6.azkban的监控

azkaban自带的监控
flow自带的邮件功能
SLA
总结
写程序监控job情况
监控azkaban的元数据库
使用azkaban API监控
总结

azkaban自带的监控

azkban目前仅仅支持邮件监控,又分两个部分:

flow自带的邮件功能

• First failure
  当flow中出现第一个failure时发邮件
• Flow finished
  在flow执行完后,有错误再出邮件
  发的邮件内容如下:
  

SLA

Service-Level Agreement的缩写，意思是服务等级协议.即azkaban上表示,在什么时候内满足什么条件.SLA的设置:

• FINISH
  表示flow在要什么时间点之前完成,不管是success还是failure.
• SUCCESS
  表示flow要在什么时间点之前执行功能.
  目前azkaban有两种action:
• Emain Action
  不满足SLA条件,则发送EAMIL
• kill Actinon
  不满足SLA条件,则kill flow

总结

不管是flow自带的邮件功能还是SLA,都是flow级别的预警,都不针对单个的job发邮件

写程序监控job情况

写程序监控job情况有两种方式:

监控azkaban的元数据库

写程序查询azkaban的元数据库,监控job的运行情况.以下sql可以查出0点-8点之间所有的flow,job执行情况

select
a.id as project_id,
a.name project_name,
b.flow_id,
c.status flow_status,#50success 70 failed 30 running 80 running with failure 60 killed 
from_unixtime(c.start_time/1000) flow_start_time,
from_unixtime(c.end_time/1000) flow_end_time,
d.job_id,
d.status job_status,
d.start_time job_start_time,
d.end_time job_end_time
from projects a left join 
(select project_id,max(version) version,flow_id from project_flows group by project_id,flow_id) b 
on a.id = b.project_id
left join execution_flows c
on a.id = c.project_id and b.flow_id = c.flow_id and b.version = c.version
and c.start_time>= unix_timestamp( date_format(curdate(),'%Y-%m-%d %H:%i:%S'))*1000  #早上0点
and c.end_time<=unix_timestamp(date_format(date_add(curdate(), interval 8 HOUR),'%Y-%m-%d %H:%i:%S'))*1000
left join execution_jobs d 
on a.id = d.project_id and b.version =d.version and b.flow_id=d.flow_id
and d.start_time>= unix_timestamp( date_format(curdate(),'%Y-%m-%d %H:%i:%S'))*1000  #早上0点
and d.end_time<=unix_timestamp(date_format(date_add(curdate(), interval 8 HOUR),'%Y-%m-%d %H:%i:%S'))*1000
where  a.name in ('E_S1','E_S3')
order by a.name,case  when c.start_time is  null then 9999999999999999999 else c.start_time end

使用azkaban API监控

通过azkaban的aip: http://azkaban.github.io/azkaban/docs/latest/#api-fetch-a-flow-execution job的执行情况.

总结

azkaban自带的监控只支持flow级别的监控,job级别的监控必须通过开发程序实现.

来自为知笔记(Wiz)