Introduce irisrun for automatic port management in distributed examples

[Copilot]

All examples hardcode port 29500 for torch.distributed initialization. Crashed processes leave ports bound, blocking development iteration.

Changes

New irisrun CLI tool (irisrun_cli/)

• Finds free TCP port via localhost socket binding
• Spawns processes with RANK, WORLD_SIZE, MASTER_ADDR, MASTER_PORT environment variables
• Similar interface to torchrun: irisrun --nproc_per_node=N script.py [args]

Updated examples/00_load/load_bench.py

• Checks for environment variables; falls back to hardcoded port if absent
• Maintains backward compatibility with direct execution

Documentation

• docs/irisrun.md: Usage reference and migration guide
• examples/00_load/README.md: Added irisrun examples

Usage

# Recommended: auto port allocation
irisrun --nproc_per_node=2 examples/00_load/load_bench.py --verbose

# Legacy: hardcoded port 29500
python examples/00_load/load_bench.py --num_ranks 2 --verbose

Example script adaptation pattern:

def _worker(local_rank, world_size, init_url, args):
    if "RANK" in os.environ and "WORLD_SIZE" in os.environ:
        # irisrun mode: use environment variables
        rank = int(os.environ["RANK"])
        init_method = f"tcp://{os.environ['MASTER_ADDR']}:{os.environ['MASTER_PORT']}"
    else:
        # standalone mode: use function parameters
        rank = local_rank
        init_method = init_url

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Introduce irisrun</issue_title>
<issue_description>All examples initial torch dist with hardcoded ports, when something crashes, we don't cleanly close the ports which makes development difficult. Introduce irisrun similar to torchrun and port just one example.</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Introduce irisrun #308

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.