I am trying to install a tool that depends on c++ pstreams libary. I guess I cannot install this myself.

"yum install pstreams-devel" doesn't seem to work.

I am working on luna. I am not sure who I should contact for this request. Any help would be appreciated.

Hi,

There have been a couple of threads on the cBio-cluster issue tracker pertaining to docker policy that I'd like to echo here. It's particularly a big deal now since changes made at docker hub have now complicated workflows for anyone using docker on Hal.

In the first, @tatarsky is trying to gauge what's the best way to operate docker on the Hal cluster. Currently, all nodes have their own separate storage space for previously pulled docker images, which are not accessible cluster wide. So, if I submit a docker job which pulls an image from docker hub into the queue, and it runs on node 1, that image is only stored and accessible to docker on node 1. If I want to re-run that job later, I have to run docker pull again as the first part of my job.

This is a drag, since as of December 15th, the version of docker running on all Hal nodes can no longer pull from docker hub (viz. the second issue for why this is a non-trivial problem).

@tatarsky has heroically figured out a way to avoid a cluster-wide OS upgrade, but currently only a couple of nodes have this fix. If these two nodes could also serve as the host of all pulled docker images that were made available cluster-wide, we could continue to use docker as before while we figure out the least-painful path to an OS or kernel upgrade.

TL;DR

I propose we make a change to the docker image storage policy, by making one (or both) of the currently available nodes host the docker images in its scratch space, and that this space be made available to all nodes on the network for docker jobs.