Closed BHSDuncan closed 1 month ago
@BHSDuncan we understand your concern, we used to have ubuntu/nvidia user all the time because of that we didn't this issue. I will verify with different user and push the fix and let you know.
The reason being changed from /tmp
to ansible_user_dir
is we saw some issues when we try to use /tmp
like permissions.
@BHSDuncan I have changed to /tmp
and tested with different user on remote machines. works for me, please pull the latest and let us know if you still hit any issues.
Thanks
@BHSDuncan closing this issue as it's fixed
The
k8s-install.yaml
file used inplaybooks/
makes use ofansible_user_dir
, but when doing a remote install, unless the username used for the remote is the same on the localhost (i.e. is already a user on the local system from which the playbook is kicked off), the playbook will fail with a permissions issue around line 276.I verified this by creating a user on a local machine that kicks off the playbook (in fact, just a user dir in
/home/
with the appropriate name and setting r/w permissions temporarily). Uninstalling and then re-running the playbook then works.It looks like the change was done last year sometime: https://github.com/NVIDIA/cloud-native-stack/commit/55bdd53be90398b23dff1da95681d6afc88bab44