Conference call notes 20170522

(back to Conference calls)

Notes on the 77th EasyBuild conference call, Monday May 22th 2017 (5.30pm - 6.30pm CET)

Attendees

Alphabetical list of attendees (6):

• Fotis Georgatos (Illumina, UK)
• Kenneth Hoste (HPC-UGent, Belgium)
• Alan O'Cais (JSC, Germany)
• Davide Vanzo (Vanderbilt University)
• Adam Huffman (Francis Crick Institute, UK)
• Åke Sandgren (Umeå University, Sweden)

Agenda

• outlook to EasyBuild v3.3.0
  • short overview of already merged changes (https://github.com/hpcugent/easybuild-framework/milestone/46?closed=1)
  • support for renaming sources on download (https://github.com/hpcugent/easybuild-framework/pull/2223)
  • support for detecting loaded modules (https://github.com/hpcugent/easybuild-framework/pull/2228)
• update on working towards SHA256 checksums in all easyconfigs
  • support for SHA256 added in https://github.com/hpcugent/easybuild-framework/pull/2215
  • now what...
• Q&A

Notes

SHA256 checksums

• next steps:
  • start enforcing SHA256
  • script to automate adding of checksums
    • only for new/modified files in PR to make the transition gradual
      • requires detection which files were changed, cfr. https://travis-ci.org/hpcugent/easybuild-easyconfigs/jobs/234717293#L856
    • Alan: we could also automatically inject checksums in archived easyconfigs
    • Åke: may be easier with new format for source specs, have checksums right next to sources?
      • cfr. https://github.com/hpcugent/easybuild-framework/pull/2223#issuecomment-302707125
  • how strictly do we verify checksums in PRs to go towards 100% SHA256 coverage?
    • let Kenneth issue PRs using SHA256 checksums for his sources repo?

Other

• Tkinter easyblock
• Alan: Singularity & MPI
  • works well with recent OpenMPI 2.x
  • Singularity bind-mounts libraries via auto-detection
• Davide: Rmpi with intel toolchains doesn't include patch anymore?
• Åke: Python 3.6.1 install failed because of low stack limit that made Intel compilers fail when building hashlib.sha3
• Davide: permission denied when installing setuptools as extension in Python 3.5.2
  • cfr. https://lists.ugent.be/wws/arc/easybuild/2017-05/msg00038.html
• Fotis: installing 'fat' installations of R/Python/... can be an maintenance issue
  • better to have bare R + basic extensions + user-level?
  • only one kind of R, easier choice between bare vs fat install
  • Fotis: rate of changes/updates should drive design
  • at JSC: Python bar on GCCcore level vs Python extensions on MPI level
    • in terms of performance, this matters for numpy
    • see issue in numpy repo by Damian: https://github.com/numpy/numpy/issues/8823
  • R bundles is already being done for Bioconductor: https://github.com/hpcugent/easybuild-easyconfigs/blob/master/easybuild/easyconfigs/r/R-bundle-Bioconductor
  • Davide: Python installation on low level with setuptools & co, numpy/scipy in separate bundle
  • may also help with fat easyconfigs

2017b toolchains?

• intel/2017.05 is a candidate for intel/2017b
  • https://github.com/hpcugent/easybuild-easyconfigs/pull/4618
• Åke: official word on support for GCC 7.1.0 as base for Intel 2017 update 4?
  • Kenneth: will check with Intel support