CentOS 6 repo EOL makes yum unusable.

[hmacdope]

As detailed in https://github.com/FoldingAtHome/openmm-core/issues/432 and https://github.com/FoldingAtHome/openmm-core/issues/364 the Centos6 EOL makes this image mostly unusable as you can't use yum

Three possible solutions

1. As suggested by @dotsdl Build a new jchodera/omnia-linux-anvil:texlive18-cuda92 based on the image used by conda-forge currently, based on CentOS 8. Then, use it in place of the old image for Linux builds.

2. We can patch the repos in the Dockerfile

RUN sed -i 's/#baseurl/baseurl/g' /etc/yum.repos.d/CentOS-* \
  && sed -i 's/mirrorlist=/#mirrorlist/g' /etc/yum.repos.d/CentOS-* \
  && sed -i 's/mirror.centos.org/vault.centos.org/g' /etc/yum.repos.d/CentOS-*
  && sed -i 's/enabled=1/enabled=0/g' /etc/yum/pluginconf.d/fastestmirror.conf

3. Stop using these all together in favour of official conda-forge cuda images

[dotsdl]

I believe at this point omnia as a channel is no longer maintained, as we've been able to move most or all projects over to conda-forge instead (omnia predates conda-forge).

I think it may make sense to mark this repo as unmaintained.

[Lnaden]

This is still maintained, mostly by me, whenever there is a critical existence failure which forces me to update this repo. The product of the repo is still used by OpenMM to build its dev builds on the The Omnia Dev Recipes, which is almost entirely OpenMM at this point since everything else is on Conda-Forge on its own.

As it stands now, there are a sequence of docker images which are all built up on each other in Dockerhub, separated by branches of this repo (which admittedly is confusing, but i tried to make the branch and image names build on each other)

Stop using these all together in favour of official conda-forge cuda images

I want to start with this bullet point because I have spent alot of time thinking about this; and that won't be exactly viable for all the reasons I listed here: https://github.com/omnia-md/conda-dev-recipes/issues/216

We can patch the repos in the Dockerfile

This seems way too hacky, even for me. sed-editing yum configs seems way to delicate, and would require to edit the base docker file of our image stack anyways, which leads me to the first point...

As suggested by @dotsdl Build a new jchodera/omnia-linux-anvil:texlive18-cuda92 based on the image used by conda-forge currently, based on CentOS 8

I didn't realize they had moved up to CentOS 8, last time i overhauled this, they were on CentOS 7. I would be in favor of this much more than any other option. It would also simplify the build because then I wouldn't have to manually install glibc!

[Lnaden]

@hmacdope Do you all need/would benefit from CentOS 8 based builds? I can block out time to do that this week if so, and if it won't break the OpenMM builds which I'll need to ask about as well. I can't see the linked issues in particular because they are likely private, but I wanted to ask about that.

[hmacdope]

This is not strictly necessary for the Folding@Home OpenMM-core as it is built on quay.io/condaforge/linux-anvil-cuda:11.2 apologies for the confusion. The older version of the docker build system used jchodera/omnia-linux-anvil:texlive18-cuda92 which led me to raising this issue.

I suppose that the image could still be improved by moving to Centos8 independent of FAH needs, but this is dependent on downstream users needs, time required, if it aint broke ... etc etc.

[Lnaden]

Thats fine. I just wanted to check before I committed to doing something. Thanks for the update!