Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.
fix: retry on 500 and 503 for gemini [LET-4185]
jnjpng37d32be58b6e
diff --git a/letta/llm_api/google_vertex_client.py b/letta/llm_api/google_vertex_client.pyindex ff7a37b22..62d93ee88 100644--- a/letta/llm_api/google_vertex_client.py+++ b/letta/llm_api/google_vertex_client.py@@ -3,6 +3,7 @@ import uuidfrom typing import List, Optionalfrom google import genai+from google.genai import errorsfrom google.genai.types import (FunctionCallingConfig,FunctionCallingConfigMode,@@ -67,11 +68,21 @@ class GoogleVertexClient(LLMClientBase):retry_count = 1should_retry = Truewhile should_retry and retry_count <= self.MAX_RETRIES:- response = await client.aio.models.generate_content(- model=llm_config.model,- contents=request_data["contents"],- config=request_data["config"],- )+ try:+ response = await client.aio.models.generate_content(+ model=llm_config.model,+ contents=request_data["contents"],+ config=request_data["config"],+ )+ except errors.APIError as e:+ # Retry on 503 and 500 errors as well, usually ephemeral from Gemini+ if e.code == 503 or e.code == 500:+ logger.warning(f"Received {e}, retrying {retry_count}/{self.MAX_RETRIES}")+ retry_count += 1+ continue+ raise e+ except Exception as e:+ raise eresponse_data = response.model_dump()is_malformed_function_call = self.is_malformed_function_call(response_data)if is_malformed_function_call:diff --git a/tests/integration_test_send_message.py b/tests/integration_test_send_message.pyindex 756adb5d2..b47a4d542 100644--- a/tests/integration_test_send_message.py+++ b/tests/integration_test_send_message.py@@ -1334,6 +1334,7 @@ def test_background_token_streaming_tool_call(messages=messages_to_send,stream_tokens=True,background=True,+ request_options={"timeout_in_seconds": 300},)verify_token_streaming = (llm_config.model_endpoint_type in ["anthropic", "openai", "bedrock"] and "claude-3-5-sonnet" not in llm_config.model