Deadline and the render farm

TumblePipe includes a tumblehead.apps.deadline module for submitting jobs to Thinkbox Deadline, and a custom Deadline plugin (UV) for running Python tasks on workers with reproducible dependencies via Astral UV.

Farm worker prerequisites

Each worker needs the same environment as an artist workstation:

  • WSL2 / Ubuntu (on Windows workers).

  • UV, ffmpeg, openimageio-tools, opencolorio-tools — install via the commands in Installation.

  • Drive mappings — /etc/fstab mounts must match the workstations exactly, so the worker can resolve the same project paths artists use.

Without matching drive mappings, jobs that reference P:\... on the Windows side won’t resolve on the worker and the job will fail.

The Deadline UV plugin

The UV plugin is maintained as a separate repository:

  • Repo — tumblehead/deadline-uv-plugin

  • Install — copy the plugin directory into <DeadlineRepository>/custom/plugins/UV/ and restart the workers that should pick it up.

Submitting a job from TumblePipe

tumblehead.apps.deadline wraps deadlinecommand with three primitives:

  • Job — a single task (script + optional requirements.txt + args).

  • Batch — a named collection of jobs with optional dependencies.

  • Deadline — the submission client.

from pathlib import Path

from tumblehead.apps.deadline import Batch, Deadline, Job

job = Job(
    Path("/path/to/script.py"),
    Path("/path/to/requirements.txt"),  # pass None if no UV env is needed
    "arg1",
    "arg2",
)

job.name = "My Render Job"
job.pool = "general"
job.group = "karma"
job.priority = 50
job.start_frame = 1
job.end_frame = 100
job.env.update({"MY_VAR": "value"})

batch = Batch("My batch")
batch.add_job(job)

farm = Deadline()
farm.submit(batch, Path("/path/to/jobs-dir"))

The plugin is always set to UV, so the worker runs the script inside a UV-managed virtualenv built from requirements_path. Pass None for requirements_path if the script has no third-party dependencies.

Use Batch.add_dep(first, second) to mark second as a dependency of first, or Batch.add_jobs_with_deps(jobs, deps) to wire a whole graph at once. See the farm job implementations in python/tumblepipe/farm/jobs/ for complete examples.

Further reading