[HTCondor-users] Measuring wallclocktime for grid jobs

[Kartik Vedalaveni <kartik@xxxxxxxxxxx>]

Hi all, 

I am writing a Co-Scheduler for condor and thus the need to measure remote wallclocktime.

I

 looked at condor_history 1128 -l | grep RemoteWallClockTime I get the right wallclocktime which means this information is maintained somewhere.

...

GridResource = "gt5 tusker-gw1.unl.edu/jobmanager-pbs"

ImageSize_RAW = 8

LastSuspensionTime = 0

ExitCode = 0

JobStatus = 4

ShadowBday = undefined

ExecutableSize = 10

RemoteWallClockTime = 454.000000

>

KillSig = "SIGTERM"

StreamOut = false

CurrentTime = time()

RequestMemory = 2048

...

I started measuring wallclocktime by using the timestamps in the log file as I could not find the datastructure containing remotewallclocktime anywhere. The difference of times between the events JOB_EXECUTE and JOB_TERMINATED. This seemed to work fine until I hit a bump where there could be options like MASTER_UPDATE_INTERVAL in condor configuration, that could update the condor logs with the delay. I need to measure wallclock accurately independent of configuration on the clusters on the grid.

Can someone please shed light as to from where I can measure RemoteWallClockTime right after the completion of a job. I am already making use of libcondorapi for reading events from log file.