fix(fireworks): retain root client refs to prevent aiohttp session leak

[Mason Daugherty (mdrxy)]

Summary

• ChatFireworks.validate_environment creates Fireworks / AsyncFireworks instances, extracts .chat.completions, and immediately discards the parents
• Starting with fireworks-ai>=0.19, FireworksClient.__init__ eagerly creates an aiohttp.ClientSession when an async event loop is running
• Orphaning the parents means these sessions are never closed, producing Unclosed client session warnings on stderr
• Stores parent instances as root_client / root_async_client (matching langchain-openai pattern) so transports stay reachable and are cleaned up by GC

Reproduction

In an async context (e.g., LangGraph agent, Harbor eval runner):

import asyncio, gc, warnings
warnings.simplefilter("always")

async def main():
    with warnings.catch_warnings(record=True) as caught:
        warnings.simplefilter("always")
        from langchain_fireworks import ChatFireworks
        model = ChatFireworks(model="accounts/fireworks/models/llama-v3p1-8b-instruct", api_key="test-key")
        del model
        gc.collect()
        await asyncio.sleep(0)
        gc.collect()

    unclosed = [w for w in caught if issubclass(w.category, ResourceWarning) and "unclosed" in str(w.message).lower()]
    print(f"{len(unclosed)} leaked sessions")  # 4 before fix, 0 after

asyncio.run(main())

Additional context

• The fireworks-ai sync close() also doesn't clean up the aiohttp session created in async contexts — separate upstream issue
• The regression test will only catch the leak once the fireworks-ai pin is bumped to >=0.19; on 0.15.x the aiohttp session isn't created so the test passes vacuously

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 1 untouched benchmark
⏩ 39 skipped benchmarks¹

---

_{Comparing mason/fix-fireworks-client-session-leak (d50251e) with master (d22df94)}

Footnotes

1. 39 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩