Conference call notes 20181003

(back to Conference calls)

Notes on the 111th EasyBuild conference call, Wednesday Oct 3rd 2018 (17:00 - 18:00 CEST)

Attendees

Alphabetical list of attendees (4):

• Fotis Georgatos (SDSC, Switzerland)
• Kenneth Hoste (HPC-UGent, Belgium)
• Bart Oldeman (Compute Canada)
• Davide Vanzo (Vanderbilt University, US)

Agenda

• update on upcoming EasyBuild v3.7.1
• Q&A

EasyBuild v3.7.1

• framework (https://github.com/easybuilders/easybuild-framework/milestone/59)
  • problems with Java/1.8 wrapper via .modulerc in HMNS setting and with Lmod 6.x: https://github.com/easybuilders/easybuild-framework/issues/2601 We have a good resolution on that by explicitly checking .modulerc{.lua} files.
  • Fotis: should add option to skip module tool version check.
  • Bart working on solution for https://github.com/easybuilders/easybuild-framework/issues/2592 (--try-toolchain overly picky)
• easyblocks (https://github.com/easybuilders/easybuild-easyblocks/milestone/51)
  • Issue with ModuleRC easyblock when using HierarchicalMNS in https://github.com/easybuilders/easybuild-easyblocks/issues/1534 fixed in https://github.com/easybuilders/easybuild-easyblocks/pull/1535
  • TensorFlow easyblock: was downloading binary packages for dependencies such as numpy:
    • Issue https://github.com/easybuilders/easybuild-easyblocks/issues/1536 fixed in https://github.com/easybuilders/easybuild-easyblocks/pull/1537, but needs adapting various easyconfigs to deal with protobuf dependency.
    • Related: TensorFlow forum: https://groups.google.com/a/tensorflow.org/forum/#!forum/build
  • When side-stepping config.guess in ConfigureMake easyblock, we if setting --build (as done in 3.7.0) one also needs --host: https://github.com/easybuilders/easybuild-easyblocks/pull/1532. Fixes building Hypre.
• easyconfigs (https://github.com/easybuilders/easybuild-easyconfigs/milestone/54)
  • TensorFlow easyconfig changes referred to above:
    • https://github.com/easybuilders/easybuild-easyconfigs/pull/6940

Other

• Kenneth:
  • Segmentation faults for Fortran code using large stack memory allocations such as VASP when using the SystemTap mitigation for CVE-2018-14634 (see https://access.redhat.com/security/cve/cve-2018-14634). Updating the kernel does not exhibit the same issue.
  • Slurm: any has experience with monitoring tools to see how well jobs use resources, in particular CPU efficiency?
    • Davide: something is in the works, parsed data in this notebook: https://github.com/nss-data-science-cohort-2/data-question-2-lethal-storm/blob/master/accre-combined.ipynb
    • Fotis: SDSC provides http://renkulab.io (register yourself there, in case you are looking for a Jupyterhub instance, for notebook experimentation).
    • Kenneth will also ask people at Jülich.