test(fireworks): stabilize integration tests with rate limiting and retries (#37590)

Fireworks integration tests have been flaky against the live API with
429s. Adds a shared, xdist-aware rate limiter and a global retry policy
so transient rate-limit errors no longer fail the suite. Mirrors the
same fix recently applied to `langchain-mistralai`.

libs/partners/fireworks/tests/integration_tests/_rate_limiter.py

@@ -0,0 +1,18 @@
+"""Shared rate limiter for Fireworks integration tests.
+
+Scaled by ``PYTEST_XDIST_WORKER_COUNT`` so aggregate QPS across all xdist
+workers stays bounded near the target rate.
+"""
+
+from __future__ import annotations
+
+import os
+
+from langchain_core.rate_limiters import InMemoryRateLimiter
+
+_TARGET_REQUESTS_PER_SECOND = 0.5
+_WORKER_COUNT = max(1, int(os.environ.get("PYTEST_XDIST_WORKER_COUNT", "1")))
+
+rate_limiter = InMemoryRateLimiter(
+    requests_per_second=_TARGET_REQUESTS_PER_SECOND / _WORKER_COUNT,
+)

libs/partners/fireworks/tests/integration_tests/test_chat_models.py

@@ -14,6 +14,7 @@ from pydantic import BaseModel, Field
 from typing_extensions import TypedDict
 
 from langchain_fireworks import ChatFireworks
+from tests.integration_tests._rate_limiter import rate_limiter
 
 _MODEL = "accounts/fireworks/models/gpt-oss-120b"
 
@@ -21,7 +22,9 @@ _MODEL = "accounts/fireworks/models/gpt-oss-120b"
 @pytest.mark.parametrize("strict", [None, True, False])
 def test_tool_choice_bool(strict: bool | None) -> None:  # noqa: FBT001
     """Test that tool choice is respected with different strict values."""
-    llm = ChatFireworks(model="accounts/fireworks/models/kimi-k2p6")
+    llm = ChatFireworks(
+        model="accounts/fireworks/models/kimi-k2p6", rate_limiter=rate_limiter
+    )
 
     class MyTool(BaseModel):
         name: str
@@ -59,7 +62,9 @@ def test_tool_choice_bool(strict: bool | None) -> None:  # noqa: FBT001
 
 async def test_astream() -> None:
     """Test streaming tokens from ChatFireworks."""
-    llm = ChatFireworks(model="accounts/fireworks/models/kimi-k2p6")
+    llm = ChatFireworks(
+        model="accounts/fireworks/models/kimi-k2p6", rate_limiter=rate_limiter
+    )
 
     full: BaseMessageChunk | None = None
     chunks_with_token_counts = 0
@@ -96,7 +101,7 @@ async def test_astream() -> None:
 
 async def test_abatch_tags() -> None:
     """Test batch tokens from ChatFireworks."""
-    llm = ChatFireworks(model=_MODEL)
+    llm = ChatFireworks(model=_MODEL, rate_limiter=rate_limiter)
 
     result = await llm.abatch(
         ["I'm Pickle Rick", "I'm not Pickle Rick"], config={"tags": ["foo"]}
@@ -107,7 +112,7 @@ async def test_abatch_tags() -> None:
 
 async def test_ainvoke() -> None:
     """Test invoke tokens from ChatFireworks."""
-    llm = ChatFireworks(model=_MODEL)
+    llm = ChatFireworks(model=_MODEL, rate_limiter=rate_limiter)
 
     result = await llm.ainvoke("I'm Pickle Rick", config={"tags": ["foo"]})
     assert isinstance(result.content, str)
@@ -115,7 +120,7 @@ async def test_ainvoke() -> None:
 
 def test_invoke() -> None:
     """Test invoke tokens from ChatFireworks."""
-    llm = ChatFireworks(model=_MODEL)
+    llm = ChatFireworks(model=_MODEL, rate_limiter=rate_limiter)
 
     result = llm.invoke("I'm Pickle Rick", config={"tags": ["foo"]})
     assert isinstance(result.content, str)
@@ -157,7 +162,9 @@ def _get_joke_class(
 
 @pytest.mark.parametrize("schema_type", ["pydantic", "typeddict", "json_schema"])
 def test_structured_output_json_schema(schema_type: str) -> None:
-    llm = ChatFireworks(model="accounts/fireworks/models/kimi-k2p6")
+    llm = ChatFireworks(
+        model="accounts/fireworks/models/kimi-k2p6", rate_limiter=rate_limiter
+    )
     schema, validation_function = _get_joke_class(schema_type)  # type: ignore[arg-type]
     chat = llm.with_structured_output(schema, method="json_schema")
 

libs/partners/fireworks/tests/integration_tests/test_standard.py

@@ -8,6 +8,7 @@ from langchain_tests.integration_tests import (  # type: ignore[import-not-found
 )
 
 from langchain_fireworks import ChatFireworks
+from tests.integration_tests._rate_limiter import rate_limiter
 
 
 class TestFireworksStandard(ChatModelIntegrationTests):
@@ -20,6 +21,7 @@ class TestFireworksStandard(ChatModelIntegrationTests):
         return {
             "model": "accounts/fireworks/models/kimi-k2p6",
             "temperature": 0,
+            "rate_limiter": rate_limiter,
         }
 
     @pytest.mark.xfail(reason="Not yet implemented.")