[Serve] Return service name and endpoint from sky.serve.up

[fozziethebeat]

Fixes #3545

This returns the generated (or specified) service name and the generated endpoint. This moves closer to supporting sky.serve.core.up as a library function.

Manually ran the following:

import sky
from sky.serve.service_spec import SkyServiceSpec 

run_str = """
docker run  \
--runtime nvidia \
--gpus all \
--shm-size 1g \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=${HUGGINGFACE_TOKEN}" \
--ipc=host \
-p 8000:8000 \
vllm/vllm-openai:latest \
--trust-remote-code \                                                                            
--tensor-parallel-size $SKYPILOT_NUM_GPUS_PER_NODE \
--max-model-len 18000 \
--chat-template '<s>[INST] You are a fair judge assistant tasked with providing clear, objective feedback based on specific criteria, ensuring each assessment reflects the absolute standards set for performance. </s>{% for message in messages %}{% if message["role"] == "user" %}{{ "[INST] " + message["content"] + " [/INST]" }}{% elif message["role"] == "assistant" %}{{ message["content"] + "</s> " }}{% endif %}{% endfor %}' \
--model prometheus-eval/prometheus-7b-v2.0
"""
task = sky.Task(
    envs={
        "HUGGINGFACE_TOKEN": "<token>",
    },
    run=run_str,
)

task.set_service(SkyServiceSpec(
    "/v1/models",
    1200,
    1,
))
resources = sky.Resources(
    cloud=sky.clouds.AWS(),
    ports="8080",
    region="us-west-2",
    accelerators="A10G:1",
    instance_type="g5.8xlarge",
)
task.set_resources(resources)
service_name, endpoint = sky.serve.core.up(task)

Tested (run the relevant ones):

• Code formatting: bash format.sh
• Any manual or new tests for this PR (please specify below)
• All smoke tests: pytest tests/test_smoke.py
• Relevant individual smoke tests: pytest tests/test_smoke.py::test_fill_in_the_name
• Backward compatibility tests: conda deactivate; bash -i tests/backward_compatibility_tests.sh

[cblmemo]

Thanks for adding this Feature! Left a comment about docstr 🫡

[cblmemo]

nit: is it possible to use a similar format with

skypilot/sky/execution.py

Lines 444 to 449 in 5a2f1b8

	Returns:
	job_id: Optional[int]; the job ID of the submitted job. None if the
	backend is not CloudVmRayBackend, or no job is submitted to
	the cluster.
	handle: Optional[backends.ResourceHandle]; the handle to the cluster. None
	if dryrun.

Also cc @Michaelvll for discussion: do you think it would be helpful to return the controller handle, like sky.launch?

[Michaelvll]

It feels fine to me to only return the service name and the endpoint for now, as the controller handle can always be retrieved with sky.status.

[fozziethebeat]

docstr change made~

[Michaelvll]

Thanks for adding the return value for up and sharing the use case @fozziethebeat! LGTM!

Just an FYI, we should be able to use sky.serve.up which is slightly simpler than sky.serve.core.up