Mailing List ArchivesPublic Access |
![[Computer Systems Lab]](https://www-auth.cs.wisc.edu/pics/csl_logo.gif){kind=logo}
|
Mailing List ArchivesPublic Access |
|
Hi Brain, Thanks for the clarification.Would you please read the snippet of text below. It tells about a problem with the way CPU counts are calculated in the current version (3.2) of that BDII code. I've had to fix it.
Can you tell me who maintains that BDII; perhaps they want to include this?
Cheers,
Ste
--- WHAT I WROTE ABOUT THE BDII ---
Fix the BDII provider to use real slots provided, not just
detected CPUS
The BDII provider, /var/lib/bdii/gip/provider/htcondor-ce-provider, that
comes with the system is a bit naive. It just says that if a node that
has, say, 16 possible hyperthreads (cores * 2, what have you) then it
must have 16 job slots on it. This is not so. Some of our 16 hyperthread
nodes can only have max 12 jobs running, due to memory.
To fix this in a set-up like ours, with partitionable slots, this patch is used (see below). What does this mean? It means Iâm counting up only the slots with ID = 1. This tells me the allocation, which the right logic for this job.
To put in the patch, around line 129 of /var/lib/bdii/gip/provider/htcondor-ce-provider, change this constraint 'State=!="Owner"' to 'SlotTypeID == 1'. And, from line 129 to line 145, change all âDetectedCpusâ to âTotalSlotCpusâ. You will then get the right counts in the BDII output.
On 14/02/2019 16:21, Brian Lin wrote:
Stephen, Sorry, I forgot to address that earlier! The version of HTCondor-CE in the 8.8.x and 8.9.x HTCondor repositories will include the BDII package. - Brian On 2/14/19 10:18 AM, Stephen Jones wrote:Hi Brian, The trouble is that we might be going to have quite a few sites taking on HTCondor-CE in the UK, since the APEL accounting is now ready. https://twikiai07.cern.ch/twiki/bin/view/LCG/HtCondorCeAccounting And CREAM will be gone in under two years. For the time being, we will want that htcondor-ce-bdii package. Will these 8.8.x and 8.9.x repositories have the full suite, including the bdii, or will we still need to make it from the source code? Cheers, Ste On 14/02/2019 15:32, Brian Lin wrote:Hi Stephen, We don't ship htcondor-ce-bdii in the OSG repositories since BDII isn't used in the OSG. However, we do ship HTCondor-CE in the HTCondor repositories but there have been some build inefficiencies that caused a hiccup in getting the latest versions into our yum repositories. We should have HTCondor-CE 3.2.1 available in the 8.8.x and 8.9.x repositories either today or tomorrow! - Brian On 2/14/19 8:09 AM, Stephen Jones wrote:Hi, I'm using HTCondor-CE here at our T2 grid site. I did not find htcondor-ce-bdii-3.2.0-1.el7.centos.noarch.rpm in the OSG repo, e.g. http://repo.opensciencegrid.org/osg/3.2/el7/development/x86_64/ Hence I had to make it from scratch (please see below.) This took a lot of time. I'm sure there must be an easier way to get hold of htcondor-ce-bdii-3.2.0-1.el7.centos.noarch.rpm Does anybody know how this product is managed? Cheers, Ste ------------------------------------------------  Making the HTCondor-CE RPMs (to get the BDII rpm) Ordinarily, once might expect the htcondor-ce-bdii rpm to be in the OSG release repositories, for example: http://repo.opensciencegrid.org/osg/3.2/el7/development/x86_64/ Although the htcondor-ce components are there, the bdii component is not. So since the HTCondor-CE BDII rpm is not available, I made the whole system, from source. Here are my notes for doing so. BTW: It might be worth checking at OSG, CERN or elsewhere to see if the BDII RPM is now part of the release. If so, you can get them directly from one of those places. Anyway, this is the run down to make your own. On some CentOS7.n system software development system, make a user called rpmuser and install the rpm, boost-devel and cmake packages. Clone the htcondor-ce git repo. # mkdir dev # cd dev # git clonehttps://github.com/opensciencegrid/htcondor-ce Find the commit tag associated with 3.2 (Note: It is a9c1104febcbaf20a8380284d9d1213eb504afa5) # cd htcondor-ce/ # git log > /tmp/log # vi /tmp/log Prepare the source material; # cd .. # mv htcondor-ce/ htcondor-ce-3.2.0 # tar -cvf htcondor-ce-3.2.0.tar htcondor-ce-3.2.0/ # gzip htcondor-ce-3.2.0.tar # cp htcondor-ce-3.2.0.tar.gz ~/rpmbuild/SOURCES/ # cp htcondor-ce-3.2.0/rpm/htcondor-ce.spec ~/rpmbuild/SPECS/ Make the RPMs # cd ~/rpmbuild/SPECS/ # rpmbuild -ba htcondor-ce.spec The rpms will wind up in ~/rpmbuild/RPMS/noarch/htcondor-ce-* # ls -rt ~/rpmbuild/RPMS/noarch/htcondor-ce-* htcondor-ce-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-bdii-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-view-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-condor-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-pbs-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-lsf-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-sge-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-slurm-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-bosco-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-client-3.2.0-1.el7.centos.noarch.rpm htcondor-ce-collector-3.2.0-1.el7.centos.noarch.rpm Put these in the local repo. Special note: whenever you put an rpm into the local repo, you have to run this command in that directory. # createrepo .
-- Steve Jones sjones@xxxxxxxxxxxxxxxx Grid System Administrator office: 220 High Energy Physics Division tel (int): 43396 Oliver Lodge Laboratory tel (ext): +44 (0)151 794 3396 University of Liverpool http://www.liv.ac.uk/physics/hep/