-
Notifications
You must be signed in to change notification settings - Fork 0
EESSI hackathon Dec'22
Kenneth Hoste edited this page Jan 5, 2023
·
4 revisions
- when: 14-21 Dec 2022
- main goal: focused effort on various tasks in EESSI
-
expectations:
- joining kickoff/sync/show&tell meetings
- spending a couple of hours that week on one or more of the outlined tasks (in group)
- take extensive notes (to integrate into documentation later)
- https://github.com/EESSI/hackathons/tree/main/2022-12
- show & tell meeting:
- slides: (no slides)
- recording: (no recording)
- #hackathon EESSI Slack channel: https://eessi-hpc.slack.com/archives/C02NB46EK9P
- GitHub repo for EESSI hackathon(s): https://github.com/EESSI/hackathons
-
Wed 14 Dec 2022, 09:00 UTC: kickoff
- clarify expectations
- overview of tasks
- getting organised: who works on what, form groups
-
Mon 19 Dec 2022, 09:00 UTC: sync
- quick progress report per group
- briefly discuss next steps
-
Wed 21 Dec 2022, 13:00 UTC: show & tell
- each group briefly demos/presents what they worked on
- outline follow-up steps
- Zoom meeting with breakout rooms
- #hackathon channel in EESSI Slack
- CitC Slurm cluster in AWS
- see https://github.com/EESSI/hackathons/tree/main/2022-12/citc
- if you don't have an account yet, contact Kenneth
- Cyclecloud Slurm cluster in Azure
- see https://github.com/EESSI/hackathons/tree/main/2022-12/cyclecloud-azure
- WIP, Kenneth still needs to set up partitions (together with Hugo?)
- contact Kenneth to get access
NC (Standard_NC6) -> Haswell
*** NCv2 (Standard_NC6s_v2) -> Broadwell
*** HC (Standard_HC44rs)-> Skylake
Dv5 (Standard_D4_v5) -> Ice Lake
*** HBv2 (Standard_HB120rs_v2) -> Rome
HBv3 -> Milan-X
*** HBv4 -> Genoa (Genoa-X when in GA)
- Kenneth
- out-of-office most of Fri 16 Dec
- out-of-office Mon 19 Dec pm
- Alan
- ...
- Thomas
- 14: 10-14, 15: 10-16 (16-17 XALT monthly), 16: 9-15, 19: 9-10, 12:30-15, 20: 9-15, 21: 9-10, 12:30-15
- evenings / weekend
- Bob
- afk on Wed 14 Dec
- Jure
- ...
- Sebastien
- may have some time on Thursday without network in the morning
- several meetings along the day on Friday 16 Dec
- not available on Monday between 10:00 and 13:30
- out-of-office on Wed 21 Dec
- Hugo
- oof on Friday 16 Dec
- Michael
- ...
(see also https://terminplaner4.dfn.de/jNOHCCL2nNTEzrWn)
- joining: Kenneth
- maybe: Thomas, Jure, Alan, Sebastien, Hugo
- notes:
- open PRs:
- SciPy-bundle with foss/2021a (https://github.com/EESSI/software-layer/pull/160)
- OpenFOAM v9 (https://github.com/EESSI/software-layer/pull/195)
- OSU-Micro-Benchmarks (https://github.com/EESSI/software-layer/pull/201)
- snakemake (https://github.com/EESSI/software-layer/pull/197)
- 'mag' Nextflow pipeline (https://github.com/EESSI/software-layer/pull/202)
- open PRs:
- joining: Thomas, Kenneth, Bob
- maybe: Alan, Sebastien, Hugo
- notes
- see https://github.com/EESSI/eessi-bot-software-layer
- particularly issues labeled for hackathon
- discuss/design:
- joining: Alan, Hugo, Michael
- maybe: Thomas, Kenneth, Jure, Bob
- notes
- joining: Hugo
- maybe: Kenneth, Bob
- notes
- joining: Kenneth, Bob
- maybe: Hugo
- notes
- see https://github.com/EESSI/compatibility-layer/pull/155
- related
- new compat layer version: https://github.com/EESSI/compatibility-layer/pull/160
- joining: Jure
- goals:
- apptainer support
- inclusion into eessi (contrib folder? compat layer?)
- probably in scripts/ dir in filesystem-layer repo?
- Interesting work by CSC on containerizing software installations
- CVMFS has a utility to squashfs
- goals:
- joining: Jure
- maybe: Alan, Hugo, Michael, Bob
- notes:
- goals:
- (Jure) script to export software stack to container
- (Bob) data repository
- goals:
- joining: Jure, Thomas, Michael, Kenneth
- excused: Alan, Sebastién
- merged PRs:
- updated EasyBuild hooks for MetaBAT (https://github.com/EESSI/software-layer/pull/203)
- SciPy-bundle with foss/2021a (https://github.com/EESSI/software-layer/pull/160)
- use correct OpenBLAS easyblock to install OpenBLAS/0.3.15-GCC-10.3.0 for
*/generic
+ ignore failing FlexiBLAS tests forppc64le/generic
(https://github.com/EESSI/software-layer/pull/211)
- opened PRs
- opened PRs:
- providing an overview of a PR's status in its description (https://github.com/EESSI/eessi-bot-software-layer/pull/127)
- merged PRs
- easyconfig PR for CUDA samples (https://github.com/easybuilders/easybuild-easyconfigs/pull/16914)
- opened PRs
- Add CUDA support to software_layer (https://github.com/EESSI/software-layer/pull/212)
- 1st step in breaking large PR down to smaller PRs
- draft since step since EESSI NVIDIA GPU support script is missing
- already included in https://github.com/EESSI/software-layer/pull/172, but needs some cleanup
- Would it be possible to install CUDA runtime via EasyBuild?
- Add CUDA support to software_layer (https://github.com/EESSI/software-layer/pull/212)
- individual PRs coming for
- cleaned up NVIDIA GPU support script, docs, testing, Lmod hooks, ...
- ...
- (no action yet)
- see https://github.com/EESSI/hackathons/tree/16_export_software_stack/2022-12
- singularity and apptainer don't want to coexist (at least as packaged by epel)
- because apptainer also ships 'singularity' command (and more stuff)
- apptainer rpm package is aliased to singularity to create additional fun
- uninstalling singularity also uninstalls apptainer
- apptainer is mostly a drop-in replacement for singulairty
- added some logic to create
From:
stanza in image definition based on the host distro- used to be
rocky:linux
nowrockylinux:...
- tested on alma/centos/debian/rhel/rocky/ubuntu
- used to be
- current state of script: ready to contribute to EESSI/filesystem-layer
- joining: Thomas, Michael, Kenneth, Jörg
- excused: ...
- fix for installing Bowtie2 on non-x86_64 merged: https://github.com/easybuilders/easybuild-easyconfigs/pull/16946
- PR for adding R v4.1.0 to EESSI pilot 2021.12 fixed
- https://github.com/EESSI/software-layer/pull/210
- tested on
aarch64/graviton2
- test underway for
ppc64le/power9le
...
- opened PRs:
- providing an overview of a PR's status in its description (https://github.com/EESSI/eessi-bot-software-layer/pull/127)
- see also https://hackmd.io/T1Jln5l2QSyJh0yg3t7Xaw
- need to follow up on this after Christmas
- (Thomas) was caught in other work :-(
- however also related to the bot
- debugging https://github.com/trz42/software-layer/pull/49#issuecomment-1325473297 using https://github.com/EESSI/eessi-bot-software-layer/pull/85 revealed wrong handling of http* proxy settings which causes problems for EB to fetch sources
- described issue in https://github.com/EESSI/eessi-bot-software-layer/pull/85#issuecomment-1361252105
- will work on fix for that first
- providing an overview of a PR's status in its description (https://github.com/EESSI/eessi-bot-software-layer/pull/127)
- (Thomas) general observations:
- while much better prepared to take on some tasks, ended up making very little progress (other urgent work, sickness, ...)
- need to be more self-honest on how much time one can spend
- would prefer to work in group of at least two
- maybe rethink overall concept of "hackathon", rather do sprints from time to time, i.e., focus on making progress with specific tasks, issues, PRs than a group working largely individually on isolated issues
- current format could work well to add software to EESSI
- have to wait for installs to complete anyway
- much less so for implementation work, better to have focused 2-3h pair programming sessions for this
- PR to move utils.sh into scripts dir + make
create_init_tarball.sh
more genericcreate_directory_tarballs.sh
: https://github.com/EESSI/software-layer/pull/213-
ingest-tarball.sh should be extended to also support ingesting
scripts
- similar to update for data: https://github.com/EESSI/filesystem-layer/pull/138
-
ingest-tarball.sh should be extended to also support ingesting
- PR to install CUDA into software-layer
- https://github.com/EESSI/software-layer/pull/212
- some more changes needed, see comments by Alan
-
- also use scripts/ subdir
- requires https://github.com/easybuilders/easybuild-framework/pull/4119
- Michael has also worked on additional PRs that were fleshed out from https://github.com/EESSI/software-layer/pull/172
- idea is to leave PR #172 open for reference, and get everything done in separate smaller PRs
- newer Lmod version also required
- that's planned for next pilot version, see also https://github.com/EESSI/gentoo-overlay/pull/84
- getting started info was added to README @ https://github.com/EESSI/test-suite
- (no progress here... :( )
- same state as on Monday
-
new compat layer
- building new version of compat layer currently still requires root for some things
- blocker for being able to let bot build compat layer
- could be fixed by
- not using Ansible anymore, just use a script instead
- running Ansible in container instead
-
(Bob)
- new repo for
datasets
CVMFS repo: https://github.com/EESSI/datasets- Hugo's PR should be moved to this repo (https://github.com/EESSI/software-layer/pull/170)
- PR to update ingest script for datasets: https://github.com/EESSI/filesystem-layer/pull/138
- new repo for
-
(Thomas)
- should we also look into providing documentation for the software we provide?
-
(Jörg)
- would be interesting to have a way of submitting installation requests (for things already supported in EasyBuild)
- bot could automatically test whether it works in EESSI
- only for trusted contributors
- only one some platforms (bot could be configured to run test builds of "untested" PRs)
- requirement could be to provide a script to test the software (to add to test suite, docs, ...)
- bot could automatically report which installations are missing when a PR is opened