From b4f9e14a44789b408c3ceab56d01671f33a7b3d0 Mon Sep 17 00:00:00 2001
From: milan <milan@berri.ai>
Date: Tue, 28 Jul 2026 14:14:53 +0000
Subject: [PATCH 1/7] feat(dd_span_tagger): emit litellm_user_email span tag
 for JWT-authenticated requests

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/proxy/dd_span_tagger.py               |  8 +++++-
 .../proxy/test_common_request_processing.py   | 27 ++++++++++++++++++-
 2 files changed, 33 insertions(+), 2 deletions(-)

diff --git a/litellm/proxy/dd_span_tagger.py b/litellm/proxy/dd_span_tagger.py
index 08b7d928d0e..1fda0c638d4 100644
--- a/litellm/proxy/dd_span_tagger.py
+++ b/litellm/proxy/dd_span_tagger.py
@@ -38,11 +38,15 @@ class DDSpanTagger:
         - ``litellm.key_alias``      — human-readable alias for the API key
         - ``litellm.key_hash``       — hashed API key (safe to log; never the raw secret)
         - ``litellm.requested_model``— model name as sent by the client
+        - ``litellm_user_email``     — email of the authenticated user
 
         Use cases:
         - Trace all requests from a specific user/key: filter by ``litellm.key_alias`` or
           ``litellm.key_hash``.
         - Trace all requests for a specific model: filter by ``litellm.requested_model``.
+        - Trace all requests from a specific person under JWT auth, where there is no key_alias:
+          filter by ``@litellm_user_email:"user@example.com"``. The value comes from the JWT claim
+          mapped by ``litellm_jwtauth.user_email_jwt_field``, or from the user row for virtual keys.
 
         Note: key_alias / key_hash are not available for unauthenticated (e.g. 401) requests.
         """
@@ -53,8 +57,10 @@ class DDSpanTagger:
                 set_active_span_tag("litellm.key_hash", str(user_api_key_dict.token))
             if requested_model:
                 set_active_span_tag("litellm.requested_model", str(requested_model))
+            if user_api_key_dict.user_email:
+                set_active_span_tag("litellm_user_email", str(user_api_key_dict.user_email))
         except Exception:
             verbose_proxy_logger.debug(
-                "Failed to tag active ddtrace span with key/model tags",
+                "Failed to tag active ddtrace span with key/model/user tags",
                 exc_info=True,
             )
diff --git a/tests/test_litellm/proxy/test_common_request_processing.py b/tests/test_litellm/proxy/test_common_request_processing.py
index 58f81cdad35..287e6f2f80c 100644
--- a/tests/test_litellm/proxy/test_common_request_processing.py
+++ b/tests/test_litellm/proxy/test_common_request_processing.py
@@ -2318,12 +2318,13 @@ class TestStreamingOverheadHeader:
 class TestDDSpanTaggerTagRequest:
     """Tests for DDSpanTagger.tag_request - key/model DD span tagging."""
 
-    def _make_user_api_key_dict(self, key_alias=None, token=None):
+    def _make_user_api_key_dict(self, key_alias=None, token=None, user_email=None):
         from litellm.proxy._types import UserAPIKeyAuth
 
         d = UserAPIKeyAuth()
         d.key_alias = key_alias
         d.token = token
+        d.user_email = user_email
         return d
 
     def test_tags_key_alias_and_model(self):
@@ -2364,6 +2365,30 @@ class TestDDSpanTaggerTagRequest:
 
         mock_set_tag.assert_called_once_with("litellm.requested_model", "claude-3-5-sonnet")
 
+    def test_tags_user_email(self):
+        """user_email is tagged so JWT-authenticated requests are traceable per person."""
+        user_key = self._make_user_api_key_dict(user_email="user@example.com")
+
+        with patch("litellm.proxy.dd_span_tagger.set_active_span_tag") as mock_set_tag:
+            DDSpanTagger.tag_request(
+                user_api_key_dict=user_key,
+                requested_model=None,
+            )
+
+        mock_set_tag.assert_called_once_with("litellm_user_email", "user@example.com")
+
+    def test_no_user_email_tag_when_absent(self):
+        """No user email tag when the authenticated identity has no email."""
+        user_key = self._make_user_api_key_dict(key_alias="my-prod-key", user_email=None)
+
+        with patch("litellm.proxy.dd_span_tagger.set_active_span_tag") as mock_set_tag:
+            DDSpanTagger.tag_request(
+                user_api_key_dict=user_key,
+                requested_model="gpt-4o",
+            )
+
+        assert all(call.args[0] != "litellm_user_email" for call in mock_set_tag.call_args_list)
+
 
 class TestHasAttributeErrorInChain:
     """Tests for _has_attribute_error_in_chain helper."""

From 819faa01c336d825619b3592ed1c1d763a4156d4 Mon Sep 17 00:00:00 2001
From: milan <milan@berri.ai>
Date: Tue, 28 Jul 2026 14:40:04 +0000
Subject: [PATCH 2/7] refactor(dd_span_tagger): use dotted litellm.user_email
 tag for consistency

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/proxy/dd_span_tagger.py                            | 6 +++---
 tests/test_litellm/proxy/test_common_request_processing.py | 4 ++--
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/litellm/proxy/dd_span_tagger.py b/litellm/proxy/dd_span_tagger.py
index 1fda0c638d4..cafae9bb2a3 100644
--- a/litellm/proxy/dd_span_tagger.py
+++ b/litellm/proxy/dd_span_tagger.py
@@ -38,14 +38,14 @@ class DDSpanTagger:
         - ``litellm.key_alias``      — human-readable alias for the API key
         - ``litellm.key_hash``       — hashed API key (safe to log; never the raw secret)
         - ``litellm.requested_model``— model name as sent by the client
-        - ``litellm_user_email``     — email of the authenticated user
+        - ``litellm.user_email``     — email of the authenticated user
 
         Use cases:
         - Trace all requests from a specific user/key: filter by ``litellm.key_alias`` or
           ``litellm.key_hash``.
         - Trace all requests for a specific model: filter by ``litellm.requested_model``.
         - Trace all requests from a specific person under JWT auth, where there is no key_alias:
-          filter by ``@litellm_user_email:"user@example.com"``. The value comes from the JWT claim
+          filter by ``@litellm.user_email:"user@example.com"``. The value comes from the JWT claim
           mapped by ``litellm_jwtauth.user_email_jwt_field``, or from the user row for virtual keys.
 
         Note: key_alias / key_hash are not available for unauthenticated (e.g. 401) requests.
@@ -58,7 +58,7 @@ class DDSpanTagger:
             if requested_model:
                 set_active_span_tag("litellm.requested_model", str(requested_model))
             if user_api_key_dict.user_email:
-                set_active_span_tag("litellm_user_email", str(user_api_key_dict.user_email))
+                set_active_span_tag("litellm.user_email", str(user_api_key_dict.user_email))
         except Exception:
             verbose_proxy_logger.debug(
                 "Failed to tag active ddtrace span with key/model/user tags",
diff --git a/tests/test_litellm/proxy/test_common_request_processing.py b/tests/test_litellm/proxy/test_common_request_processing.py
index 287e6f2f80c..312c3d2a884 100644
--- a/tests/test_litellm/proxy/test_common_request_processing.py
+++ b/tests/test_litellm/proxy/test_common_request_processing.py
@@ -2375,7 +2375,7 @@ class TestDDSpanTaggerTagRequest:
                 requested_model=None,
             )
 
-        mock_set_tag.assert_called_once_with("litellm_user_email", "user@example.com")
+        mock_set_tag.assert_called_once_with("litellm.user_email", "user@example.com")
 
     def test_no_user_email_tag_when_absent(self):
         """No user email tag when the authenticated identity has no email."""
@@ -2387,7 +2387,7 @@ class TestDDSpanTaggerTagRequest:
                 requested_model="gpt-4o",
             )
 
-        assert all(call.args[0] != "litellm_user_email" for call in mock_set_tag.call_args_list)
+        assert all(call.args[0] != "litellm.user_email" for call in mock_set_tag.call_args_list)
 
 
 class TestHasAttributeErrorInChain:

From b224b15b9d24c15590a81c3c4bb5399d40b7d9e0 Mon Sep 17 00:00:00 2001
From: milan <milan@berri.ai>
Date: Fri, 21 Aug 2026 02:41:55 +0000
Subject: [PATCH 3/7] feat(model_hub): surface model_info.description in model
 group info and Model Hub UI

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/router.py                             |  6 +++++
 litellm/types/router.py                       |  1 +
 tests/test_litellm/test_router.py             | 25 +++++++++++++++++++
 .../src/components/AIHub/ModelHubTable.tsx    |  6 +++++
 .../components/AIHub/ModelHubTableColumns.tsx |  1 +
 .../components/PublicModelHubTableColumns.tsx |  1 +
 .../src/components/public_model_hub.tsx       |  6 +++++
 ui/litellm-dashboard/src/lib/http/schema.d.ts |  2 ++
 8 files changed, 48 insertions(+)

diff --git a/litellm/router.py b/litellm/router.py
index e9eeab53934..90968800460 100644
--- a/litellm/router.py
+++ b/litellm/router.py
@@ -9334,6 +9334,9 @@ class Router:
             model_litellm_params = model.get("litellm_params", {})
             model_info_dict = model.get("model_info", {})
 
+            _raw_description = model_info_dict.get("description")
+            _deployment_description: str | None = _raw_description if isinstance(_raw_description, str) else None
+
             # get model tpm
             _deployment_tpm: int | None = None
             if _deployment_tpm is None:
@@ -9415,6 +9418,7 @@ class Router:
                     **{
                         "model_group": user_facing_model_group_name,
                         "providers": [llm_provider],
+                        "description": _deployment_description,
                         **model_info,
                     }
                 )
@@ -9428,6 +9432,8 @@ class Router:
                 # supports_function_calling == True
                 if llm_provider not in model_group_info.providers:
                     model_group_info.providers.append(llm_provider)
+                if model_group_info.description is None and _deployment_description is not None:
+                    model_group_info.description = _deployment_description
                 if (
                     model_info.get("max_input_tokens", None) is not None
                     and model_info["max_input_tokens"] is not None
diff --git a/litellm/types/router.py b/litellm/types/router.py
index 99a4603ae49..b33e387c508 100644
--- a/litellm/types/router.py
+++ b/litellm/types/router.py
@@ -637,6 +637,7 @@ class ModelGroupInfo(BaseModel):
     supports_function_calling: bool = Field(default=False)
     supported_openai_params: list[str] | None = Field(default=[])
     configurable_clientside_auth_params: CONFIGURABLE_CLIENTSIDE_AUTH_PARAMS = None
+    description: str | None = None
 
     def __init__(self, **data) -> None:
         for field_name, field_type in get_type_hints(self.__class__).items():
diff --git a/tests/test_litellm/test_router.py b/tests/test_litellm/test_router.py
index 9894fcef163..765e479e378 100644
--- a/tests/test_litellm/test_router.py
+++ b/tests/test_litellm/test_router.py
@@ -1626,6 +1626,31 @@ def test_model_group_info_cost_none_when_db_model_info_has_no_cost():
         assert result.output_cost_per_token is None
 
 
+def test_model_group_info_description_from_model_info():
+    """
+    model_info.description set in the config should surface on ModelGroupInfo,
+    including when only a later deployment in the group carries it.
+    """
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": "gpt-4",
+                "litellm_params": {"model": "gpt-4", "api_key": "fake"},
+                "model_info": {},
+            },
+            {
+                "model_name": "gpt-4",
+                "litellm_params": {"model": "gpt-4", "api_key": "fake2"},
+                "model_info": {"description": "State-of-the-art language model."},
+            },
+        ]
+    )
+
+    result = router.get_model_group_info("gpt-4")
+    assert result is not None
+    assert result.description == "State-of-the-art language model."
+
+
 @pytest.mark.parametrize(
     "value,expected",
     [
diff --git a/ui/litellm-dashboard/src/components/AIHub/ModelHubTable.tsx b/ui/litellm-dashboard/src/components/AIHub/ModelHubTable.tsx
index 299104b271b..5fadbe3300d 100644
--- a/ui/litellm-dashboard/src/components/AIHub/ModelHubTable.tsx
+++ b/ui/litellm-dashboard/src/components/AIHub/ModelHubTable.tsx
@@ -647,6 +647,12 @@ const ModelHubTable: React.FC<ModelHubTableProps> = ({ accessToken, publicPage,
                       ))}
                     </div>
                   </div>
+                  {selectedModel.description && (
+                    <div className="col-span-2">
+                      <p className="font-medium">Description:</p>
+                      <p className="mt-1">{selectedModel.description}</p>
+                    </div>
+                  )}
                 </div>
               </div>
 
diff --git a/ui/litellm-dashboard/src/components/AIHub/ModelHubTableColumns.tsx b/ui/litellm-dashboard/src/components/AIHub/ModelHubTableColumns.tsx
index 9f74771f3b1..b025bcf62aa 100644
--- a/ui/litellm-dashboard/src/components/AIHub/ModelHubTableColumns.tsx
+++ b/ui/litellm-dashboard/src/components/AIHub/ModelHubTableColumns.tsx
@@ -31,6 +31,7 @@ export interface ModelHubData {
   supports_function_calling: boolean;
   supported_openai_params?: string[];
   is_public_model_group: boolean;
+  description?: string;
   [key: string]: any;
 }
 
diff --git a/ui/litellm-dashboard/src/components/PublicModelHubTableColumns.tsx b/ui/litellm-dashboard/src/components/PublicModelHubTableColumns.tsx
index ab0ed976149..f3673df8025 100644
--- a/ui/litellm-dashboard/src/components/PublicModelHubTableColumns.tsx
+++ b/ui/litellm-dashboard/src/components/PublicModelHubTableColumns.tsx
@@ -24,6 +24,7 @@ export interface ModelGroupInfo {
   health_status?: string;
   health_response_time?: number;
   health_checked_at?: string;
+  description?: string;
   [key: string]: any;
 }
 
diff --git a/ui/litellm-dashboard/src/components/public_model_hub.tsx b/ui/litellm-dashboard/src/components/public_model_hub.tsx
index 171b7992325..cd10068b9e6 100644
--- a/ui/litellm-dashboard/src/components/public_model_hub.tsx
+++ b/ui/litellm-dashboard/src/components/public_model_hub.tsx
@@ -918,6 +918,12 @@ const PublicModelHub: React.FC<PublicModelHubProps> = ({ accessToken, isEmbedded
                           })}
                         </div>
                       </div>
+                      {selectedModel.description && (
+                        <div className="col-span-2">
+                          <p className="font-medium">Description:</p>
+                          <p className="mt-1">{selectedModel.description}</p>
+                        </div>
+                      )}
                     </div>
 
                     {/* Wildcard Routing Note */}
diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts
index fff48e14ecf..39b3255efb3 100644
--- a/ui/litellm-dashboard/src/lib/http/schema.d.ts
+++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts
@@ -29481,6 +29481,8 @@ export interface components {
         ModelGroupInfoProxy: {
             /** Configurable Clientside Auth Params */
             configurable_clientside_auth_params?: (string | components["schemas"]["ConfigurableClientsideParamsCustomAuth-Output"])[] | null;
+            /** Description */
+            description?: string | null;
             /** Health Checked At */
             health_checked_at?: string | null;
             /** Health Response Time */

From 342b93fe0a92f32b8c2fb4ccf61441cb0a2cf3b1 Mon Sep 17 00:00:00 2001
From: milan <milan@berri.ai>
Date: Fri, 21 Aug 2026 03:27:35 +0000
Subject: [PATCH 4/7] refactor(router): aggregate model group description
 without in-place mutation

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/router.py | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/litellm/router.py b/litellm/router.py
index 90968800460..d65c13d57b2 100644
--- a/litellm/router.py
+++ b/litellm/router.py
@@ -9313,6 +9313,10 @@ class Router:
         model_list: Final = self.get_model_list(model_name=model_group)
         if model_list is None:
             return None
+        model_group_description: Final[str | None] = next(
+            (d for d in (m.get("model_info", {}).get("description") for m in model_list) if isinstance(d, str)),
+            None,
+        )
         for model in model_list:
             is_match = False
             if (
@@ -9334,9 +9338,6 @@ class Router:
             model_litellm_params = model.get("litellm_params", {})
             model_info_dict = model.get("model_info", {})
 
-            _raw_description = model_info_dict.get("description")
-            _deployment_description: str | None = _raw_description if isinstance(_raw_description, str) else None
-
             # get model tpm
             _deployment_tpm: int | None = None
             if _deployment_tpm is None:
@@ -9418,7 +9419,7 @@ class Router:
                     **{
                         "model_group": user_facing_model_group_name,
                         "providers": [llm_provider],
-                        "description": _deployment_description,
+                        "description": model_group_description,
                         **model_info,
                     }
                 )
@@ -9432,8 +9433,6 @@ class Router:
                 # supports_function_calling == True
                 if llm_provider not in model_group_info.providers:
                     model_group_info.providers.append(llm_provider)
-                if model_group_info.description is None and _deployment_description is not None:
-                    model_group_info.description = _deployment_description
                 if (
                     model_info.get("max_input_tokens", None) is not None
                     and model_info["max_input_tokens"] is not None

From 6d2c4899b0f9fa36c59c793657234f75b6a83704 Mon Sep 17 00:00:00 2001
From: yassin <yassin@berri.ai>
Date: Sun, 13 Sep 2026 09:44:22 +0000
Subject: [PATCH 5/7] fix(health): skip background health check DB writes when
 the latest-row read fails

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/proxy/db/health_check_latest.py       |  8 ++++--
 .../health_endpoints/_health_endpoints.py     |  7 +++--
 .../proxy/db/test_health_check_latest.py      | 10 +++++++
 .../proxy/test_health_check_functions.py      | 27 +++++++++++++++----
 4 files changed, 43 insertions(+), 9 deletions(-)

diff --git a/litellm/proxy/db/health_check_latest.py b/litellm/proxy/db/health_check_latest.py
index 35bc838379c..21438f095bb 100644
--- a/litellm/proxy/db/health_check_latest.py
+++ b/litellm/proxy/db/health_check_latest.py
@@ -74,10 +74,14 @@ class LatestHealthCheckRow(BaseModel):
 _ROWS_ADAPTER: Final = TypeAdapter(tuple[LatestHealthCheckRow, ...])
 
 
+async def query_latest_health_checks(prisma_client: PrismaClient) -> tuple[LatestHealthCheckRow, ...]:
+    rows: Final = await prisma_client.db.query_raw(LATEST_HEALTH_CHECKS_SQL)
+    return _ROWS_ADAPTER.validate_python(rows)
+
+
 async def fetch_latest_health_checks(prisma_client: PrismaClient) -> tuple[LatestHealthCheckRow, ...]:
     try:
-        rows: Final = await prisma_client.db.query_raw(LATEST_HEALTH_CHECKS_SQL)
-        return _ROWS_ADAPTER.validate_python(rows)
+        return await query_latest_health_checks(prisma_client)
     except Exception as query_err:  # noqa: BLE001  # health decorates other reads; a driver error must not fail them
         verbose_proxy_logger.error("Error getting all latest health checks: %s", query_err)
         return ()
diff --git a/litellm/proxy/health_endpoints/_health_endpoints.py b/litellm/proxy/health_endpoints/_health_endpoints.py
index db6ec754c6e..64fd59bbe44 100644
--- a/litellm/proxy/health_endpoints/_health_endpoints.py
+++ b/litellm/proxy/health_endpoints/_health_endpoints.py
@@ -45,7 +45,10 @@ from litellm.proxy.auth.auth_utils import (
 from litellm.proxy.auth.model_checks import get_key_models
 from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
 from litellm.proxy.db.exception_handler import PrismaDBExceptionHandler
-from litellm.proxy.db.health_check_latest import LatestHealthCheckRow
+from litellm.proxy.db.health_check_latest import (
+    LatestHealthCheckRow,
+    query_latest_health_checks,
+)
 from litellm.proxy.db.proxy_worker_heartbeat import count_live_proxy_workers
 from litellm.proxy.health_check import (
     ADMIN_ONLY_HEALTH_DISPLAY_PARAMS,
@@ -876,7 +879,7 @@ async def _save_background_health_checks_to_db(
         )
 
         # Step 3: Get latest health checks for all models in one query to compare status
-        latest_checks: Final = await prisma_client.get_all_latest_health_checks()
+        latest_checks: Final = await query_latest_health_checks(prisma_client)
         latest_checks_map: Final = {}
         for check in latest_checks:
             # Use model_id as primary key, fallback to model_name
diff --git a/tests/test_litellm/proxy/db/test_health_check_latest.py b/tests/test_litellm/proxy/db/test_health_check_latest.py
index 529e4d8f2e9..6322891ae9e 100644
--- a/tests/test_litellm/proxy/db/test_health_check_latest.py
+++ b/tests/test_litellm/proxy/db/test_health_check_latest.py
@@ -8,6 +8,7 @@ from litellm.proxy.db.health_check_latest import (
     LATEST_HEALTH_CHECKS_SQL,
     fetch_latest_health_checks,
     fetch_latest_health_checks_for_models,
+    query_latest_health_checks,
 )
 
 
@@ -83,6 +84,15 @@ async def test_fetch_all_degrades_to_no_rows_when_the_query_fails():
     assert await fetch_latest_health_checks(prisma) == ()
 
 
+@pytest.mark.asyncio
+async def test_query_all_raises_when_the_query_fails_instead_of_reading_as_an_empty_table():
+    """The background save decides what to write from this read; a failure has to be told apart from no rows."""
+    prisma = _prisma([])
+    prisma.db.query_raw.side_effect = RuntimeError("db down")
+    with pytest.raises(RuntimeError, match="db down"):
+        await query_latest_health_checks(prisma)
+
+
 @pytest.mark.asyncio
 async def test_fetch_all_degrades_to_no_rows_for_a_malformed_row():
     assert await fetch_latest_health_checks(_prisma([{"unexpected": "shape"}])) == ()
diff --git a/tests/test_litellm/proxy/test_health_check_functions.py b/tests/test_litellm/proxy/test_health_check_functions.py
index c0c853ae2c5..7d6c5d3cebe 100644
--- a/tests/test_litellm/proxy/test_health_check_functions.py
+++ b/tests/test_litellm/proxy/test_health_check_functions.py
@@ -374,7 +374,7 @@ async def test_save_background_health_checks_to_db():
     """Test the main background health check save function"""
     mock_prisma = MagicMock()
     mock_prisma.save_health_check_result = AsyncMock()
-    mock_prisma.get_all_latest_health_checks = AsyncMock(return_value=[])
+    mock_prisma.db.query_raw = AsyncMock(return_value=[])
 
     model_list = [
         {
@@ -398,9 +398,9 @@ async def test_save_background_health_checks_to_db():
         "background_health_check",
     )
 
-    # Should call get_all_latest_health_checks and save_health_check_result, and report completion
+    # Should read the latest rows and save_health_check_result, and report completion
     assert persisted is True
-    mock_prisma.get_all_latest_health_checks.assert_called_once()
+    mock_prisma.db.query_raw.assert_awaited_once()
     mock_prisma.save_health_check_result.assert_called_once()
 
     call_kwargs = mock_prisma.save_health_check_result.call_args[1]
@@ -493,7 +493,7 @@ def _one_model_setup():
 @pytest.mark.asyncio
 async def test_save_background_health_checks_to_db_returns_false_when_a_write_fails():
     mock_prisma = MagicMock()
-    mock_prisma.get_all_latest_health_checks = AsyncMock(return_value=[])
+    mock_prisma.db.query_raw = AsyncMock(return_value=[])
     mock_prisma.save_health_check_result = AsyncMock(return_value=None)
     model_list, healthy_endpoints, unhealthy_endpoints = _one_model_setup()
 
@@ -504,6 +504,23 @@ async def test_save_background_health_checks_to_db_returns_false_when_a_write_fa
     assert (persisted, mock_prisma.save_health_check_result.await_count) == (False, 1)
 
 
+@pytest.mark.asyncio
+async def test_save_background_health_checks_to_db_writes_nothing_when_the_latest_row_read_fails(mock_prisma):
+    """
+    A failed dedup read must not read as an empty table. Treated that way, every model was written on every
+    cycle by every pod while the read kept failing, which is what filled the table in production.
+    """
+    mock_prisma.db.query_raw = AsyncMock(side_effect=RuntimeError("db down"))
+    mock_prisma.save_health_check_result = AsyncMock(return_value={"id": "row"})
+    model_list, healthy_endpoints, unhealthy_endpoints = _one_model_setup()
+
+    persisted = await _save_background_health_checks_to_db(
+        mock_prisma, model_list, healthy_endpoints, unhealthy_endpoints, 1234567890.0, "background_health_check"
+    )
+
+    assert (persisted, mock_prisma.save_health_check_result.await_count) == (False, 0)
+
+
 @pytest.mark.asyncio
 async def test_save_background_health_checks_to_db_no_prisma():
     """Test graceful handling when no prisma client"""
@@ -515,7 +532,7 @@ async def test_save_background_health_checks_to_db_no_prisma():
 async def test_save_background_health_checks_to_db_exception_handling():
     """Test exception handling in background health check save"""
     mock_prisma = MagicMock()
-    mock_prisma.get_all_latest_health_checks = AsyncMock(side_effect=Exception("DB Error"))
+    mock_prisma.db.query_raw = AsyncMock(side_effect=Exception("DB Error"))
 
     model_list = [
         {

From 834290290144b121d027401e3d27e3299f93a2ba Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 02:03:48 +0000
Subject: [PATCH 6/7] fix(images): keep drop_params out of images/edits
 multipart

Per-model drop_params and additional_drop_params were flattened into the
OpenAI images/edits body. Strip them only when building image-edit
non_default_params. Do not add them to all_litellm_params: that list is
popped before completion and embedding apply the drop_params flag

Fixes #40153

Co-authored-by: Zsanz3 <Zsanz3@users.noreply.github.com>
---
 litellm/images/main.py                        |  7 +++--
 .../images/test_image_edit_extra_params.py    | 29 +++++++++++++++++++
 2 files changed, 34 insertions(+), 2 deletions(-)

diff --git a/litellm/images/main.py b/litellm/images/main.py
index 6a94e7c8df2..18255e88c24 100644
--- a/litellm/images/main.py
+++ b/litellm/images/main.py
@@ -64,6 +64,9 @@ from litellm.utils import (
 # Cache for ImageEditRequestUtils to avoid repeated __getattr__ calls
 _ImageEditRequestUtils_cache: Optional["ImageEditRequestUtils"] = None
 
+# Kept out of all_litellm_params because that list is stripped before drop_params is applied.
+_IMAGE_EDIT_CONTROL_KWARGS: Final = frozenset(("drop_params", "additional_drop_params"))
+
 
 def _get_ImageEditRequestUtils() -> "ImageEditRequestUtils":
     """Get ImageEditRequestUtils, loading it lazily if needed."""
@@ -759,8 +762,8 @@ def image_edit(
         litellm_params_list: Final = all_litellm_params
         default_params: Final = openai_params + litellm_params_list
         non_default_params: Final = {
-            k: v for k, v in kwargs.items() if k not in default_params
-        }  # model-specific params - pass them straight to the model/provider
+            k: v for k, v in kwargs.items() if k not in default_params and k not in _IMAGE_EDIT_CONTROL_KWARGS
+        }
         litellm_logging_obj: Final[LiteLLMLoggingObj] = kwargs.get("litellm_logging_obj")
         litellm_call_id: Final[str | None] = kwargs.get("litellm_call_id", None)
         model_info: Final = kwargs.get("model_info", None)
diff --git a/tests/test_litellm/images/test_image_edit_extra_params.py b/tests/test_litellm/images/test_image_edit_extra_params.py
index 088faafa9f3..1d1fce2bf63 100644
--- a/tests/test_litellm/images/test_image_edit_extra_params.py
+++ b/tests/test_litellm/images/test_image_edit_extra_params.py
@@ -121,6 +121,35 @@ def test_image_edit_forwards_scalar_array_as_repeated_fields():
     assert b"style_a" in body and b"style_b" in body and b"style_c" in body
 
 
+@pytest.mark.parametrize("drop_params", [True, False])
+def test_image_edit_strips_per_model_drop_params_from_multipart(drop_params):
+    """Per-model drop_params / additional_drop_params are LiteLLM control
+    flags. If they survive into non_default_params they are flattened onto the
+    OpenAI images/edits multipart body and the provider rejects the call with
+    Unknown parameter: 'drop_params'. Issue #40153."""
+    captured = {}
+    client = HTTPHandler(client=httpx.Client(transport=httpx.MockTransport(_capture_image_edit_request(captured))))
+
+    litellm.image_edit(
+        model="openai/gpt-image-1",
+        image=PNG_BYTES,
+        prompt="Make the background blue",
+        api_key="sk-test",
+        api_base="https://edit.example/v1",
+        client=client,
+        seed=42,
+        drop_params=drop_params,
+        additional_drop_params=["quality_level"],
+    )
+
+    fields = _multipart_text_fields(captured["content_type"], captured["body"])
+    assert "drop_params" not in fields
+    assert "additional_drop_params" not in fields
+    assert fields["seed"] == "42"
+    assert fields["prompt"] == "Make the background blue"
+    assert fields["model"] == "gpt-image-1"
+
+
 @pytest.mark.asyncio
 async def test_aimage_edit_forwards_extra_body():
     """aimage_edit used to drop extra_headers/extra_query/extra_body when

From 13638fb3b9c6deeee8e3402af2aa289e41e68a87 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Wed, 16 Sep 2026 07:34:28 +0000
Subject: [PATCH 7/7] test(images): annotate drop_params in image-edit
 multipart regression

Greptile flagged the parametrized test argument as untyped.

Co-authored-by: Zsanz3 <Zsanz3@users.noreply.github.com>
---
 tests/test_litellm/images/test_image_edit_extra_params.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/test_litellm/images/test_image_edit_extra_params.py b/tests/test_litellm/images/test_image_edit_extra_params.py
index 1d1fce2bf63..5b61b474c71 100644
--- a/tests/test_litellm/images/test_image_edit_extra_params.py
+++ b/tests/test_litellm/images/test_image_edit_extra_params.py
@@ -122,7 +122,7 @@ def test_image_edit_forwards_scalar_array_as_repeated_fields():
 
 
 @pytest.mark.parametrize("drop_params", [True, False])
-def test_image_edit_strips_per_model_drop_params_from_multipart(drop_params):
+def test_image_edit_strips_per_model_drop_params_from_multipart(drop_params: bool):
     """Per-model drop_params / additional_drop_params are LiteLLM control
     flags. If they survive into non_default_params they are flattened onto the
     OpenAI images/edits multipart body and the provider rejects the call with