聊聊PowerJob日志的上报及存储

序

本文主要研究一下PowerJob的日志上报及存储

OmsLoggerFactory.build

tech/powerjob/worker/log/OmsLoggerFactory.java

public class OmsLoggerFactory {

    public static OmsLogger build(Long instanceId, String logConfig, WorkerRuntime workerRuntime) {
        LogConfig cfg;
        if (StringUtils.isEmpty(logConfig)) {
            cfg = new LogConfig();
        } else {
            try {
                cfg = JsonUtils.parseObject(logConfig, LogConfig.class);
            } catch (Exception ignore) {
                cfg = new LogConfig();
            }
        }

        switch (LogType.of(cfg.getType())) {
            case LOCAL:
                return new OmsLocalLogger(cfg);
            case STDOUT:
                return new OmsStdOutLogger(cfg);
            case NULL:
                return new OmsNullLogger();
            case LOCAL_AND_ONLINE:
                return new OmsServerAndLocalLogger(cfg, instanceId, workerRuntime.getOmsLogHandler());
            default:
                return new OmsServerLogger(cfg, instanceId, workerRuntime.getOmsLogHandler());
        }
    }
}

默认logConfig为null，cfg是new LogConfig()，其build出来的是OmsServerLogger

OmsServerLogger

tech/powerjob/worker/log/impl/OmsServerLogger.java

public class OmsServerLogger extends AbstractOmsLogger {

    private final long instanceId;
    private final OmsLogHandler omsLogHandler;

    public OmsServerLogger(LogConfig logConfig, long instanceId, OmsLogHandler omsLogHandler) {
        super(logConfig);
        this.instanceId = instanceId;
        this.omsLogHandler = omsLogHandler;
    }

    @Override
    public void debug0(String messagePattern, Object... args) {
        process(LogLevel.DEBUG, messagePattern, args);
    }

    @Override
    public void info0(String messagePattern, Object... args) {
        process(LogLevel.INFO, messagePattern, args);
    }

    @Override
    public void warn0(String messagePattern, Object... args) {
        process(LogLevel.WARN, messagePattern, args);
    }

    @Override
    public void error0(String messagePattern, Object... args) {
        process(LogLevel.ERROR, messagePattern, args);
    }

    private void process(LogLevel level, String messagePattern, Object... args) {
        String logContent = genLogContent(messagePattern, args);
        omsLogHandler.submitLog(instanceId, level, logContent);
    }

}

OmsServerLogger的process方法调用的是OmsLogHandler的submitLog方法

submitLog

tech/powerjob/worker/background/OmsLogHandler.java

@Slf4j
public class OmsLogHandler {

    private final String workerAddress;
    private final Transporter transporter;
    private final ServerDiscoveryService serverDiscoveryService;

    // 处理线程，需要通过线程池启动
    public final Runnable logSubmitter = new LogSubmitter();
    // 上报锁，只需要一个线程上报即可
    private final Lock reportLock = new ReentrantLock();
    // 生产者消费者模式，异步上传日志
    private final BlockingQueue<InstanceLogContent> logQueue = Queues.newLinkedBlockingQueue(10240);

    // 每次上报携带的数据条数
    private static final int BATCH_SIZE = 20;
    // 本地囤积阈值
    private static final int REPORT_SIZE = 1024;

    public OmsLogHandler(String workerAddress, Transporter transporter, ServerDiscoveryService serverDiscoveryService) {
        this.workerAddress = workerAddress;
        this.transporter = transporter;
        this.serverDiscoveryService = serverDiscoveryService;
    }

    /**
     * 提交日志
     * @param instanceId 任务实例ID
     * @param logContent 日志内容
     */
    public void submitLog(long instanceId, LogLevel logLevel, String logContent) {

        if (logQueue.size() > REPORT_SIZE) {
            // 线程的生命周期是个不可循环的过程，一个线程对象结束了不能再次start，只能一直创建和销毁
            new Thread(logSubmitter).start();
        }

        InstanceLogContent tuple = new InstanceLogContent(instanceId, System.currentTimeMillis(), logLevel.getV(), logContent);
        boolean offerRet = logQueue.offer(tuple);
        if (!offerRet) {
            log.warn("[OmsLogHandler] [{}] submit log failed, maybe your log speed is too fast!", instanceId);
        }
    }

    //......
}    

OmsLogHandler的submitLog方法每次先判断logQueue大小是否大于REPORT_SIZE(1024)，是则启动logSubmitter线程，否则放入logQueue队列

LogSubmitter

tech/powerjob/worker/background/OmsLogHandler.java

    private class LogSubmitter implements Runnable {

        @Override
        public void run() {

            boolean lockResult = reportLock.tryLock();
            if (!lockResult) {
                return;
            }

            try {

                final String currentServerAddress = serverDiscoveryService.getCurrentServerAddress();
                // 当前无可用 Server
                if (StringUtils.isEmpty(currentServerAddress)) {
                    if (!logQueue.isEmpty()) {
                        logQueue.clear();
                        log.warn("[OmsLogHandler] because there is no available server to report logs which leads to queue accumulation, oms discarded all logs.");
                    }
                    return;
                }

                List<InstanceLogContent> logs = Lists.newLinkedList();

                while (!logQueue.isEmpty()) {
                    try {
                        InstanceLogContent logContent = logQueue.poll(100, TimeUnit.MILLISECONDS);
                        logs.add(logContent);

                        if (logs.size() >= BATCH_SIZE) {
                            WorkerLogReportReq req = new WorkerLogReportReq(workerAddress, Lists.newLinkedList(logs));
                            // 不可靠请求，WEB日志不追求极致
                            TransportUtils.reportLogs(req, currentServerAddress, transporter);
                            logs.clear();
                        }

                    }catch (Exception ignore) {
                        break;
                    }
                }

                if (!logs.isEmpty()) {
                    WorkerLogReportReq req = new WorkerLogReportReq(workerAddress, logs);
                    TransportUtils.reportLogs(req, currentServerAddress, transporter);
                }

            }finally {
                reportLock.unlock();
            }
        }
    }

LogSubmitter不断地从logQueue.poll数据，在logs的大小大于等于BATCH_SIZE(20)时通过TransportUtils.reportLogs给server上报日志

reportLogs

tech/powerjob/worker/common/utils/TransportUtils.java

    public static void reportLogs(WorkerLogReportReq req, String address, Transporter transporter) {
        final URL url = easyBuildUrl(ServerType.SERVER, S4W_PATH, S4W_HANDLER_REPORT_LOG, address);
        transporter.tell(url, req);
    }

reportLogs请求S4W_HANDLER_REPORT_LOG

processWorkerLogReport

tech/powerjob/server/core/handler/AbWorkerRequestHandler.java

    @Override
    @Handler(path = S4W_HANDLER_REPORT_LOG, processType = ProcessType.NO_BLOCKING)
    public void processWorkerLogReport(WorkerLogReportReq req) {

        WorkerLogReportEvent event = new WorkerLogReportEvent()
                .setWorkerAddress(req.getWorkerAddress())
                .setLogNum(req.getInstanceLogContents().size());
        try {
            processWorkerLogReport0(req, event);
            event.setStatus(WorkerLogReportEvent.Status.SUCCESS);
        } catch (RejectedExecutionException re) {
            event.setStatus(WorkerLogReportEvent.Status.REJECTED);
        } catch (Throwable t) {
            event.setStatus(WorkerLogReportEvent.Status.EXCEPTION);
            log.warn("[WorkerRequestHandler] process worker report failed!", t);
        } finally {
            monitorService.monitor(event);
        }
    }

server端的processWorkerLogReport接收WorkerLogReportReq，执行processWorkerLogReport0方法

WorkerRequestHandlerImpl

tech/powerjob/server/core/handler/WorkerRequestHandlerImpl.java

    protected void processWorkerLogReport0(WorkerLogReportReq req, WorkerLogReportEvent event) {
        // 这个效率应该不会拉垮吧...也就是一些判断 + Map#get 吧...
        instanceLogService.submitLogs(req.getWorkerAddress(), req.getInstanceLogContents());
    }

WorkerRequestHandlerImpl的processWorkerLogReport0执行的是instanceLogService.submitLogs

submitLogs

tech/powerjob/server/core/instance/InstanceLogService.java

    @Async(value = PJThreadPool.LOCAL_DB_POOL)
    public void submitLogs(String workerAddress, List<InstanceLogContent> logs) {

        List<LocalInstanceLogDO> logList = logs.stream().map(x -> {
            instanceId2LastReportTime.put(x.getInstanceId(), System.currentTimeMillis());

            LocalInstanceLogDO y = new LocalInstanceLogDO();
            BeanUtils.copyProperties(x, y);
            y.setWorkerAddress(workerAddress);
            return y;
        }).collect(Collectors.toList());

        try {
            CommonUtils.executeWithRetry0(() -> localInstanceLogRepository.saveAll(logList));
        }catch (Exception e) {
            log.warn("[InstanceLogService] persistent instance logs failed, these logs will be dropped: {}.", logs, e);
        }
    }

InstanceLogService的submitLogs是个异步方法，它将InstanceLogContent转换为LocalInstanceLogDO，然后执行localInstanceLogRepository.saveAll保存

LocalJpaConfig

tech/powerjob/server/persistence/config/LocalJpaConfig.java

@Configuration
@EnableTransactionManagement
@EnableJpaRepositories(
        // repository包名
        basePackages = LocalJpaConfig.LOCAL_PACKAGES,
        // 实体管理bean名称
        entityManagerFactoryRef = "localEntityManagerFactory",
        // 事务管理bean名称
        transactionManagerRef = "localTransactionManager"
)
public class LocalJpaConfig {

    public static final String LOCAL_PACKAGES = "tech.powerjob.server.persistence.local";

    private static Map<String, Object> genDatasourceProperties() {

        JpaProperties jpaProperties = new JpaProperties();
        jpaProperties.setOpenInView(false);
        jpaProperties.setShowSql(false);

        HibernateProperties hibernateProperties = new HibernateProperties();
        // 每次启动都删除数据（重启后原来的Instance已经通过故障转移更换了Server，老的日志数据也没什么意义了）
        hibernateProperties.setDdlAuto("create");
        return hibernateProperties.determineHibernateProperties(jpaProperties.getProperties(), new HibernateSettings());
    }

    @Bean(name = "localEntityManagerFactory")
    public LocalContainerEntityManagerFactoryBean initLocalEntityManagerFactory(@Qualifier("omsLocalDatasource") DataSource omsLocalDatasource,EntityManagerFactoryBuilder builder) {
        return builder
                .dataSource(omsLocalDatasource)
                .properties(genDatasourceProperties())
                .packages(LOCAL_PACKAGES)
                .persistenceUnit("localPersistenceUnit")
                .build();
    }

    @Bean(name = "localTransactionManager")
    public PlatformTransactionManager initLocalTransactionManager(@Qualifier("localEntityManagerFactory") LocalContainerEntityManagerFactoryBean localContainerEntityManagerFactoryBean) {
        return new JpaTransactionManager(Objects.requireNonNull(localContainerEntityManagerFactoryBean.getObject()));
    }

    @Bean(name = "localTransactionTemplate")
    public TransactionTemplate initTransactionTemplate(@Qualifier("localTransactionManager") PlatformTransactionManager ptm) {
        TransactionTemplate tt =  new TransactionTemplate(ptm);
        // 设置隔离级别
        tt.setIsolationLevel(TransactionDefinition.ISOLATION_DEFAULT);
        return tt;
    }
}

LocalJpaConfig针对tech.powerjob.server.persistence.local的dao采用了omsLocalDatasource数据源

MultiDatasourceConfig

tech/powerjob/server/persistence/config/MultiDatasourceConfig.java

@Configuration
public class MultiDatasourceConfig {

    private static final String H2_DRIVER_CLASS_NAME = "org.h2.Driver";
    private static final String H2_JDBC_URL_PATTERN = "jdbc:h2:file:%spowerjob_server_db";
    private static final int H2_MIN_SIZE = 4;
    private static final int H2_MAX_ACTIVE_SIZE = 10;

    @Primary
    @Bean("omsRemoteDatasource")
    @ConfigurationProperties(prefix = "spring.datasource.core")
    public DataSource initOmsCoreDatasource() {
        return DataSourceBuilder.create().build();
    }

    @Bean("omsLocalDatasource")
    public DataSource initOmsLocalDatasource() {
        String h2Path = OmsFileUtils.genH2WorkPath();
        HikariConfig config = new HikariConfig();
        config.setDriverClassName(H2_DRIVER_CLASS_NAME);
        config.setJdbcUrl(String.format(H2_JDBC_URL_PATTERN, h2Path));
        config.setAutoCommit(true);
        // 池中最小空闲连接数量
        config.setMinimumIdle(H2_MIN_SIZE);
        // 池中最大连接数量
        config.setMaximumPoolSize(H2_MAX_ACTIVE_SIZE);

        // JVM 关闭时删除文件
        try {
            FileUtils.forceDeleteOnExit(new File(h2Path));
        }catch (Exception ignore) {
        }
        return new HikariDataSource(config);
    }
}

MultiDatasourceConfig定义了两个数据源，一个是远程的数据源，比如mysql，一个是本地的h2数据源

processFinishedInstance

tech/powerjob/server/core/instance/InstanceManager.java

    public void processFinishedInstance(Long instanceId, Long wfInstanceId, InstanceStatus status, String result) {

        log.info("[Instance-{}] process finished, final status is {}.", instanceId, status.name());

        // 上报日志数据
        HashedWheelTimerHolder.INACCURATE_TIMER.schedule(() -> instanceLogService.sync(instanceId), 60, TimeUnit.SECONDS);

        // workflow 特殊处理
        if (wfInstanceId != null) {
            // 手动停止在工作流中也认为是失败（理论上不应该发生）
            workflowInstanceManager.move(wfInstanceId, instanceId, status, result);
        }

        // 告警
        if (status == InstanceStatus.FAILED) {
            alert(instanceId, result);
        }
        // 主动移除缓存，减小内存占用
        instanceMetadataService.invalidateJobInfo(instanceId);
    }

InstanceManager的processFinishedInstance方法会延时60s执行instanceLogService.sync(instanceId)

sync

tech/powerjob/server/core/instance/InstanceLogService.java

    @Async(PJThreadPool.BACKGROUND_POOL)
    public void sync(Long instanceId) {

        Stopwatch sw = Stopwatch.createStarted();
        try {
            // 先持久化到本地文件
            File stableLogFile = genStableLogFile(instanceId);
            // 将文件推送到 MongoDB

            FileLocation dfsFL = new FileLocation().setBucket(Constants.LOG_BUCKET).setName(genMongoFileName(instanceId));

            try {
                dFsService.store(new StoreRequest().setLocalFile(stableLogFile).setFileLocation(dfsFL));
                log.info("[InstanceLog-{}] push local instanceLogs to mongoDB succeed, using: {}.", instanceId, sw.stop());
            }catch (Exception e) {
                log.warn("[InstanceLog-{}] push local instanceLogs to mongoDB failed.", instanceId, e);
            }

        }catch (Exception e) {
            log.warn("[InstanceLog-{}] sync local instanceLogs failed.", instanceId, e);
        }
        // 删除本地数据库数据
        try {
            instanceId2LastReportTime.remove(instanceId);
            CommonUtils.executeWithRetry0(() -> localInstanceLogRepository.deleteByInstanceId(instanceId));
            log.info("[InstanceLog-{}] delete local instanceLog successfully.", instanceId);
        }catch (Exception e) {
            log.warn("[InstanceLog-{}] delete local instanceLog failed.", instanceId, e);
        }
    }

InstanceLogService的sync方法先通过genStableLogFile将日志持久化到server端的本地日志文件，接着将该任务实例日志的元信息(哪个任务实例、在哪个server、本地日志文件的路径)存储到dFsService(它有oss、gridfs、minio、mysql四种实现，具体看server的配置文件是启动哪个，如果是mysql则是存储到powerjob_files表中)，最后通过localInstanceLogRepository.deleteByInstanceId清空该任务实例在h2中的LOCAL_INSTANCE_LOG表的记录

genStableLogFile

    private File genStableLogFile(long instanceId) {
        String path = genLogFilePath(instanceId, true);
        int lockId = ("stFileLock-" + instanceId).hashCode();
        try {
            segmentLock.lockInterruptibleSafe(lockId);

            return localTransactionTemplate.execute(status -> {

                File f = new File(path);
                if (f.exists()) {
                    return f;
                }

                try {
                    // 创建父文件夹（文件在开流时自动会被创建）
                    FileUtils.forceMkdirParent(f);

                    // 本地存在数据，从本地持久化（对应 SYNC 的情况）
                    if (instanceId2LastReportTime.containsKey(instanceId)) {
                        try (Stream<LocalInstanceLogDO> allLogStream = localInstanceLogRepository.findByInstanceIdOrderByLogTime(instanceId)) {
                            stream2File(allLogStream, f);
                        }
                    }else {

                        FileLocation dfl = new FileLocation().setBucket(Constants.LOG_BUCKET).setName(genMongoFileName(instanceId));
                        Optional<FileMeta> dflMetaOpt = dFsService.fetchFileMeta(dfl);
                        if (!dflMetaOpt.isPresent()) {
                            OmsFileUtils.string2File("SYSTEM: There is no online log for this job instance.", f);
                            return f;
                        }

                        dFsService.download(new DownloadRequest().setTarget(f).setFileLocation(dfl));
                    }
                    return f;
                }catch (Exception e) {
                    CommonUtils.executeIgnoreException(() -> FileUtils.forceDelete(f));
                    throw new RuntimeException(e);
                }
            });
        }finally {
            segmentLock.unlock(lockId);
        }
    }

    private static String genLogFilePath(long instanceId, boolean stable) {
        if (stable) {
            return OmsFileUtils.genLogDirPath() + String.format("%d-stable.log", instanceId);
        }else {
            return OmsFileUtils.genLogDirPath() + String.format("%d-temporary.log", instanceId);
        }
    }    

genStableLogFile它先判断该server是否有存储该任务实例的日志文件(~/powerjob/server/online_log/%d-stable.log)，有则直接返回；否则判断该instanceId2LastReportTime是否包含该任务实例，包含则从localInstanceLogRepository拉取日志然后写入到文件；不包含则通过dFsService.fetchFileMeta拉取元信息，然后下载到本地再返回

相关表结构

LOCAL_INSTANCE_LOG

CREATE TABLE PUBLIC.LOCAL_INSTANCE_LOG (
    ID BIGINT NOT NULL AUTO_INCREMENT,
    INSTANCE_ID BIGINT,
    LOG_CONTENT CHARACTER VARYING,
    LOG_LEVEL INTEGER,
    LOG_TIME BIGINT,
    WORKER_ADDRESS CHARACTER VARYING(255),
    CONSTRAINT CONSTRAINT_8 PRIMARY KEY (ID)
);
CREATE INDEX IDXPJ6CD8W5EAW8QBKMD84I8KYS7 ON PUBLIC.LOCAL_INSTANCE_LOG (INSTANCE_ID);
CREATE UNIQUE INDEX PRIMARY_KEY_8 ON PUBLIC.LOCAL_INSTANCE_LOG (ID);

powerjob_files

CREATE TABLE `powerjob_files` (
  `id` bigint NOT NULL AUTO_INCREMENT COMMENT 'ID',
  `bucket` varchar(255) COLLATE utf8mb4_general_ci NOT NULL COMMENT '分桶',
  `name` varchar(255) COLLATE utf8mb4_general_ci NOT NULL COMMENT '文件名称',
  `version` varchar(255) COLLATE utf8mb4_general_ci NOT NULL COMMENT '版本',
  `meta` varchar(255) COLLATE utf8mb4_general_ci DEFAULT NULL COMMENT '元数据',
  `length` bigint NOT NULL COMMENT '长度',
  `status` int NOT NULL COMMENT '状态',
  `data` longblob NOT NULL COMMENT '文件内容',
  `extra` varchar(255) COLLATE utf8mb4_general_ci DEFAULT NULL COMMENT '其他信息',
  `gmt_create` datetime NOT NULL COMMENT '创建时间',
  `gmt_modified` datetime DEFAULT NULL COMMENT '更新时间',
  PRIMARY KEY (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=6 DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_general_ci;

小结

• PowerJob的worker端的OmsServerLogger的process方法调用的是OmsLogHandler的submitLog方法，它每次先判断logQueue大小是否大于REPORT_SIZE(1024)，是则启动logSubmitter线程，否则放入logQueue队列LogSubmitter不断地从logQueue.poll数据，在logs的大小大于等于BATCH_SIZE(20)时通过TransportUtils.reportLogs给server上报日志
• server端的AbWorkerRequestHandler的processWorkerLogReport接收WorkerLogReportReq，执行processWorkerLogReport0方法，它执行的是instanceLogService.submitLogs；InstanceLogService的submitLogs是个异步方法，它将InstanceLogContent转换为LocalInstanceLogDO，然后执行localInstanceLogRepository.saveAll保存；server端有两份数据源，一份是mysql，一份是h2，而localInstanceLog存储到的是h2的LOCAL_INSTANCE_LOG表
• 另外server端在任务实例结束时会执行InstanceManager的processFinishedInstance方法，它会延时60s执行instanceLogService.sync(instanceId)；sync方法先通过genStableLogFile将日志持久化到server端的本地日志文件，接着将该任务实例日志的元信息(哪个任务实例、在哪个server、本地日志文件的路径)存储到dFsService(它有oss、gridfs、minio、mysql四种实现，具体看server的配置文件是启动哪个，如果是mysql则是存储到powerjob_files表中)，最后通过localInstanceLogRepository.deleteByInstanceId清空该任务实例在h2中的LOCAL_INSTANCE_LOG表的记录