Grestemplate1

Added a script to probe nvidia-smi and return list of info to template out gres.conf per file.
No nvidia-smi or no devices means an empty gres file.
tested on mc300

See merge request !76

update the provision_homedir script again



See merge request !77

depending upon node. This is done by a script nvidia-probe.py which probes node for correct information

copy skel files into user home

new functions were tested independently, but the entire j2 script is untested
if skelroot is given, it looks for skelroot/common/skel and copies all within to $home
otherwise, it looks for $home/../common/skel and copies that to $home

See merge request !73

Fixed calculate etc hosts

Chris, the build process could not go further correctly without fixing the etc hosts, please merge it.

See merge request !75

allow provision_slurm to take a path to the lockfile so its on a directory share…

…d between managment nodes

See merge request !71

only try to install nhc if its not already isntalled



See merge request !70

only install munge if it isn't already installed



See merge request !69

allow provision_slurm to take a path to the lockfile so its on a directory shared between managment nodes

update provision_homedir, pass the mntpt variable to indicate where homedirectories are mounted



See merge request !68

Gpu



See merge request !67

update the mellanox role so drivers will build support for new kernels. Also new driver version



See merge request !66

tweaking the extra packages. Make sure our experimental mellanox repo is disable…

…d. Clean the cache before installation (as disks may fill up)

See merge request !64

tweaking the extra packages. Make sure our experimental mellanox repo is disabled. Clean the cache before installation (as disks may fill up)

update the extra pacakges role to make it easier to cluster specific package lists

See merge request !63

Nfs client

See merge request !62

  1) attempts in nfs-client to notify two handlers at once (this didn't work)
  2) attempts to mount a file system twice (I don't know why we included this in the first place, mounts should work first time)
  3) checks using mountpoint -q (this a) didn't work when updating a mount b) only checked the first mount)

remove m3_networking as it is highly site specific

See merge request !61

allow us to make a filesystem on an ephemeral disk as well as using a cinder vol…

…ume supplied with ansible metadata

See merge request !57