test(mistralai): stabilize integration tests with rate limiting and retries (#37588)

Mistral integration tests have been flaky against the live API. This
adds a shared, xdist-aware rate limiter and a global retry policy so
transient 429s no longer fail the suite.

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

@@ -0,0 +1,18 @@
+"""Shared rate limiter for Mistral 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/mistralai/tests/integration_tests/test_chat_models.py

@@ -13,11 +13,12 @@ from pydantic import BaseModel
 from typing_extensions import TypedDict
 
 from langchain_mistralai.chat_models import ChatMistralAI
+from tests.integration_tests._rate_limiter import rate_limiter
 
 
 async def test_astream() -> None:
     """Test streaming tokens from ChatMistralAI."""
-    llm = ChatMistralAI()
+    llm = ChatMistralAI(rate_limiter=rate_limiter)
 
     full: BaseMessageChunk | None = None
     chunks_with_token_counts = 0
@@ -70,7 +71,7 @@ def _check_parsed_result(result: Any, schema: Any) -> None:
 
 @pytest.mark.parametrize("schema", [Book, BookDict, Book.model_json_schema()])
 def test_structured_output_json_schema(schema: Any) -> None:
-    llm = ChatMistralAI(model="ministral-8b-latest")  # type: ignore[call-arg]
+    llm = ChatMistralAI(model="ministral-8b-latest", rate_limiter=rate_limiter)  # type: ignore[call-arg]
     structured_llm = llm.with_structured_output(schema, method="json_schema")
 
     messages = [
@@ -91,7 +92,7 @@ def test_structured_output_json_schema(schema: Any) -> None:
 
 @pytest.mark.parametrize("schema", [Book, BookDict, Book.model_json_schema()])
 async def test_structured_output_json_schema_async(schema: Any) -> None:
-    llm = ChatMistralAI(model="ministral-8b-latest")  # type: ignore[call-arg]
+    llm = ChatMistralAI(model="ministral-8b-latest", rate_limiter=rate_limiter)  # type: ignore[call-arg]
     structured_llm = llm.with_structured_output(schema, method="json_schema")
 
     messages = [
@@ -116,6 +117,7 @@ def test_retry_parameters(caplog: pytest.LogCaptureFixture) -> None:
     mistral = ChatMistralAI(
         timeout=1,  # Very short timeout to trigger timeouts
         max_retries=3,  # Should retry 3 times
+        rate_limiter=rate_limiter,
     )
 
     # Simple test input that should take longer than 1 second to process
@@ -148,7 +150,7 @@ def test_retry_parameters(caplog: pytest.LogCaptureFixture) -> None:
 
 
 def test_reasoning() -> None:
-    model = ChatMistralAI(model="magistral-medium-latest")  # type: ignore[call-arg]
+    model = ChatMistralAI(model="magistral-medium-latest", rate_limiter=rate_limiter)  # type: ignore[call-arg]
     input_message = {
         "role": "user",
         "content": "Hello, my name is Bob.",
@@ -172,7 +174,11 @@ def test_reasoning() -> None:
 
 
 def test_reasoning_v1() -> None:
-    model = ChatMistralAI(model="magistral-medium-latest", output_version="v1")  # type: ignore[call-arg]
+    model = ChatMistralAI(  # type: ignore[call-arg]
+        model="magistral-medium-latest",
+        output_version="v1",
+        rate_limiter=rate_limiter,
+    )
     input_message = {
         "role": "user",
         "content": "Hello, my name is Bob.",

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

@@ -7,6 +7,7 @@ from langchain_tests.integration_tests import (  # type: ignore[import-not-found
 )
 
 from langchain_mistralai import ChatMistralAI
+from tests.integration_tests._rate_limiter import rate_limiter
 
 
 class TestMistralStandard(ChatModelIntegrationTests):
@@ -16,7 +17,11 @@ class TestMistralStandard(ChatModelIntegrationTests):
 
     @property
     def chat_model_params(self) -> dict:
-        return {"model": "mistral-large-latest", "temperature": 0}
+        return {
+            "model": "mistral-large-latest",
+            "temperature": 0,
+            "rate_limiter": rate_limiter,
+        }
 
     @property
     def supports_json_mode(self) -> bool: