fix(proxy): keep bug report notice out of client error bodies

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/litellm_core_utils/bug_report.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import dataclasses
 import os
 import platform
 import traceback
@@ -15,8 +16,10 @@ from litellm._version import version as litellm_version
 ISSUE_URL_BASE: Final = "https://github.com/BerriAI/litellm/issues/new"
 MAX_URL_LENGTH: Final = 6000
 MAX_MESSAGE_CHARS: Final = 600
+MAX_CONTEXT_CHARS: Final = 200
 MAX_FRAMES: Final = 12
 DISABLE_ENV_VAR: Final = "LITELLM_DISABLE_BUG_REPORT_LINK"
+NOTICE_PREFIX: Final = "This looks like a bug in LiteLLM rather than in your request."
 _SHORTENED_MESSAGE_LENGTHS: Final = (480, 360, 240, 120, 0)
 
 Surface = Literal["sdk", "proxy"]
@@ -66,7 +69,7 @@ def _get_litellm_frames(exc: BaseException) -> tuple[str, ...]:
 def _redact_context(value: str | None) -> str | None:
     if value is None:
         return None
-    return redact_secrets(value)[:MAX_MESSAGE_CHARS]
+    return redact_secrets(value)[:MAX_CONTEXT_CHARS]
 
 
 def build_bug_report(
@@ -142,11 +145,19 @@ def bug_report_issue_url(report: BugReport) -> str:
         *(_issue_url(report, message, frames[index:]) for index in range(len(frames) + 1)),
         *(_issue_url(report, message[:length], ()) for length in _SHORTENED_MESSAGE_LENGTHS if length < len(message)),
     )
-    return next((candidate for candidate in candidates if len(candidate) <= MAX_URL_LENGTH), _issue_url(report, "", ()))
+    return next(
+        (candidate for candidate in candidates if len(candidate) <= MAX_URL_LENGTH),
+        _issue_url(dataclasses.replace(report, call_type=None, model=None, custom_llm_provider=None), "", ()),
+    )
+
+
+def strip_bug_report_notice(message: str) -> str:
+    index: Final = message.find(NOTICE_PREFIX)
+    return message if index == -1 else message[:index].rstrip()
 
 
 def bug_report_notice(report: BugReport) -> str:
     return (
-        "This looks like a bug in LiteLLM rather than in your request. File it with one click "
+        f"{NOTICE_PREFIX} File it with one click "
         f"(prefilled and redacted, review before submitting): {bug_report_issue_url(report)}"
     )

litellm/proxy/common_request_processing.py

@@ -50,6 +50,7 @@ from litellm.litellm_core_utils.bug_report import (
     bug_report_enabled,
     bug_report_notice,
     build_bug_report,
+    strip_bug_report_notice,
 )
 from litellm.litellm_core_utils.core_helpers import (
     get_or_create_metadata_bucket,
@@ -3681,8 +3682,11 @@ class ProxyBaseLLMRequestProcessing:
                         )
                     )
                 )
+        client_message: Final = getattr(e, "message", error_msg)
         raise ProxyException(
-            message=redact_internal_details_from_client_message(getattr(e, "message", error_msg)),
+            message=redact_internal_details_from_client_message(
+                strip_bug_report_notice(client_message) if isinstance(client_message, str) else error_msg
+            ),
             type=openai_error_type(e, _code),
             param=openai_error_param(e),
             openai_code=getattr(e, "code", None),

litellm/proxy/utils.py

@@ -59,6 +59,7 @@ from litellm.litellm_core_utils.bug_report import (
     bug_report_enabled,
     bug_report_notice,
     build_bug_report,
+    strip_bug_report_notice,
 )
 from litellm.proxy._types import (
     CommonProxyErrors,
@@ -7844,7 +7845,7 @@ def handle_exception_on_proxy(e: Exception, litellm_call_id: str | None = None)
     if _status_code == status.HTTP_500_INTERNAL_SERVER_ERROR and bug_report_enabled() and isinstance(e, BaseException):
         verbose_proxy_logger.error(bug_report_notice(build_bug_report(e, surface="proxy")))
     return ProxyException(
-        message=str(e),
+        message=strip_bug_report_notice(str(e)),
         type=ProxyErrorTypes.internal_server_error,
         param=openai_error_param(e),
         headers=headers,

tests/test_litellm/litellm_core_utils/test_bug_report.py

@@ -13,7 +13,9 @@ from litellm.litellm_core_utils.bug_report import (
     MAX_URL_LENGTH,
     bug_report_enabled,
     bug_report_issue_url,
+    bug_report_notice,
     build_bug_report,
+    strip_bug_report_notice,
 )
 from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
 
@@ -54,6 +56,18 @@ def test_issue_url_is_bounded_for_long_messages():
     assert "RuntimeError" in query["description"][0]
 
 
+def test_issue_url_is_bounded_for_long_non_ascii_context():
+    report = build_bug_report(
+        ValueError("故障" * 10_000),
+        surface="sdk",
+        model="模型" * 300,
+        custom_llm_provider="供給" * 300,
+        call_type="呼出" * 300,
+    )
+
+    assert len(bug_report_issue_url(report)) <= MAX_URL_LENGTH
+
+
 def test_issue_url_builds_without_a_traceback():
     exc = RuntimeError("no traceback")
     assert exc.__traceback__ is None
@@ -80,3 +94,11 @@ def test_proxy_provider_uses_translation_domain():
     query = parse_qs(urlparse(bug_report_issue_url(report)).query)
 
     assert query["domain"] == ["LLM translation: a specific provider's request or response"]
+
+
+def test_strip_bug_report_notice():
+    report = build_bug_report(RuntimeError("boom"), surface="sdk")
+    notice = bug_report_notice(report)
+
+    assert strip_bug_report_notice(f"boom\n{notice}") == "boom"
+    assert strip_bug_report_notice("boom") == "boom"

tests/test_litellm/proxy/test_common_request_processing.py

@@ -14,7 +14,12 @@ from fastapi.responses import JSONResponse, StreamingResponse
 
 import litellm
 from litellm._uuid import uuid
-from litellm.litellm_core_utils.bug_report import DISABLE_ENV_VAR, ISSUE_URL_BASE
+from litellm.litellm_core_utils.bug_report import (
+    DISABLE_ENV_VAR,
+    ISSUE_URL_BASE,
+    bug_report_notice,
+    build_bug_report,
+)
 from litellm.constants import (
     CLIENT_REQUESTED_MODEL_SCOPE_KEY,
     MAX_LITELLM_CALL_ID_LENGTH,
@@ -4259,6 +4264,21 @@ class TestHandleLLMApiExceptionRetryAfter:
         proxy_exc = await self._invoke(ValueError("some other failure"))
         assert "retry-after" not in proxy_exc.headers
 
+    async def test_handle_llm_api_exception_strips_bug_report_notice_from_client_message(self, caplog):
+        report = build_bug_report(RuntimeError("boom"), surface="sdk")
+        notice = bug_report_notice(report)
+        exc = litellm.APIConnectionError(
+            message=f"boom\n{notice}",
+            model="gpt-4o",
+            llm_provider="openai",
+        )
+
+        with caplog.at_level("ERROR"):
+            proxy_exc = await self._invoke(exc)
+
+        assert ISSUE_URL_BASE not in proxy_exc.message
+        assert ISSUE_URL_BASE in caplog.text
+
     async def test_handle_llm_api_exception_retry_after_survives_callback_headers(self):
         from litellm.types.router import RouterRateLimitError