azkaban 中对于任务运行日志的处理

文章目录

• • azkaban任务日志处理方式
  • azkaban日志mysql表信息
  • 源码中逻辑处理流程

azkaban任务日志处理方式

azkaban是将任务的运行日志存储在mysql 表中的，根据任务的exec_id,name,attempt，即可获取对应任务的日志，当日志文件比较大时，日志将按照50KB的规格来进行分段储存，每一段日志都插入一条数据，详细内容如下文。

azkaban日志mysql表信息

mysql表的schema如下：

CREATE TABLE `execution_logs` (
  `exec_id` int(11) NOT NULL,
  `name` varchar(640) NOT NULL,
  `attempt` int(11) NOT NULL,
  `enc_type` tinyint(4) DEFAULT NULL,
  `start_byte` int(11) NOT NULL,
  `end_byte` int(11) DEFAULT NULL,
  `log` longblob,
  `upload_time` bigint(20) DEFAULT NULL,
  PRIMARY KEY (`exec_id`,`name`,`attempt`,`start_byte`),
  KEY `ex_log_attempt` (`exec_id`,`name`,`attempt`),
  KEY `ex_log_index` (`exec_id`,`name`),
  KEY `ex_log_upload_time` (`upload_time`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

• exec_id 任务的执行id
• name 任务flow的名称
• attempt 重试次数
• enc_type 日志编码类型（plain/gzip）
• log 日志文件内容
• upload_time 日志上传时间

源码中逻辑处理流程

private void uploadLogFile(final DatabaseTransOperator transOperator, final int execId,
    final String name,
    final int attempt, final File[] files, final EncodingType encType)
    throws SQLException {
  // 50K buffer... if logs are greater than this, we chunk.
  // However, we better prevent large log files from being uploaded somehow
  final byte[] buffer = new byte[50 * 1024];
  int pos = 0;
  int length = buffer.length;
  int startByte = 0;
  try {
    for (int i = 0; i < files.length; ++i) {
      final File file = files[i];

      final BufferedInputStream bufferedStream =
          new BufferedInputStream(new FileInputStream(file));
      try {
        int size = bufferedStream.read(buffer, pos, length);
        while (size >= 0) {
          if (pos + size == buffer.length) {
            // Flush here.
            uploadLogPart(transOperator, execId, name, attempt, startByte,
                startByte + buffer.length, encType, buffer, buffer.length);

            pos = 0;
            length = buffer.length;
            startByte += buffer.length;
          } else {
            // Usually end of file.
            pos += size;
            length = buffer.length - pos;
          }
          size = bufferedStream.read(buffer, pos, length);
        }
      } finally {
        IOUtils.closeQuietly(bufferedStream);
      }
    }

    // Final commit of buffer.
    if (pos > 0) {
      uploadLogPart(transOperator, execId, name, attempt, startByte, startByte
          + pos, encType, buffer, pos);
    }
  } catch (final SQLException e) {
    logger.error("Error writing log part.", e);
    throw new SQLException("Error writing log part", e);
  } catch (final IOException e) {
    logger.error("Error chunking.", e);
    throw new SQLException("Error chunking", e);
  }
}

private void uploadLogPart(final DatabaseTransOperator transOperator, final int execId,
    final String name,
    final int attempt, final int startByte, final int endByte,
    final EncodingType encType,
    final byte[] buffer, final int length)
    throws SQLException, IOException {
  final String INSERT_EXECUTION_LOGS = "INSERT INTO execution_logs "
      + "(exec_id, name, attempt, enc_type, start_byte, end_byte, "
      + "log, upload_time) VALUES (?,?,?,?,?,?,?,?)";

  byte[] buf = buffer;
  if (encType == EncodingType.GZIP) {
    buf = GZIPUtils.gzipBytes(buf, 0, length);
  } else if (length < buf.length) {
    buf = Arrays.copyOf(buffer, length);
  }

  transOperator.update(INSERT_EXECUTION_LOGS, execId, name, attempt,
      encType.getNumVal(), startByte, startByte + length, buf, DateTime.now()
          .getMillis());
}

标签：运行日志,attempt,name,buffer,任务,azkaban,int,length,final
来源： https://blog.51cto.com/u_15278282/2933645