orcd
/
flame-roles
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

use qos partitions

This commit is contained in:
dvosler 2025-02-25 06:06:42 -05:00
parent c35dae0327
commit 925aec8a17
1 changed files with 48 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
slurm_qos/tasks/main.yml
View File

@ -62,32 +62,72 @@
{% if item.item.organization is defined %}Organization={{ item.item.organization }}{% endif %}
{% if item.item.description is defined %}Description="{{ item.item.description }}"{% endif %}
{% if item.item.parent is defined %}Parent={{ item.item.parent }}{% endif %}
{% if item.item.default_qos is defined %}DefaultQOS={{ item.item.default_qos }}{% endif %}
{% if item.item.qos is defined %}QOS={{ item.item.qos | join(',') }}{% endif %}
loop: "{{ account_check.results }}"
loop_control:
label: "{{ item.item.name }}"
when: item.stdout is not search(item.item.name)
changed_when: true
- name: Ensure Slurm users exist
- name: Ensure Slurm users exist with partition limits
when: slurm_qos_users | default([]) | length > 0
block:
- name: Check Slurm users
- name: Check existing Slurm users
ansible.builtin.command: "sacctmgr -Pn list users format=user"
register: user_list
changed_when: false
failed_when: false
- name: Add Slurm user
- name: Add Slurm user if not already present
ansible.builtin.command: >
sacctmgr -i add user name={{ item.username }}
{% if item.sponsor is defined %}Account={{ item.sponsor }}{% endif %}
cluster={{ slurm_cluster_name }}
account={{ item.sponsor | default('orcd') }}
{% if item.partitions is defined and item.partitions | length > 0 %}
partition={{ item.partitions | map(attribute='name') | join(',') }}
{% else %}
partition=debug,preempt
{% endif %}
{% if item.qos_list is defined and item.qos_list | length > 0 %}
QOS={{ item.qos_list | join(',') }}
{% else %}
QOS=debug_qos,preempt_qos
{% endif %}
{% if item.qos_default is defined %}
DefaultQOS={{ item.qos_default }}
{% else %}
DefaultQOS=preempt_qos
{% endif %}
{% if item.comment is defined %}Comment="{{ item.comment }}"{% endif %}
{% if item.qos_list is defined and item.qos_list | length > 0 %}QOS={{ item.qos_list | join(',') }}{% endif %}
{% if item.qos_default is defined %}DefaultQOS={{ item.qos_default }}{% endif %}
loop: "{{ slurm_qos_users }}"
loop_control:
label: "{{ item.username }}"
when: user_list.stdout is not search(item.username)
changed_when: true
register: user_creation_result
failed_when: user_creation_result.rc != 0
- name: Set partition-specific node limits
ansible.builtin.command: >
sacctmgr -i modify user where
name={{ item.0.username }}
cluster={{ slurm_cluster_name }}
account={{ item.0.sponsor | default('orcd') }}
partition={{ item.1.name }}
set
GrpTRES={{ item.1.grptres | default(
slurm_qos_partition_defaults[item.1.name].grptres if item.1.name in slurm_qos_partition_defaults else
'node=' + (
item.1.max_nodes | default(
slurm_qos_partition_defaults[item.1.name].max_nodes if item.1.name in slurm_qos_partition_defaults else
'4' if item.1.name == 'preempt' else
'2' if item.1.name == 'debug' else
''
)
) | string
) }}
loop: "{{ slurm_qos_users | subelements('partitions') }}"
loop_control:
label: "{{ item.0.username }} - {{ item.1.name }}"
changed_when: true
#GrpTRES={{ item.1.grptres | default(slurm_qos_partition_defaults[item.1.name].grptres if item.1.name in slurm_qos_partition_defaults else 'node=' + (item.1.max_nodes | default(slurm_qos_partition_defaults[item.1.name].max_nodes if item.1.name in slurm_qos_partition_defaults else '4' if item.1.name == 'preempt' else '2' if item.1.name == 'debug' else '')) | string) }}