How to gather the job startTime info in PAI?

[JianfeiHu]

Organization Name: MSRAIT
Short summary about the issue/question:我试图统计用户对系统的使用情况，计算用户对资源的使用时间,从REST服务器获取作业列表，发现只有createdTime和completedTime。createdTime是用户提交作业的时间，而不是job实际运行的时间。PAI是否记录作业实际运行的时间？

OpenPAI Environment:

• OpenPAI version:0.8.2
• Cloud provider or hardware configuration:
• OS (e.g. from /etc/os-release):
• Kernel (e.g. uname -a):
• Hardware (e.g. core number, memory size, storage size, GPU type etc.):
• Others:

[yqwang-ms]

FrameworkLauncher记录了YARN app实际运行时间，
private Long applicationLaunchedTimestamp;
private Long applicationCompletedTimestamp;

如：

restserver暴露出来的是：
appLaunchedTime
appCompletedTime
你要不试试这两个？

（这只是最后一次retry的job的起始和终止时间）

[JianfeiHu]

@yqwang-ms我在使用这种方法获得appLaunchedTime，但是对一部分job查询时会出现这种情况。
http://*5:9086/v1/Frameworks/nni_exp_NftmjP6l_trial_jOcTK

I'm changing the IP to secured, you guys can contact offline for the job ip.

[yqwang-ms]

这个job应该被人删掉了，你看看nni那边是不是有删除逻辑

[scarlett2018]

I'm closing the issue as the original question had been resolved. @JianfeiHu - please follow up with @yds05 or Chi for the nni job log issue.

[yqwang-ms]

@JianfeiHu We have just refined the "作业实际开始运行的时间" in frameworkcontroller (so will be supported in Pure K8S PAI soon).

For more details, please check:
microsoft/frameworkcontroller#35

For a job attempt:
WholeDuration = CompletionTime - StartTime
RunningDuration = CompletionTime - RunTime
WaitingDuration = WholeDuration - RunningDuration

For the whole job which may have multiple attempts:
WholeDuration = CompletionTime - StartTime
RunningTime = Sum_all_attempts (CompletionTime - RunTime)
WaitingDuration = WholeDuration - RunningDuration