Bugs encountered when running ansible-playbook controller.yml --tags=luna today.

[jw-26-sudo]

Hi Antoine,

I am trying to create a TrinityX version 16 (non-HA cluster) using Rocky 9.8 OS via Proxmox VM, for Ubuntu family OS image diskfull support.

I an writing to report some bugs encountered when running bash prepare, then ansible-playbook controller.yml --tags=luna today.

1. The first bug is the directory: /trinity/local/etc/ssl/ does not exist. I had to do a mkdir -p /trinity/local/etc/ssl/ and rerun playbook.

2. The second bug encountered is based on the below screenshots:

Please assist to check on the issues and provide advice on where I have made the mistakes.

Thank you.

Regards,
Jing Wei

[aphmschonewille]

Hi JW. It looks you are cherry picking. If you like to just have luna, run this:

ansible-playbook controller.yml --tags=repos,packages,ssl,ssl-cert,resolv,bind,luna

However these are very tuned steps. I'd recommend to run a full playbook and tune all.yml to add or exclude parts.

[jw-26-sudo]

Hi Antoine,

I have followed your suggestion, but unable to:

1. view /usr/local/sbin/obol group list
2. obol user add -P --groups admins <name_to_give>
3. I ran ansible-playbook controller.yml --tags=obol >> Failed binding to ldap while viewing /usr/local/sbin/obol group list and using obol user add command.
4. The ondemand and ondemand-gems are not installed.

I hope you can enlighten me with this "However these are very tuned steps. I'd recommend to run a full playbook and tune all.yml to add or exclude parts."

I have previously done all necessary tunings and ran a full playbook, but unable to acheive diskfull/diskless for Ubuntu OS compute images.

Appreciate if you can provide some workaround solutions. 

Thank you.

Regards,
Jing Wei

[aphmschonewille]

Hi Jing Wei,
Sorry for the slow reply, and let me correct myself first: the tag list I gave you was missing
trix-tree. That's the role that creates the Trinity directory tree, including
/trinity/local/etc/ssl — so you'd have hit your first bug again even following my suggestion.
That also explains why that one was so confusing to debug. In trix-tree, the task that sets the
paths is tagged always, so {{ trix_ssl }} resolves on any run and everything looks right — but
the tasks that actually create the directories are not tagged that way. On a cherry-picked run the
paths therefore exist only as variables, and a later role tries to write into a directory nobody
made.

Rather than hand-picking roles, use the authentication tag
— it's a curated bundle covering
exactly what you need here: init, trix-tree, ssl-cert,
openldap, ds389, obol, sssd.

ansible-playbook controller.yml --tags=repos,packages,resolv,bind,authentication,luna

That covers points 1–3 of your follow-up. --tags=obol on its own fails with "Failed binding to
ldap" because it runs obol without the directory server obol
talks to — openldap or ds389, depending
on trix_use_ds389. The authentication tag runs them together, in the right order.
For point 4, ondemand is a separate tag — --tags=ood (plus ood-apps), and it needs enable_ood: true in all.yml.

Two caveats on the above, which is really what I was getting
at with "run the full playbook": that
list is still hand-picked and I can't promise it's complete — the dependencies between roles aren't
something a tag list expresses well. And note ssl in my original list isn't a real tag at all;
ssl-cert is the one, and it comes with authentication anyway.

On the Ubuntu images — that's the actual problem here, and it's the part I can't help with as it
stands. "Unable to achieve diskfull/diskless" doesn't tell us where it broke. Could you run
ansible-playbook compute-ubuntu.yml as root, on a controller built from a full playbook run,
and paste the failing task and its error? With that we can chase it properly. (Running it as a
normal user fails in a confusing way — the Ansible collections are installed under root's home, so
you get an unrelated "couldn't resolve module" error.)

Also worth updating before you spend more time on this: 16u1 is out now.

Regards,
Antoine

[jw-26-sudo]

Hi Antoine,

Please advise if we can try the following on TrinityX version 16u1:

1. Comment out SLURM associated components, since we do not need SLURM for our cluster testing.

2. Run a full playbook after making necessary amendments/adjustments.

Appreciate if additional advice can be given.

Thank you.

Regards,
Jing Wei

[aphmschonewille]

Just make the all.yml setting for schedulers like this: [] and place # in front of slurm there. It's cleaner as it wont install other slurm related parts.