fix(opencode): never use a per-request trace id as the Go session id

This commit is contained in:
Sean Murphy 2026-09-14 08:40:56 -07:00
parent 683e66334f
commit f83c840e9d
2 changed files with 97 additions and 2 deletions

View file

@ -4,6 +4,7 @@ from functools import lru_cache
from types import MappingProxyType
from typing import Final
from litellm.constants import SESSION_ID_GENERATED_METADATA_KEY
from litellm.llms.base_llm.chat.transformation import BaseLLMException
from litellm.secret_managers.main import get_secret_str
@ -201,6 +202,29 @@ def resolve_opencode_api_base(surface: str, api_base: str | None = None) -> str
OPENCODE_SESSION_HEADER: Final = "x-opencode-session"
def _conversation_session_id(litellm_params: Mapping[str, object]) -> str | None:
"""The caller's own session id, or None when the request carries none.
Trace ids and session ids the proxy generated for a session-less request are
per request, so they are never used: they would change the session every turn.
A ``litellm_session_id`` equal to the trace id was derived from it, not sent.
"""
metadata_maps: Final = tuple(
metadata
for metadata in (litellm_params.get("metadata"), litellm_params.get("litellm_metadata"))
if isinstance(metadata, Mapping)
)
if any(metadata.get(SESSION_ID_GENERATED_METADATA_KEY) for metadata in metadata_maps):
return None
session_id: Final = litellm_params.get("litellm_session_id")
candidates: Final = (
*(metadata.get("session_id") for metadata in metadata_maps),
litellm_params.get("session_id"),
session_id if session_id != litellm_params.get("litellm_trace_id") else None,
)
return next((candidate for candidate in candidates if isinstance(candidate, str) and candidate), None)
def with_opencode_session_header(
surface: str,
headers: Mapping[str, str],
@ -214,6 +238,5 @@ def with_opencode_session_header(
"""
if surface != "go" or any(name.lower() == OPENCODE_SESSION_HEADER for name in headers):
return {**headers} # mutable-ok: request handlers keep mutating the headers they are given
known: Final = litellm_params.get("litellm_session_id") or litellm_params.get("litellm_trace_id")
session_id: Final = known if isinstance(known, str) else str(uuid.uuid4())
session_id: Final = _conversation_session_id(litellm_params) or str(uuid.uuid4())
return {**headers, OPENCODE_SESSION_HEADER: session_id} # mutable-ok: request handlers keep mutating the headers

View file

@ -15,6 +15,7 @@ from httpx import Response
import litellm
import pytest
from litellm.constants import SESSION_ID_GENERATED_METADATA_KEY
from litellm.llms.openai.chat.gpt_transformation import (
OpenAIChatCompletionStreamingHandler,
)
@ -362,6 +363,77 @@ class TestMockedCompletion:
assert uuid.UUID(respx_mock.calls[0].request.headers["x-opencode-session"]).version == 4
def test_go_chat_uses_session_id_from_metadata(self, respx_mock, monkeypatch):
"""A session id sent in request metadata keeps the Go session stable across turns."""
respx_mock.post("https://opencode.ai/zen/go/v1/chat/completions").mock(
return_value=Response(200, json=_make_response("deepseek-v4-flash", "ok"))
)
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
litellm.completion(
model="opencode_go/deepseek-v4-flash",
messages=[{"role": "user", "content": "hi"}],
api_key="sk-fake",
metadata={"session_id": "conv-1"},
)
assert respx_mock.calls[0].request.headers["x-opencode-session"] == "conv-1"
def test_go_chat_never_sends_the_per_request_trace_id(self, respx_mock, monkeypatch):
"""A trace id changes every request, so sending it would break prompt caching on every turn."""
respx_mock.post("https://opencode.ai/zen/go/v1/chat/completions").mock(
return_value=Response(200, json=_make_response("deepseek-v4-flash", "ok"))
)
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
litellm.completion(
model="opencode_go/deepseek-v4-flash",
messages=[{"role": "user", "content": "hi"}],
api_key="sk-fake",
litellm_trace_id="trace-123",
)
sent = respx_mock.calls[0].request.headers["x-opencode-session"]
assert sent != "trace-123"
assert uuid.UUID(sent).version == 4
def test_go_chat_never_sends_a_trace_id_from_metadata(self, respx_mock, monkeypatch):
"""A metadata trace id also stands in for a missing session id, but it is still per request."""
respx_mock.post("https://opencode.ai/zen/go/v1/chat/completions").mock(
return_value=Response(200, json=_make_response("deepseek-v4-flash", "ok"))
)
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
litellm.completion(
model="opencode_go/deepseek-v4-flash",
messages=[{"role": "user", "content": "hi"}],
api_key="sk-fake",
metadata={"trace_id": "trace-abc"},
)
sent = respx_mock.calls[0].request.headers["x-opencode-session"]
assert sent != "trace-abc"
assert uuid.UUID(sent).version == 4
def test_go_chat_ignores_a_session_id_the_proxy_generated(self, respx_mock, monkeypatch):
"""The proxy fills a missing session id with the per-request trace id, which is no stable session."""
respx_mock.post("https://opencode.ai/zen/go/v1/chat/completions").mock(
return_value=Response(200, json=_make_response("deepseek-v4-flash", "ok"))
)
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
litellm.completion(
model="opencode_go/deepseek-v4-flash",
messages=[{"role": "user", "content": "hi"}],
api_key="sk-fake",
litellm_session_id="trace-123",
metadata={"session_id": "trace-123", SESSION_ID_GENERATED_METADATA_KEY: True},
)
sent = respx_mock.calls[0].request.headers["x-opencode-session"]
assert sent != "trace-123"
assert uuid.UUID(sent).version == 4
def test_zen_chat_sends_no_session_header(self, respx_mock, monkeypatch):
"""Only the Go surface requires the header."""
respx_mock.post("https://opencode.ai/zen/v1/chat/completions").mock(