Add mounts role with /tmp defaulted to tmpfs on login & compute nodes#888
Conversation
elelaysh
left a comment
There was a problem hiding this comment.
Would it make sense to have a more granular enable/disable?
By adding mount_tmp_enable: true and handling an enable field in a mounts dict value (can be | default(True))
Yeah that's exactly the sort of discussion I was expecting/after, thanks. I was wondering about removing the entire I'll try that, thanks. |
…stackhpc#888) * add DEX support for ondemand * fix merge conflicts * fix lint errors * cleanup PAM install/configure * install DEX and OIDC Apache module in stackhpc builds * fix markdown lint errors * bump CI image * wip; add tmpfs /tmp * add compute-init support for mounts * update compute-init docs for mounts role * fix linter errors * bump CI image * add docs * add mounts[].enabled and mounts_tmp_enabled * update builder config for consistency * bump CI image
…#888) * add DEX support for ondemand * fix merge conflicts * fix lint errors * cleanup PAM install/configure * install DEX and OIDC Apache module in stackhpc builds * fix markdown lint errors * bump CI image * wip; add tmpfs /tmp * add compute-init support for mounts * update compute-init docs for mounts role * fix linter errors * bump CI image * add docs * add mounts[].enabled and mounts_tmp_enabled * update builder config for consistency * bump CI image
mountsrole to support arbitrary mounts via fstab, with compute_init support.mountsrole enabled by default forcomputeandloginnodes.tmpfsof 10% of RAM. This prevents users from filling the root disk.Note that the tmpfs functionality is actually provided by the systemd
tmp.mountunit (masked by default in RockyLinux genericcloud images). However this is automatically triggered by appropriate changes to fstab, and using Ansible'sposix.mountmodule to drive it is is much easier to integrate correctly than via overrrides using the currentsystemdunit. It also allows for a more general solution for additional mount-points which can evolve towards #814. Testing shows changes to the tmpfs size are correctly applied by re-running Ansible.