HTCondor Project List Archives



[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-devel] group quotas are confusing



It looks bad in the manual.

http://research.cs.wisc.edu/condor/manual/v7.7/3_4User_Priorities.html#SECTION00447000000000000000

+AccountingGroup = "group_physics"

disagrees with

http://research.cs.wisc.edu/condor/manual/v7.7/3_4User_Priorities.html#SECTION00448000000000000000

+AccountingGroup = "<group>.<subgroup>.<user>"

Note: "<group>[.<subgroup>*].<user>" would be better.

"Note that a job specifying a value for the AccountingGroup ClassAd attribute that lacks at least one period in the specification will cause the job to not be considered part of a group when negotiating, even if the group name (highest within the hierarchy) has a quota. Furthermore, there will be no warnings that the group quota is not in effect for the job, as this syntax defines group accounting."

This note is not as bold as it used to be.

http://research.cs.wisc.edu/condor/manual/v7.5/3_4User_Priorities.html#SECTION00448000000000000000

"_NOTE_: An accounting group value lacking the period will cause the job to not be considered part of the group when negotiating, even if the group name has a quota. Furthermore, there will be no warnings that the group quota is not in effect for the job, as this syntax defines group accounting."

And given that everyone trips over this at least once, it should be even more bold as a huge gotcha.

The example disagreement possibly goes back to 6.7 manual.

http://research.cs.wisc.edu/condor/manual/v6.7/3_5User_Priorities.html#SECTION00456000000000000000

Best,


matt

On 03/20/2012 03:57 PM, Lans Carstensen wrote:
You need to set AccountingGroup to be "GROUPNAME.OWNER", e.g.

+AccountingGroup = "nwp_few.user"

-- Lans Carstensen

On Tue, Mar 20, 2012 at 12:48 PM, Nathan Panike<nwp@xxxxxxxxxxx>  wrote:
I was testing out preemption, so I set the following
configuration: The idea is that I could submit  some jobs in the
"nwp_few" accounting group which would then begin running, then start
some up in the "nwp_lots" accounting group, and see them preempt the
jobs in the "nwp_few" group. Unfortunately, now all I get is "Reason for
last match failure: group quota exceeded" for every job.

What am I missing?

Nathan Panike

JAVA_BENCHMARK_TIME = 0
MAIL = /bin/mail
# If I disable this, I can run jobs
GROUP_NAMES = nwp_lots, nwp_few
GROUP_QUOTA_DYNAMIC_nwp_lots = 1.0
GROUP_QUOTA_DYNAMIC_nwp_few = 1.0
GROUP_PRIO_FACTOR_nwp_lots = 1.0
GROUP_PRIO_FACTOR_nwp_few = 100.0
GROUP_ACCEPT_SURPLUS = True
SLOT_TYPE_1=cpus=32
NUM_SLOTS=1
NUM_SLOTS_TYPE_1 = 1
SLOT_TYPE_1_PARTITIONABLE=true
NUM_CPUS = 32
PREEMPTION_REQUIREMENTS = True

A submit file looks like

executable = /bin/sleep
arguments = 300
universe = vanilla
output = dslots.$(cluster).$(process).out
log = dslots.log
+AccountingGroup = "nwp_few"
error = dslots.$(cluster).$(process).err
request_memory = 4
request_cpus = 32
queue
request_cpus = 32
queue
request_cpus = 32
queue
request_cpus = 32
queue

and another submit file looks like

executable = 1281.sh
arguments = $(cluster)
output = 1281.$(cluster).$(process).out
error = 1281.$(cluster).$(process).err
log = 1281.log
+AccountingGroup = "nwp_lots"
request_cpus = 1
request_memory = 10
queue
_______________________________________________
Condor-devel mailing list
Condor-devel@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/condor-devel
_______________________________________________
Condor-devel mailing list
Condor-devel@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/condor-devel