PyTorch profiling when serving with two nodes

[jokerwyt]

Hi.

I'm going to follow https://docs.sglang.ai/references/benchmark_and_profiling.html. When I use DP and serve with two nodes, what's the proper way to do pytorch profiling?
should I set env variable export SGLANG_TORCH_PROFILER_DIR=/root/sglang/profile_log when launching both instances? Should they share a profiler director via an NFS or something else?

related issue: #2110

Thank you!