activiti的job设计浅析

序

activiti自己弄了一个job的机制,解决了集群中job的的若干问题:

• 集群中如何保证一个job只有一个机器执行

• job在处理过程中失败了(其他异常或server重启),这个时候如何处理这些failed job,重新跑起来

ACT_RU_JOB表结构

字段名称	字段描述	数据类型	主键	为空	说明
ID_		varchar(64)	√
REV_		integer		√
TYPE_		varchar(255)
LOCK_EXP_TIME_	锁定释放时间	timestamp(3)		√
LOCK_OWNER_	挂起者	varchar(255)		√
EXCLUSIVE_		boolean		√
EXECUTION_ID_		varchar(64)		√
PROCESS_INSTANCE_ID_		varchar(64)		√
PROC_DEF_ID_		varchar(64)		√
RETRIES_		integer		√
EXCEPTION_STACK_ID_	异常id	varchar(64)		√
EXCEPTION_MSG_	异常信息	varchar(4000)		√
DUEDATE_	到期时间	timestamp(3)		√
REPEAT_	重复	varchar(255)		√
HANDLER_TYPE_	处理类型	varchar(255)		√
HANDLER_CFG_		varchar(4000)		√
TENANT_ID_		varchar(255)		√

查询待处理job

SELECT RES.* FROM
ACT_RU_JOB RES
LEFT OUTER JOIN
ACT_RU_EXECUTION PI ON PI.ID_ = RES.PROCESS_INSTANCE_ID_ 
WHERE (RETRIES_ > 0)
AND (DUEDATE_ is null or DUEDATE_ <= ?)
AND (LOCK_OWNER_ is null or LOCK_EXP_TIME_ <= ?)
AND ( (RES.EXECUTION_ID_ is null) or (PI.SUSPENSION_STATE_ = 1) );

查询条件为

• retries值>0

• duedate为空或者duedate小于当前

• lock_owner为空或者lock_exp_time小于当前

抢占任务

AcquireJobsCmd

protected void lockJob(CommandContext commandContext, JobEntity job, String lockOwner, int lockTimeInMillis) {    
    job.setLockOwner(lockOwner);
    GregorianCalendar gregorianCalendar = new GregorianCalendar();
    gregorianCalendar.setTime(commandContext.getProcessEngineConfiguration().getClock().getCurrentTime());
    gregorianCalendar.add(Calendar.MILLISECOND, lockTimeInMillis);
    job.setLockExpirationTime(gregorianCalendar.getTime());    
  }

设置lock_owner以及lock_exp_time
lockTimeInMillis比如5 60 1000,5分钟

job执行成功

private void removeJobs() {
    for (Job job: getJobs()) {
      ((JobEntity) job).delete();
    }
  }

从job表中删除

job执行失败

if (activity == null || activity.getFailedJobRetryTimeCycleValue() == null) {
      log.debug("activitiy or FailedJobRetryTimerCycleValue is null in job " + jobId + "'. only decrementing retries.");
      job.setRetries(job.getRetries() - 1);
      job.setLockOwner(null);
      job.setLockExpirationTime(null);
      if (job.getDuedate() == null) {
        // add wait time for failed async job
        job.setDuedate(calculateDueDate(commandContext, processEngineConfig.getAsyncFailedJobWaitTime(), null));
      } else {
        // add default wait time for failed job
        job.setDuedate(calculateDueDate(commandContext, processEngineConfig.getDefaultFailedJobWaitTime(), job.getDuedate()));
      }
      
    }

• retries减1

• 清空lock_owner

• 清空lock_exp_time

• 重新设置due_date，延时重试

/** define the default wait time for a failed async job in seconds */
  protected int asyncFailedJobWaitTime = 10;