[P/D][Misc] Enable profiling in disagg setup

[NickLucche]

Quick add to enable profiling on PD disaggregated setup for development.

Test with

# Setting this allows creating a symlink to Justfile from another dir
set working-directory := "/home/nicolo/perf-test/"
# Needed for the proxy server
vllm-directory := "/home/nicolo/vllm/" 

MODEL := "meta-llama/Llama-3.1-8B-Instruct"
PREFILL_TP_SIZE := "1"
DECODE_TP_SIZE := "1"
PREFILL_GPUS := "0"
DECODE_GPUS := "1"
LOGGING_LEVEL := "INFO"

port PORT: 
  @python port_allocator.py {{PORT}}
prefill:
    VLLM_TORCH_PROFILER_DIR=/tmp/prefiller_profiler \
    VLLM_LOGGING_LEVEL={{LOGGING_LEVEL}} \
    VLLM_NIXL_SIDE_CHANNEL_PORT=$(just port 5557) \
    CUDA_VISIBLE_DEVICES={{PREFILL_GPUS}} \
    vllm serve {{MODEL}} \
      --port $(just port 8100) \
      --enforce-eager \
      --disable-log-requests \
      --tensor-parallel-size {{PREFILL_TP_SIZE}} \
      --kv-transfer-config '{"kv_connector":"NixlConnector","kv_role":"kv_both"}'
decode:
    VLLM_TORCH_PROFILER_DIR=/tmp/decoder_profiler \
    VLLM_LOGGING_LEVEL={{LOGGING_LEVEL}} \
    VLLM_NIXL_SIDE_CHANNEL_PORT=$(just port 5558) \
    CUDA_VISIBLE_DEVICES={{DECODE_GPUS}} \
    vllm serve {{MODEL}} \
      --port $(just port 8200) \
      --enforce-eager \
      --disable-log-requests \
      --tensor-parallel-size {{DECODE_TP_SIZE}} \
      --kv-transfer-config '{"kv_connector":"NixlConnector","kv_role":"kv_both"}'
proxy:
    python "{{vllm-directory}}tests/v1/kv_connector/nixl_integration/toy_proxy_server.py" \
      --port $(just port 8192) \
      --prefiller-port $(just port 8100) \
      --decoder-port $(just port 8200)

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀