The objective of this solution is to assign CPU and GPU minute quotas to users and accounts on Slurm clusters, and to block Slurm job submissions and modifications when these quotas are reached.
The solution takes into account the time preallocated to jobs that are not yet completed. These jobs must be accounted for to prevent users/accounts from submitting jobs in parallel that, when added together, could exceed the quota once they are accounted for upon completion. By controlling the sum of "consumed + preallocated" at both the user and account levels, we ensure that reserved but not yet used capacity is properly accounted for and that the system is not over-committed.
The solution is built around a SQLite database located at /var/lib/state/slurm-quota/slurm-quota.db.
This database is used by 2 programs:
job_submit.lua, a Lua script designed to be used as a Slurm submission plugin.slurm-quota, a Python script with several subcommands intended to be executed by Slurm, administrators, and cluster users.
When the Lua submission plugin is enabled in the Slurm configuration, the job_submit.lua script is automatically called during each job submission or modification (sbatch, srun, scontrol, etc.) to validate the request before it is accepted into the system. The script can thus apply custom rules (such as quota control for example) and reject jobs that do not comply with the defined policies.
The provided job_submit.lua script calculates the requested CPU minutes from num_tasks × time_limit. It also calculates the requested GPU minutes from the GPU resources specified in the job's TRES fields (tres_per_job, tres_per_task, tres_per_node, tres_per_socket), taking into account the load factors configured by GPU type. It then checks if a numeric quota is set in the database (quota_cpu_minutes != -1 for CPU, quota_gpu_minutes != -1 for GPU). The plugin compares the calculated CPU and GPU minute value for the job to the available share, defined as the quota minus the sum of "already consumed" and "already preallocated". This check is performed for both the user and the account, for both CPU and GPU. If the request exceeds any of the available shares (CPU or GPU), the submission or modification is refused, with an explicit message to the user.
When a job submission is accepted, the Lua script creates a corresponding preallocation in the jobs_preallocations table and associates this preallocation with a generated unique UUID identifier, the Slurm username, and the Slurm account. In case of an accepted job modification, the Lua script updates the preallocation assigned to the job in the database.
The Lua script generates a UUID at submission time to uniquely identify jobs and to be able to track the preallocated time until their completion. The job ID unfortunately cannot be used for this purpose because it is not yet available at the time of the job_submit callback (Slurm assigns a job ID later only if the job is accepted by the job_submit.lua script). The generated UUID identifier is stored in the job's admin_comment field, so it can be retrieved by the solution to reassociate the preallocation with the job during other steps.
The slurm-quota-charge-wrapper wrapper script is designed to be executed by Slurm's job completion script plugin (JobCompType=script). When this functionality is enabled, Slurm executes this script every time a job completes or is cancelled. This wrapper script actually executes the slurm-quota charge command. The wrapper is used for 2 reasons:
- Slurm does not allow directly specifying arguments to the command executed by the script completion plugin. The wrapper allows this limitation to be bypassed.
- Slurm systematically redirects JobComp script output to
/dev/null. By using the wrapper as an intermediate layer, it is possible to redirect the output of theslurm-quota chargecommand to a dedicated log file (/var/log/slurm/charge/slurm-quota-charge.log) to ensure that all processing information and any errors are preserved to trace operations.
Upon job completion, the slurm-quota charge command retrieves the UUID from admin_comment and the allocated GPU resources from AllocTRES (via sacct). It calculates the effective consumption in CPU minutes according to PROCS × (END − START) / 60 and in GPU minutes according to the allocated GPUs, their type, and the configured load factors. It credits these consumptions to the user and to the account, and deletes the corresponding preallocation in the database. This step ensures that the difference between "reserved" and "actually used" is correctly reconciled for both dimensions (CPU and GPU).
In the SQLite database, there are 4 tables:
users: It contains user names, consumed CPU and GPU minutes, and assigned quotas.accounts: It contains Slurm account names, consumed CPU and GPU minutes, and assigned quotas.jobs_preallocations: It contains CPU and GPU minutes preallocated to non-completed jobs, withjob_uuid,username, andaccount.gpu_factors: It contains load factors by GPU type, allowing calculation of billed GPU minutes based on the GPU type used.
We record the amount of preallocated time per job rather than a global value per user to allow fine-grained updates during modifications (increase or decrease of time_limit/num_tasks), targeted deletion of the preallocation upon completion, and robust cleanup of orphans. A global value would hide the detail per job and significantly complicate adjustments and cancellations, with an increased risk of inconsistencies.
The SQLite database file must have the system user slurm as owner, with mode 0644 to restrict modification permission to the slurm user (used by slurmctld for the Lua script and the jobcomp script) and to administrators with the root account. Other users only have read-only access to the database. The slurm-quota script automatically creates the database with the charge, user-quota, and account-quota commands (intended for Slurm and administrators), setting the correct permissions on the file.
The commands slurm-quota user-quota, slurm-quota account-quota, slurm-quota user-gpu-quota, and slurm-quota account-gpu-quota respectively allow assigning CPU and GPU quotas to users and accounts.
The slurm-quota adjust command (restricted to root) allows manually adjusting consumed CPU/GPU time for one user or account with an explicitly signed delta.
Default quotas used when the solution auto-creates a user or account can be displayed with slurm-quota default-quotas and updated with slurm-quota set-default-quotas. These defaults are applied only to newly auto-created entries and do not modify existing users/accounts.
The solution allows setting GPU load factors. This is a multiplicative coefficient applied to the calculation of consumed GPU minutes based on the GPU type used. This factor allows adjusting, for each GPU type, the actual consumption weighting, taking into account the different value or computing power of the models (for example, assigning a factor of 0.5 to an h100 GPU amounts to counting 10 minutes of usage as only 5 minutes consumed). The default factor is 1.0 if no specific factor is configured for a given type. Thus, administrators can finely adapt GPU billing based on GPU models.
The slurm-quota set-gpu-factor command allows configuring load factors by GPU type (restricted to root). The slurm-quota gpu-factors command displays the currently configured GPU load factors.
The slurm-quota serve command starts a small HTTP/JSON server designed to work with systemd "socket activation". A slurm-quota.socket socket unit listens on TCP port 9911 and launches the slurm-quota.service service on demand upon the first connection. The server can automatically stops after a configurable period of inactivity (10 minutes by default). The API exposes a single /stats route that returns a JSON object of the form { users: [...], accounts: [...] }. Optional query parameters can be used to filter responses: username filters users and limits accounts to this user's Slurm associations (e.g. /stats?username=alice), while account returns only the requested account stats in the accounts array (e.g. /stats?account=hpc).
The slurm-quota stats command consumes this HTTP/JSON API by default (URL configurable via the SLURM_QUOTA_URL environment variable, default http://127.0.0.1:9911/). It queries /stats and displays a readable table in the terminal. By specifying a user (--user or positional username), an account (--account), or the --all option, the command transmits the appropriate filters to the service. User and account selectors are mutually exclusive. If the service is not available or in case of connection failure to the server, execution fails with an error message.
Additionally, a logrotate configuration file is provided (slurm-quota-charge.logrotate) to prevent the log file fed by the slurm-quota-charge-wrapper wrapper from growing too large over time.
The slurm-quota-web application is a web dashboard that retrieves the same statistics from the HTTP API (GET /stats) and renders them as HTML tables and quota usage bars. It can run standalone with Flask built-in HTTP server for local testing, or be launched by a production-ready HTTP server (for example Apache with mod_wsgi) as a WSGI application.
RPM packages are published for Enterprise Linux 9 (RHEL 9, Rocky Linux 9, AlmaLinux 9, CentOS Stream 9, and similar) in the Rackslab packages repository.
- Install the Rackslab repository keyring:
sudo curl https://pkgs.rackslab.io/keyring.asc --output /etc/pki/rpm-gpg/RPM-GPG-KEY-Rackslab- Create
/etc/yum.repos.d/rackslab.repowith this content:
[rackslab]
name=Rackslab
baseurl=https://pkgs.rackslab.io/rpm/el9/main/$basearch/
gpgcheck=1
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-RackslabThe following packages are available:
slurm-quota: common files for all nodes (CLI, manpage, bash completion)slurm-quota-controller: controller-only files (job_submit.lua, wrapper, systemd units, logrotate, migration script)slurm-quota-web: optional web application with HTML dashboard
On the controller node:
- Install controller and common packages:
sudo dnf install slurm-quota slurm-quota-controller- Start and enable the socket-activated API service:
sudo systemctl enable --now slurm-quota.socket- Configure Slurm plugins in
slurm.conf:
Edit the Slurm configuration to set up these parameters:
JobCompType=jobcomp/script
JobCompLoc=/etc/slurm/slurm-quota-charge-wrapper
JobSubmitPlugins=lua
AccountingStorageTRES=gres/gpu:<type1>,gres/gpu:<type2>The AccountingStorageTRES parameter enables recording of complementary resource allocations (e.g., GPU, licenses) in addition to generic resources (e.g., nodes, cores, memory) in the Slurm accounting database. It is necessary to enable tracking of all GPU types in the cluster so that the slurm-quota charge command can determine the GPUs allocated to completed jobs and account for the time consumed on these GPUs.
On compute/login nodes:
- Install the common package:
sudo dnf install slurm-quota- Configure the controller API endpoint for all users in
/etc/profile.d/slurm-quota.sh:
export SLURM_QUOTA_URL=http://controller:9911/For the optional web dashboard:
- Install the web dashboard package on the node running Apache:
sudo dnf install slurm-quota-web- Install Apache/mod_wsgi packages:
sudo dnf install httpd mod_wsgi httpd-tools- Configure Apache virtual host with mod_wsgi:
<VirtualHost *:80>
ServerName quota.example.org
# Optional: point web app to remote API endpoint
SetEnv SLURM_QUOTA_URL http://127.0.0.1:9911/
WSGIDaemonProcess slurm-quota-web processes=2 threads=5 display-name=%{GROUP}
WSGIProcessGroup slurm-quota-web
WSGIScriptAlias / /usr/libexec/slurm-quota/slurm-quota-web
# If you mount in a subdir (example: /quota), use:
# WSGIScriptAlias /quota /usr/libexec/slurm-quota/slurm-quota-web
Alias /static/ /usr/share/slurm-quota-web/static/
# Keep the static prefix aligned with WSGIScriptAlias:
# - WSGIScriptAlias / -> Alias /static/
# - WSGIScriptAlias /quota -> Alias /quota/static/
<Directory /usr/share/slurm-quota-web/static>
Require all granted
</Directory>
<Directory /usr/libexec/slurm-quota>
Require all granted
</Directory>
ErrorLog /var/log/httpd/slurm-quota-web-error.log
CustomLog /var/log/httpd/slurm-quota-web-access.log combined
</VirtualHost>- Enable and reload Apache:
sudo systemctl enable --now httpd
sudo apachectl configtest
sudo systemctl reload httpd- Optional: enable HTTP Basic authentication with
htpasswd:
sudo htpasswd -c /etc/httpd/conf.d/slurm-quota-web.htpasswd adminThen add inside the same <VirtualHost>:
<Location />
AuthType Basic
AuthName "slurm-quota dashboard"
AuthUserFile /etc/httpd/conf.d/slurm-quota-web.htpasswd
Require valid-user
</Location>Security recommendations:
- Keep the backend API bound to cluster local networks when possible.
- Restrict dashboard access with auth and/or trusted networks.
- Prefer HTTPS/TLS at Apache level.
Here is the procedure to follow to install the solution on the batch controller server:
- Installation of Lua dependencies:
sudo dnf install lua-dbi lua-posix sqlite- Directories and permissions
# log directory for the wrapper
sudo mkdir -p /var/log/slurm/charge
sudo chown slurm: /var/log/slurm/charge
# data directory for the database
sudo mkdir -p /var/lib/state/slurm-quota
sudo chown slurm: /var/lib/state/slurm-quota
sudo chmod 0755 /var/lib/state/slurm-quota- Installation of the
slurm-quotacommand
sudo cp slurm-quota /usr/local/bin/slurm-quota
sudo chmod 0755 /usr/local/bin/slurm-quotaOptional: install Bash completion for slurm-quota:
sudo cp slurm-quota.bash-completion /etc/bash_completion.d/slurm-quota
sudo chmod 0644 /etc/bash_completion.d/slurm-quota- Installation of the HTTP JSON service (optional)
The HTTP JSON service allows exposing statistics via a REST API to facilitate integration with other tools. It is designed to work with systemd socket activation.
# Installation of systemd files
sudo cp slurm-quota.socket /etc/systemd/system/
sudo cp slurm-quota.service /etc/systemd/system/
sudo chmod 0644 /etc/systemd/system/slurm-quota.socket
sudo chmod 0644 /etc/systemd/system/slurm-quota.service
# Reload systemd and enable the service
sudo systemctl daemon-reload
sudo systemctl enable --now slurm-quota.socket
# Service verification
sudo systemctl status slurm-quota.socket
curl http://127.0.0.1:9911/healthThe service automatically stops after 10 minutes of inactivity (configurable via --idle-timeout in slurm-quota.service, with 0 meaning no idle timeout).
- Installation of the wrapper script
sudo cp slurm-quota-charge-wrapper /etc/slurm/slurm-quota-charge-wrapper
sudo chmod 0755 /etc/slurm/slurm-quota-charge-wrapper- Slurm submission plugin (
job_submit.lua)
sudo cp job_submit.lua /etc/slurm/job_submit.lua
sudo chmod 0644 /etc/slurm/job_submit.lua- Activation of Slurm plugins
Edit the Slurm configuration to set up these parameters:
JobCompType=jobcomp/script
JobCompLoc=/etc/slurm/slurm-quota-charge-wrapper
JobSubmitPlugins=lua
AccountingStorageTRES=gres/gpu:<type1>,gres/gpu:<type2>The AccountingStorageTRES parameter enables recording of complementary resource allocations (e.g., GPU, licenses) in addition to generic resources (e.g., nodes, cores, memory) in the Slurm accounting database. It is necessary to enable tracking of all GPU types in the cluster so that the slurm-quota charge command can determine the GPUs allocated to completed jobs and account for the time consumed on these GPUs.
- Logrotate configuration (recommended)
sudo cp slurm-quota-charge.logrotate /etc/logrotate.d/slurm-quota-charge
sudo chmod 0644 /etc/logrotate.d/slurm-quota-chargeIt is recommended to back up the SQLite database file /var/lib/state/slurm-quota/slurm-quota.db. To do this, simply run this command regularly:
sudo sqlite3 /var/lib/state/slurm-quota/slurm-quota.db ".backup /var/lib/state/slurm-quota/slurm-quota-$(date +%Y-%m-%d).db"On the other nodes of the cluster, here are the steps to follow:
- Installation of the
slurm-quotacommand
sudo cp slurm-quota /usr/local/bin/slurm-quota
sudo chmod 0755 /usr/local/bin/slurm-quotaOptional: install Bash completion for slurm-quota:
sudo cp slurm-quota.bash-completion /etc/bash_completion.d/slurm-quota
sudo chmod 0644 /etc/bash_completion.d/slurm-quota- Set the
SLURM_QUOTA_URLvariable in the user environment
The SLURM_QUOTA_URL environment variable must point to the controller node to indicate the server to query to obtain quotas. To facilitate the use of the slurm-quota stats command, this variable must be automatically set in the user environment on all nodes. For example, it is possible to add the following line in the /etc/profile.d/sh.local file:
export SLURM_QUOTA_URL=http://controller:9911/- Install dependencies:
sudo dnf install python3-flask python3-jinja2 httpd mod_wsgi httpd-tools- Install script and assets:
sudo install -Dm0755 slurm-quota-web /usr/local/bin/slurm-quota-web
sudo mkdir -p /usr/local/share/slurm-quota-web
sudo cp -r webapp/templates /usr/local/share/slurm-quota-web/
sudo cp -r webapp/static /usr/local/share/slurm-quota-web/- Configure the assets path for the app:
export SLURM_QUOTA_WEB_ASSETS_DIR=/usr/local/share/slurm-quota-web- Run standalone (HTTP built-in server, for testing only):
SLURM_QUOTA_URL=http://127.0.0.1:9911/ \
SLURM_QUOTA_WEB_ASSETS_DIR=/usr/local/share/slurm-quota-web \
/usr/local/bin/slurm-quota-web- Configure Apache with this manual installation:
<VirtualHost *:80>
ServerName quota.example.org
SetEnv SLURM_QUOTA_URL http://127.0.0.1:9911/
SetEnv SLURM_QUOTA_WEB_ASSETS_DIR /usr/local/share/slurm-quota-web
WSGIDaemonProcess slurm-quota-web processes=2 threads=5 display-name=%{GROUP}
WSGIProcessGroup slurm-quota-web
WSGIScriptAlias / /usr/local/bin/slurm-quota-web
# If you mount in a subdir (example: /quota), use:
# WSGIScriptAlias /quota /usr/local/bin/slurm-quota-web
Alias /static/ /usr/local/share/slurm-quota-web/static/
# Keep the static prefix aligned with WSGIScriptAlias:
# - WSGIScriptAlias / -> Alias /static/
# - WSGIScriptAlias /quota -> Alias /quota/static/
<Directory /usr/local/share/slurm-quota-web/static>
Require all granted
</Directory>
<Directory /usr/local/bin>
<Files slurm-quota-web>
Require all granted
</Files>
</Directory>
# Optional HTTP Basic authentication
# AuthType Basic
# AuthName "slurm-quota dashboard"
# AuthUserFile /etc/httpd/conf.d/slurm-quota-web.htpasswd
# Require valid-user
ErrorLog /var/log/httpd/slurm-quota-web-error.log
CustomLog /var/log/httpd/slurm-quota-web-access.log combined
</VirtualHost>- Optional: create
htpasswdfile:
sudo htpasswd -c /etc/httpd/conf.d/slurm-quota-web.htpasswd adminstats: Displays consumed CPU times, preallocated CPU times (with the number of jobs considered), and quotas for users and accounts.
Examples:
slurm-quota stats # displays the current user and their accounts
slurm-quota stats alice # details for user alice and their accounts
slurm-quota stats --user alice # same as positional username
slurm-quota stats --account hpc # only stats for account hpc
slurm-quota stats --all # lists all users and all accounts
slurm-quota stats --hours # same stats displayed in hoursNote: --account is mutually exclusive with user selection (--user or positional username).
Color display of the status bar can be disabled by setting the NO_COLOR environment variable.
The --hours option changes only the displayed unit in the stats output; stored values and API values remain in minutes.
serve: Launches an HTTP JSON server to expose statistics via a REST API. Designed to work with systemd socket activation.
Examples:
# Manual launch (testing)
slurm-quota serve --host 127.0.0.1 --port 9911 --idle-timeout 600
slurm-quota serve --host 127.0.0.1 --port 9911 --idle-timeout 0 # no idle timeout
# Via systemd (recommended)
sudo systemctl start slurm-quota.socket
curl http://127.0.0.1:9911/health
curl http://127.0.0.1:9911/stats
curl http://127.0.0.1:9911/stats?username=alice
curl http://127.0.0.1:9911/stats?account=hpcThe service automatically stops after a period of inactivity (600 seconds, ie. 10 minutes by default). This can be disabled with --idle-timeout 0 argument. The stats command queries this HTTP service (URL configurable via the SLURM_QUOTA_URL environment variable).
slurm-quota-web: Starts the web dashboard with the built-in HTTP server (intended for local testing; production should use Apache/mod_wsgi or another WSGI server).
Examples:
slurm-quota-web
SLURM_QUOTA_WEB_HOST=0.0.0.0 SLURM_QUOTA_WEB_PORT=8080 slurm-quota-web
SLURM_QUOTA_URL=http://controller:9911/ slurm-quota-webuser-quota(restricted to root): Sets a CPU quota for a user.
Examples:
sudo slurm-quota user-quota alice 50000 # 50k CPU minutes
sudo slurm-quota user-quota bob -1 # unlimiteduser-gpu-quota(restricted to root): Sets a GPU quota for a user.
Examples:
sudo slurm-quota user-gpu-quota alice 10000 # 10k GPU minutes
sudo slurm-quota user-gpu-quota bob -1 # unlimited GPUaccount-quota(restricted to root): Sets a CPU quota for a Slurm account.
Examples:
sudo slurm-quota account-quota projX 200000 # 200k CPU minutes for account projX
sudo slurm-quota account-quota projY -1 # unlimitedaccount-gpu-quota(restricted to root): Sets a GPU quota for a Slurm account.
Examples:
sudo slurm-quota account-gpu-quota projX 50000 # 50k GPU minutes
sudo slurm-quota account-gpu-quota projY -1 # unlimited GPUadjust(restricted to root): Adjusts consumed CPU/GPU time for one user or one account.
Examples:
sudo slurm-quota adjust --user alice --cpu --minutes=+30 # add 30 consumed CPU minutes
sudo slurm-quota adjust --user alice --gpu --minutes=-120 # subtract 120 consumed GPU minutes
sudo slurm-quota adjust --account projX --cpu --hours=+2 # add 2 consumed CPU hours (120 minutes)
sudo slurm-quota adjust --account projX --gpu --hours=-1 # subtract 1 consumed GPU hour (60 minutes)Notes:
-
The delta must be explicitly signed (
+or-), for example+30or-30. -
Subtractions are clamped to zero: consumed time never becomes negative.
-
default-quotas: Displays the default CPU/GPU quotas applied to newly auto-created users/accounts.
Example:
slurm-quota default-quotasset-default-quotas(restricted to root): Sets one or more default quotas applied when a user/account is auto-created by the submission plugin. Existing users/accounts are not modified.
Examples:
sudo slurm-quota set-default-quotas --user-cpu 50000 --account-cpu 200000
sudo slurm-quota set-default-quotas --user-gpu 10000 --account-gpu 50000
sudo slurm-quota set-default-quotas --user-cpu -1 --user-gpu -1 --account-cpu -1 --account-gpu -1gpu-factors: Displays the currently configured GPU load factors.
Example:
slurm-quota gpu-factorsset-gpu-factor(restricted to root): Configures the load factor for a GPU type. Billed GPU minutes are calculated asnumber_GPU × time_minutes × factor. The default factor is 1.0 if no factor is configured for a GPU type.
Examples:
sudo slurm-quota set-gpu-factor h100 0.5 # Factor 0.5 for h100 GPUs
sudo slurm-quota set-gpu-factor h200 0.8 # Factor 0.8 for h200 GPUs
sudo slurm-quota set-gpu-factor default 1.0 # Default factor (used if type is not specified)prune(restricted to root): Cleans data with dedicated selectors:--preallocs: remove orphaned preallocations (jobs not present in Slurm queue)--users: remove users with both consumed CPU and consumed GPU at 0--accounts: remove accounts with both consumed CPU and consumed GPU at 0--all: prune all categories above (default behavior when no selector is provided)--user <username>: limit user pruning candidates to one username--account <account>: limit account pruning candidates to one account--dry-run: report how many preallocations/users/accounts would be removed, without deleting rows
Example:
sudo slurm-quota prune # default: same as --all
sudo slurm-quota prune --preallocs # prune only orphaned preallocations
sudo slurm-quota prune --users # prune only users with 0 consumed CPU/GPU
sudo slurm-quota prune --users --user alice # prune only this eligible user
sudo slurm-quota prune --accounts # prune only accounts with 0 consumed CPU/GPU
sudo slurm-quota prune --accounts --account hpc # prune only this eligible account
sudo slurm-quota prune --dry-run # preview removals without applying themIt is normally not necessary to execute this prune command under normal conditions. It may be useful in case of malfunction of the call to the slurm-quota charge command by Slurm. Its execution is nevertheless safe, it can be executed if in doubt about the preallocated durations assigned to users.
Use this procedure to switch an existing manual deployment to RPM-managed files.
- Back up the database on the controller:
sudo sqlite3 /var/lib/state/slurm-quota/slurm-quota.db ".backup /var/lib/state/slurm-quota/slurm-quota-pre-rpm-$(date +%Y-%m-%d).db"- Remove legacy manually installed files that conflict with RPM-managed paths:
On the controller node:
# Legacy systemd unit locations used by manual installation
sudo rm -f /etc/systemd/system/slurm-quota.service
sudo rm -f /etc/systemd/system/slurm-quota.socketOn all nodes:
# Legacy manual binary/completion/manpage copies (RPM will reinstall managed files)
sudo rm -f /usr/local/bin/slurm-quota
sudo rm -f /usr/local/bin/slurm-quota-web
sudo rm -f /etc/bash_completion.d/slurm-quota
sudo rm -f /usr/local/share/man/man1/slurm-quota.1
sudo rm -f /usr/share/man/man1/slurm-quota.1
sudo rm -rf /usr/local/share/slurm-quota-web- Apply the RPM packages (recommended) procedure above (controller + compute/login nodes).
When using RPM packages, migration is automatically run during slurm-quota-controller installation/upgrade (only when the existing database file is present).
To force migration manually with RPM packages, run:
sudo /usr/libexec/slurm-quota/migrate-slurm-quotaFor manual/source-based deployments, the database migration script must be executed before updating other components:
# python3 migrate-slurm-quota
2025-12-04 10:11:42,926 - INFO - Adding array_size column to jobs_preallocations table
2025-12-04 10:11:42,938 - INFO - Migration completed: array_size column added
2025-12-04 10:11:42,939 - INFO - Database migration completed successfullyThen, the other components (job_submit.lua, slurm-quota, etc.) must be updated.
Manpages are maintained in AsciiDoc format:
man/slurm-quota.1.adocforslurm-quotaman/slurm-quota-web.1.adocforslurm-quota-web
To generate roff manpages from these files, use:
asciidoctor -b manpage -o slurm-quota.1 man/slurm-quota.1.adoc
asciidoctor -b manpage -o slurm-quota-web.1 man/slurm-quota-web.1.adocTo preview generated files locally:
man -l ./slurm-quota.1
man -l ./slurm-quota-web.1Optional user-local installation:
install -Dm644 slurm-quota.1 ~/.local/share/man/man1/slurm-quota.1
install -Dm644 slurm-quota-web.1 ~/.local/share/man/man1/slurm-quota-web.1The repository includes unit tests under tests/unit/ (one module per function under test) and functional CLI tests under tests/functional/ (one module per slurm-quota subcommand). They are standard unittest.TestCase classes; the recommended runner is pytest (as in CI), with optional coverage reports configured in pyproject.toml.
From the repository root, use a virtual environment (recommended on distributions that restrict system-wide pip, e.g. PEP 668):
python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
python -m pip install -U pip
python -m pip install ".[dev]"Run the full suite (quiet mode, coverage for the loaded slurm-quota module, terminal + coverage.xml):
python -m pytestThe development of this project was funded by ISDM-Meso, part of the University of Montpellier.
ISDM stands for Institut des Sciences des Données de Montpellier. ISDM-Meso is the ISDM mesocentre (mesocenter), i.e. a shared mid-scale research computing facility providing HPC and data services to research teams, bridging local institutional resources and national/international supercomputing centers. This tool was developed in that operational context to support the administration of Slurm-based clusters.
This project is licensed under the MIT License.