    windows系统 MySQL8.0.12详细安装步骤及基本使用教程
    IDEA的常见的设置和优化(功能)
    SpringBoot热启动让开发更便捷
  • 原文地址:https://www.cnblogs.com/felixzh/p/16037879.html
Copyright © 2020-2023  润新知