Alexander,
Thanks for the feedback. I saw a problem on another site. I suspect
something is not generic enough in the diagnose command parser. I add
Cedric to CC, as he is not register on LCG-ROLLOUT.
Probably, it'd be a great help if you could send him (let me in CC but
probably this is not necessary to send the whole list) a copy of your
maui.cfg and the output of the following commands :
- diagnose -r
- diagnose -n
Cheers,
Michel
--On vendredi 27 juillet 2007 22:23 +0300 Alexander Piavka
<[log in to unmask]> wrote:
>
> Hi Michel,
>
> Thank you for the new plugin
>
> I've just tried the new maui plugin on a PPS site which has total 6 cpu
> slots
> and one slot is always reserved to ops & dteam sam tests:
> --------------------------------------------------------
> SRCFG[sam] FLAGS=SPACEFLEX
> SRCFG[sam] TASKCOUNT=1 RESOURCES=PROCS:1
> SRCFG[sam] PERIOD=INFINITY
> SRCFG[sam] GROUPLIST=ops,opssgm,dteam,dteamsgm,dteamprd
> --------------------------------------------------------
>
> And it indeed reports correctly for non ops & dteam VOs
> --------------------------------------------------------
> dn:
> GlueCEUniqueID=ce01-pps.bgu.ac.il:2119/blah-pbs-alice,mds-vo-name=local,o
> =grid
> GlueCEInfoLRMSVersion: 2.1.6
> GlueCEInfoTotalCPUs: 5
> GlueCEStateFreeCPUs: 0
> GlueCEPolicyMaxCPUTime: 2880
> GlueCEPolicyMaxWallClockTime: 4320
> GlueCEStateStatus: Production
> --------------------------------------------------------
> while with lcg-info-dynamic-pbs it was
> --------------------------------------------------------
> dn:
> GlueCEUniqueID=ce01-pps.bgu.ac.il:2119/blah-pbs-alice,mds-vo-name=local,o
> =grid
> GlueCEInfoLRMSVersion: 2.1.6
> GlueCEInfoTotalCPUs: 6
> GlueCEStateFreeCPUs: 1
> GlueCEPolicyMaxCPUTime: 2880
> GlueCEPolicyMaxWallClockTime: 4320
> GlueCEStateStatus: Production
> --------------------------------------------------------
>
> BUT for ops & dteam VOs it also reports:
> GlueCEInfoTotalCPUs: 5
> GlueCEStateFreeCPUs: 0
> while for me is seems more correct output should be:
> GlueCEInfoTotalCPUs: 6
> GlueCEStateFreeCPUs: 1
>
> I've changed the reservation to
> --------------------------------------
> SRCFG[ops] FLAGS=SPACEFLEX
> SRCFG[ops] TASKCOUNT=1 RESOURCES=PROCS:1
> SRCFG[ops] PERIOD=INFINITY
> SRCFG[ops] OWNER=CLASS:ops
> --------------------------------------
> meaning that 1 cpu slot is reserved only for ops VO queue
> so for ops VO GlueCEInfoTotalCPUs=6 should be reported
> BUT still the same result
> -------------------------------------
> dn:
> GlueCEUniqueID=ce01-pps.bgu.ac.il:2119/blah-pbs-ops,mds-vo-name=local,o=g
> rid
> GlueCEInfoLRMSVersion: 2.1.6
> GlueCEInfoTotalCPUs: 5
> GlueCEStateFreeCPUs: 0
> GlueCEPolicyMaxTotalJobs: 6
> GlueCEPolicyMaxRunningJobs: 2
> GlueCEPolicyMaxCPUTime: 30
> GlueCEPolicyMaxWallClockTime: 180
> GlueCEStateStatus: Production
> -------------------------------------
>
> Another issue is performance, see time results of 3 sample runs:
> ----------------
> for lcg-info-dynamic-pbs
>
> real 0m0.215s
> user 0m0.070s
> sys 0m0.090s
>
> real 0m0.204s
> user 0m0.090s
> sys 0m0.050s
>
> real 0m0.318s
> user 0m0.090s
> sys 0m0.070s
> ----------------
> for lcg-info-dynamic-maui
>
> real 0m1.296s
> user 0m0.390s
> sys 0m0.150s
>
> real 0m1.058s
> user 0m0.440s
> sys 0m0.110s
>
> real 0m1.301s
> user 0m0.450s
> sys 0m0.120s
> ----------------
>
> Regrads
> Alex
>
> On Fri, 27 Jul 2007, Michel Jouvin wrote:
>
>> Hi,
>>
>> We just released first version of a new GIP plugin intended to replace
>> lcg-info-dynamic-pbs. This new plugin uses MAUI command 'diagnose' to
>> collect data about CE usage instead of Torque 'qstat'.
>>
>> The reason for this new plugin is to be able to correctly publish CE
>> information when using advanced MAUI features like Standing Reservations
>> (this allow to reserve some job slots for certain types of jobs). With
>> lcg-info-dynamic-pbs, if you declare standing reservations, they appear
>> as free job slots even if they are not accesible by the queue. The new
>> plugin properly handle that. It could be extended in the future to
>> support other advanced MAUI features. This plugin should work properly
>> even with a basic MAUI configuration without any advanced features used.
>>
>> Using this new plugin is pretty simple :
>>
>> - Download RPM from
>> <http://quattorsrv.lal.in2p3.fr/packages/glite-3.0.0-i386/externals/lcg-
>> info-dynamic-maui-1.1.0-2.noarch.rpm>
>>
>> - Install RPM on your CE (it can coexist with lcg-info-dynamic-pbs)
>>
>> - Edit /opt/lcg/var/gip/plugin/lcg-info-dynamic-ce to use the new plugin.
>> Command line should be (replace grid10... by your CE host name) :
>>
>> /opt/lcg/libexec/lcg-info-dynamic-maui -l
>> /opt/lcg/var/gip/ldif/static-file-CE-pbs.ldif -h grid10.lal.in2p3.fr
>>
>> - After checking everything is ok, you can uninstall lcg-info-dynamic-pbs
>> as the new plugin RPM advertizes lcg-info-dynamic-pbs feature and thus
>> satisfy CE_Torque metapackage dependencies.
>>
>> If you are lucky enough to use Quattor (!!!), the new plugin will be put
>> in standard templates this week-end and you just need to define variable
>> GIP_CE_USE_MAUI as true in your cluster or site templates.
>>
>>
>> If you give it a try and have any problem, please contact plugin author,
>> Cédric Duprilot <[log in to unmask]>.
>>
>> Cheers,
>>
>> Michel
>>
>> *************************************************************
>> * Michel Jouvin Email : [log in to unmask] *
>> * LAL / CNRS Tel : +33 1 64468932 *
>> * B.P. 34 Fax : +33 1 69079404 *
>> * 91898 Orsay Cedex *
>> * France *
>> *************************************************************
>>
*************************************************************
* Michel Jouvin Email : [log in to unmask] *
* LAL / CNRS Tel : +33 1 64468932 *
* B.P. 34 Fax : +33 1 69079404 *
* 91898 Orsay Cedex *
* France *
*************************************************************
|
