relay-jobs 0.1.1

Python SDK for relay job submission

relay-jobs

Python SDK for relay job submission. See full quickstart below.

Install

pip install relay-jobs

Quickstart

The SDK reads its configuration from (in order) explicit kwargs, the environment variables RELAY_URL / RELAY_TOKEN, and the CLI config file at ~/.relay/config.json (or %APPDATA%\relay\config.json on Windows). The simplest path is to run relay login once, then import and go:

import relay

job = relay.Job(name="nightly-cook", priority=relay.Priority.HIGH)
cook = job.add_task("cook", commands=[["ue4-cook", "--map", "Main"]], retries=2)
job.add_task("test", commands=[["pytest"]], depends_on=[cook])

with relay.Client() as client:
    submitted = client.submit(job)
    print(f"submitted {submitted.id}")
    final = client.wait(submitted.id, timeout=600)
    if final.status != relay.JobStatus.DONE:
        raise RuntimeError(f"job ended {final.status}")

Authoring

relay.Job and relay.Task are Pydantic models that mirror the server's JobSpec exactly. You can use the add_task builder:

job = relay.Job(name="example")
job.add_task(
    "build",
    commands=[["make", "build"]],
    env={"GOOS": "linux"},
    requires={"gpu": "true"},
    timeout_seconds=3600,
    retries=1,
    source=relay.Source(
        type="perforce",
        stream="//depot/main",
        sync=[relay.Sync(path="//depot/main/...", rev="#head")],
        unshelves=[12345],
    ),
)

Or construct from a dict if you already have a JSON spec:

job = relay.Job.model_validate(spec_dict)

depends_on accepts Task instances or names:

a = job.add_task("a", commands=[["echo", "1"]])
b = job.add_task("b", commands=[["echo", "2"]], depends_on=[a])  # or ["a"]

Client API

Method	Description
submit(job)	POST /v1/jobs. Validates locally, returns the populated Job.
get_job(id)	GET /v1/jobs/{id}.
list_jobs(status=, scheduled_job_id=)	GET /v1/jobs with optional filters.
cancel_job(id, force=False)	DELETE /v1/jobs/{id} — graceful by default; force=True requests an immediate kill on the agent.
get_tasks(job_id)	GET /v1/jobs/{id}/tasks.
get_task(id)	GET /v1/tasks/{id}.
task_logs(id)	GET /v1/tasks/{id}/logs.
follow_job(id)	Iterator over SSE Event objects until the job is terminal.
wait(id, timeout=None, poll_interval=1.0)	Block (polling) until the job is terminal.
create_schedule(...)	POST /v1/scheduled-jobs.
list_schedules() / get_schedule(id) / update_schedule(id, ...) / delete_schedule(id)	Standard CRUD.
run_schedule_now(id)	POST /v1/scheduled-jobs/{id}/run-now. Admin-only on the server.

Errors

All exceptions descend from relay.RelayError:

Class	When
ValidationError	Local Pydantic failure or server 400
AuthError	Missing token, 401, or 403
NotFound	404
Conflict	409 (e.g. cancelling a terminal job)
ServerError	5xx
HTTPError	Any other unexpected status
TimeoutError	wait() exceeded its wall-clock limit

The original httpx.Response is attached as .response on each instance for debugging.

Compatibility

• Python: 3.9, 3.10, 3.11, 3.12, 3.13.
• Server: tested against relay-server master. The SDK only consumes the existing v1 REST + SSE surface — no server-side changes needed.

Development

From the python/ directory:

python -m venv .venv && .venv/Scripts/python -m pip install -e ".[dev]"
pytest tests/unit
RELAY_INTEGRATION=1 pytest tests/integration   # requires a running relay-server

Or run from the repo root via make python-test and make python-test-integration.