kdl-deploy-tools

Collection of small tools for CI/CD & sustainability

Static sites (static_site.py)

This command line python tool can copy a web site to your filesystem (i.e. make it static), report and correct some issues.

Requirements: python 3.10+ and linux-type environment equipped with wget.

How to copy a site to a html folder:

python3 static_site.py copy -u https://dral.kdl.kcl.ac.uk

python3 static_site.py -h for more info and actions.

Down notifier (uptime.py)

This script will email a list of down sites.

python3 uptime.py

It is meant to run a few times a day as a cron job.

Requirements: python 3.5+

Before running it, create a config file under env/uptime.py with the following parameters:

EMAIL_SERVER = 'YOUR_SMTP_DOMAIN'
EMAIL_TO=[
  'RECIPIENT_1',
  'RECIPIENT_2',
]
EMAIL_FROM = 'SENDER'
UPTIME_API_KEY = 'YOUR_UPTIME_ROBOT_READ_ONLY_API_KEY'

Mercurial to Git converter (hg2git.sh)

This script converts a Mercurial repository to a Git repository and preserves the commit history.

./hg2git.sh /path/to/mercurial/repository

For more information, see the script itself.

Visual Regression Toolkit (vireg)

This tool is used for visual regression testing. It compares two images and reports any differences.

More info in the README.md file.

GPUs on SLURM (hpc-gpus)

Shows number of available gpus per type on SLURM cluster at any given moment.

python hpc-gpus.py

       GPU | left (gpu) | max  (gpu) | left (int) | max  (int)
--------------------------------------------------------------
  a100_40g |          6 |         40 |          2 |         36
  a100_80g |          0 |          0 |          1 |         41
       a30 |          0 |          0 |          0 |          8
       a40 |          0 |          0 |          0 |          6
      h100 |          0 |          0 |          0 |         11
        ib |          0 |          0 |          0 |          0
      l40s |          0 |          0 |          2 |         19
   rtx2080 |          0 |          0 |          1 |          4
   rtx3070 |          0 |          0 |          9 |         10
        t4 |          0 |          0 |          0 |          6
   titan_v |          0 |          0 |          0 |          2
  titan_xp |          0 |          0 |          0 |          2
      v100 |          0 |          0 |          0 |          4

Where int stands for interruptible_gpu partition.

GPU hoarder killer (kill-gpu-hoarder.py)

Kills processes that allocate significant GPU VRAM but remain idle for too long.

python3 kill-gpu-hoarder.py

Requirements: NVIDIA GPU, nvidia-smi, python 3.6+. Must be run as root.

Edit the constants at the top of the script to configure thresholds:

• VRAM_THRESHOLD_GB (default: 2) — minimum VRAM usage to be considered
• INACTIVE_LIMIT_MINUTES (default: 10) — how long a process can stay at 0% GPU compute before being killed
• CHECK_INTERVAL_SECONDS (default: 5) — polling interval

List models behind litellm API (litellmodels.py)

Before the first run set the following variables in your env/litellmodels.py script: API_URL (API entry point), TOKEN (your litellm API token).

python3 litellmodels.py 

arc:apex_pro    perplexity/arc-apex_pro        moonshotai/Kimi-K2.6       262144 VISION
arc:apex        hosted_vllm/arc:apex           moonshotai/Kimi-K2.6       262144 VISION
arc:nexus       hosted_vllm/arc:nexus          Qwen/Qwen3.6-35B-A3B       262144 VISION
arc:lite        ollama_chat/arc:lite           Google/Gemma4:26b           32768 VISION
arc:nano        ollama_chat/arc:nano           Google/Gemma4:e4b            8192 VISION
arc:embed       ollama/qwen3-embedding:0.6b    qwen/qwen3-embedding-0.6b    None       
medgemma1.5:4b  ollama/medgemma1.5:4b          Google/medgemma1.5:4b       32768 VISION

Analog (analog.py)

Analyse web access log files. Run python3 analog.py --help to see available actions.