jinja with grains

[brad.v...@gmail.com]

We have a grain that shows the installed GPUs on a server.  On one with just the onboard GPU, it looks like:

local:

    ---------------

    gpus:

             |_

               --------------

               model:

                        MGA G200eH3

                vendor:

                        matrox

On a server with added GPUs, it looks like:

local:

    ---------------

    gpus:

             |_

               --------------

               model:

                        MGA G200eH3

                vendor:

                        matrox

             |_

               --------------

               model:

                        GP102GL [Tesla P40]

                vendor:

                        nvidia

             |_

               --------------

               model:

                        GP102GL [Tesla P40]

                vendor:

                        nvidia

I'm trying to test for servers with an Nvidia GPU.

I have so far:

{% if salt['grains.get']('gpus:vendor') == 'nvidia' %}

{%- do salt.log.error('This server has an Nvidia GPU') -%}

{% else %}

{%- do salt.log.error('No GPU on this server') -%}

{% endif %}

and I get the 'No GPU on this server' whether or not there is an Nvidia GPU.  So, I'm obviously doing something wrong.  Any ideas?

[brad.v...@gmail.com]

In the grain display above, what does the |_ indicate?  We have a similar grain lvm which displays as:

local:

     ---------------

     lvm:

           ---------------

          vg_nvme0:

                - lv_nvme0

          vg_sys:

               - lv_audit

               - lv_data

               - lv_home

               - lv_log

               - lv_root

               - lv_tmp

              - lv_var

and we have a state that correctly identifies  when lv_nvme0 exists within vg_nvme0.  I tried to adopt that syntax to the GPUs, but it does not work.  I think the display of the grains is a hint, but I'm not sure what the '|_' indicates when displaying the gpus grains.  Does anyone know?  Thanks!

[sayf.eddi...@gmail.com]

Hello, you have access to grains directly in the jinja template throught the variable `grains`. notice that grains.gpu is a list of objects

try

{% for gpu in grains.gpu %}

  {% if gpu.vendor == 'nvidia' %}

    {%- do salt.log.error('This server has an Nvidia GPU') -%}

  {% else %}

    {%- do salt.log.error('No GPU on this server') -%}

{% endif %}
{% endfor %}

[sayf.eddi...@gmail.com]

+ {% for gpu in grains.gpus %}

-  {% for gpu in grains.gpu %}

[brad.v...@gmail.com]

That works!  I have:

{% for gpu in grains.gpus %}

{% if gpu.vendor == 'nvidia' %}
{%- do salt.log.error('This server has an Nvidia GPU') -%}
{% else %}
{%- do salt.log.error('No GPU on this server') -%}
{% endif %}

{% endfor %}

which prints out:

[ERROR  ] No GPU on this server

[ERROR  ] This server has an Nvidia GPU

[ERROR  ] This server has an Nvidia GPU

Since it hits the matrox GPU first, then the two Nvidia GPUs.  Now, I just need to figure out how to eliminate the "extra" tests.

Wait, I think I just found a simpler method.  I see we have a grain, num_gpus.  I just got on all of our salt masters and displayed that grain on every server.  They all have either 1 or 3.  So, I just need to test that grain and see if it is greater than 1!  I'll give that a shot.

Thanks!

[brad.v...@gmail.com]

I have it working with:

{% if grains['num_gpus'] > 1 %}

{%- do salt.log.error('This server has an Nvidia GPU') -%}
{% else %}
{%- do salt.log.error('No GPU on this server') -%}
{% endif %}

Now, the reason I'm trying to determine this is for servers with the Nvidia GPUs, I want to make sure the kernel-devel versions of the installed kernel packages are also installed.  On those without an Nvidia GPU, I want to remove the kernel-devel rpms.  First, we have the number of kernels installed limited to two.  I know from a command line I can run:

# This will install the devel kernel of the running kernel:

yum install "kernel-devel-uname-r == $(uname -r)"

#

# This will install the devel kernel of the non-running kernel:

yum install "kernel-devel-uname-r == $(salt-call kernelpkg.list_installed | grep -v 'local:' | head -1 | awk '{ print $2 }').x86_64"

I just need to figure out how to translate that into salt state files.  I guess my first step is get a list of installed kernels and see if there is a corresponding installed kernel-devel rpm.

[brad.v...@gmail.com]

I have it all working with doing the kernel-devel installs/removes:

{% set KERN = salt.pkg.version('kernel') %}
{% set DEVEL = salt.pkg.version('kernel-devel') %}

{% if grains['num_gpus'] > 1 %}
{%- do salt.log.error('This server has an Nvidia GPU') -%}

{% if KERN == DEVEL %}
{%- do salt.log.error('The devel packages to match the installed kernels are already installed.') -%}

print-kernel-info:
  cmd.run:
    - name: echo "kernel = {{ KREN }} - kernel-devel = {{ DEVEL }}"
   
{% else %}
{%- do salt.log.error('The devel packages do not match the installed kernels.') -%}
{% set SPLT1,SPLT2 = KERN.split(',') %}
{% if DEVEL == SPLT1 %}
{%- do salt.log.error('Missing the SPLT2 kernel-devel') -%}
 
kernel-devel2:
  pkg.installed:
    - name: kernel-devel
    - version: {{ SPLT2 }}

{% else %}

kernel-devel1:
  pkg.installed:
    - name: kernel-devel
    - version: {{ SPLT1 }}

{% endif %}
{% endif %}

{% else %}
{%- do salt.log.error('No GPU on this server') -%}

{% if DEVEL is defined %}

remove-devel-kernel:
  pkg.removed:
    - name: kernel-devel

{% endif %}
{% endif %}

thanks to everyone for their help!

[Andreas Thienemann]

First thing you are not handling is that gpus is a list and not a dict.

You need salt['grains.get']('gpus') | first or similar to get a value you can then lookup the vendor for.

The perfect jinja filter however is map() if you want to do operate on list members.

https://jinja.palletsprojects.com/en/3.1.x/templates/#jinja-filters.map and https://docs.saltproject.io/en/latest/topics/jinja/index.html arę suggested reading material.

What also helps in your situation is to write things step by step and send them through the slsutil.renderer module call:

Some example code that would illustrate what you need to do:

---

{% set data = {

    'gpus': [

      {'model': 'MGA G200eH3', 'vendor': 'matrox'},

      {'model': 'GP102GL [Tesla P40]', 'vendor': 'nvidia'},

      {'model': 'GP102GL [Tesla P40]', 'vendor': 'nvidia'},

    ]

  }

%}

simulated_grains:

  {{ data | tojson }}

mapped_vendors: {{ data.gpus | map(attribute="vendor") | join(', ') }}

{% if "nvidia" in data.gpus | map(attribute="vendor") %}

nvidia_found: true

{% else %}

nvidia_found: false

{% endif %}

If you send that through slsutil.renderer, you get some nifty output:

# salt-call slsutil.renderer /tmp/test.sls

local:

    ----------

    simulated_grains:

        ----------

        gpus:

            |_

              ----------

              model:

                  MGA G200eH3

              vendor:

                  matrox

            |_

              ----------

              model:

                  GP102GL [Tesla P40]

              vendor:

                  nvidia

            |_

              ----------

              model:

                  GP102GL [Tesla P40]

              vendor:

                  nvidia

    mapped_vendors:

        matrox, nvidia, nvidia

    nvidia_found:

        True

You can adapt that accordingly to your requirements.

Cheers,

 Andreas

--
You received this message because you are subscribed to the Google Groups "Salt-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to salt-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/salt-users/77b0b921-4fd1-4476-8e35-196a6969fb8dn%40googlegroups.com.

[Raf Czlonka]

Hi Brad,

I hadn't noticed this issue until now. Thanks for bringing it up!

This is very interesting - the above code works on some of my machine
but not on others.

# salt-call grains.item gpus
local:

----------
gpus:
|_
----------
model:

G200eR2

vendor:
matrox
|_
----------
model:

GK110GL [Tesla K20m]
vendor:
nvidia

But:

# salt-call grains.item gpus:vendor
local:
----------
gpus:vendor:
matrox

On other machines with mix of GPUs, I get 'nvidia' only instead of
'matrox' as per above.

This seems like a bug to me.

Regards,

Raf

[Raf Czlonka]

On Thu, Nov 17, 2022 at 02:59:07PM GMT, Raf Czlonka wrote:
>
> This seems like a bug to me.

Given that I rely on this daily it, I've reported it as such[0].

Please feel free to provide additional information in the comments.

Thanks again for highlighting this issue, Brad!

[0] https://github.com/saltstack/salt/issues/63079

Regards,

Raf

[brad.v...@gmail.com]

Thanks!  I will review those!  always need to learn!  :)

[Dirk Heinrichs]

Raf Czlonka:

> {% if salt['grains.get']('gpus:vendor') == 'nvidia' %}

gpus in your example is an array. You're missing an index here, or some
kind of filter.

HTH...

    Dirk

--
Dirk Heinrichs <dirk.he...@altum.de>
Matrix-Adresse: @heini:chat.altum.de
GPG Public Key: 80F1540E03A3968F3D79C382853C32C427B48049
Privacy Handbuch: https://www.privacy-handbuch.de

[Dirk Heinrichs]

brad.v...@gmail.com:

> and I get the 'No GPU on this server' whether or not there is an
> Nvidia GPU.  So, I'm obviously doing something wrong.  Any ideas?

Yes. gpus is an array, so you'll need an index or some kind of filter,
like map: https://jinja.palletsprojects.com/en/2.11.x/templates/#map