[HTCondor-users] Print formats and job cpu statistics : the deeper I look, the worse it gets

[Jeff Templon <templon@xxxxxxxxx>]

Hi,

Whatâs going on here??

$ condor_q -nobatch -pr $HOME/nan3.cpf  -allusers | grep -E "CPU|fgi"                                                                                                                      JOB_ID      User CPUS     MEMREQ        MEMUSE     ST     funCPU       intCPU        fWALL      intWALL      SlotTime Num_Starts   WorkerNode

2115385.0   fgit 64        16.0 GB      7.2 GB      R        0+19:52:05 40+15:55:39   0+19:52:05   2+17:16:13 19:52:05        2 wn-pijl-005.

Print format file:

$ cat nan3.cpf                                                                                                                                                                             SELECT NOSUMMARY

   ClusterId                     AS  JOB_ID  PRINTAS JOB_ID

   Owner                         AS User  WIDTH 4

   CpusProvisioned               AS CPUS

   MemoryProvisioned             AS "    MEMREQ  "   PRINTAS READABLE_MB

   MemoryUsage                   AS "     MEMUSE"    PRINTAS READABLE_MB

   JobStatus                     AS "    ST"         PRINTAS JOB_STATUS

   RemoteUserCpu                 AS "    funCPU"   PRINTAS CPU_TIME

   interval(RemoteUserCpu)       AS "    intCPU"

   RemoteWallClockTime           AS "      fWALL"   PRINTAS CPU_TIME

   interval(RemoteWallClockTime) AS "    intWALL"   WIDTH 12

   interval(ServerTime-JobCurrentStartDate) AS "    SlotTime"

   JobRunCount                    AS 'Num_Starts' WIDTH 4

   splitSlotName(RemoteHost)[1] OR "-" AS "  WorkerNode"  WIDTH -12

1. It looks like PRINTAS CPU_TIME is not PRINTAS, itâs PRINT â it doesnât matter what time attribute I specify, I get the same output value.  Not clear what kind of bug it is - maybe this is the intent, but then it does not belong among PRINTAS statements, see READABLE_MB above - that one is used twice, and the output value is clearly depends on the given attribute value.

2. The thing I am calling SlotTime is correct, because I know when this job most recently started, 19+ hours is correct. RemoteWallClockTime may be correct, but using the recommended recipe to to get the most current runâs WallClockTime is failing:  CommittedTime is zero (as are CommittedSlotTime and CumulativeSuspensionTime) for this job.  LastRemoteWallClockTime is identical to RemoteWallClockTime.

3. I recommend a cleanup of the attributes at some point. RemoteUserCpu is only for the current run, but RemoteWallClockTime is cumulative.  Maybe this confusion is responsible for the errors described above, either in my understanding or in the code itself.

Have a good weekend,

JT