fix unpaused user files (#5205)

* Fix user file helm start cmd + legacy file connector incompatibility

* typo

* remove unnecessary logic

* undo

* make recommended changes

* keep comment

* cleanup

* format

---------

Co-authored-by: Dominic Feliton <37809476+dominicfeliton@users.noreply.github.com>

backend/alembic/versions/b558f51620b4_pause_finished_user_file_connectors.py

@@ -0,0 +1,33 @@
+"""Pause finished user file connectors
+
+Revision ID: b558f51620b4
+Revises: 90e3b9af7da4
+Create Date: 2025-08-15 17:17:02.456704
+
+"""
+
+from alembic import op
+
+
+# revision identifiers, used by Alembic.
+revision = "b558f51620b4"
+down_revision = "90e3b9af7da4"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # Set all user file connector credential pairs with ACTIVE status to PAUSED
+    # This ensures user files don't continue to run indexing tasks after processing
+    op.execute(
+        """
+        UPDATE connector_credential_pair
+        SET status = 'PAUSED'
+        WHERE is_user_file = true
+        AND status = 'ACTIVE'
+        """
+    )
+
+
+def downgrade() -> None:
+    pass

backend/onyx/background/celery/tasks/docprocessing/tasks.py

@@ -496,7 +496,14 @@ def check_indexing_completion(
                 ConnectorCredentialPairStatus.SCHEDULED,
                 ConnectorCredentialPairStatus.INITIAL_INDEXING,
             ]:
-                cc_pair.status = ConnectorCredentialPairStatus.ACTIVE
+                # User file connectors must be paused on success
+                # NOTE: _run_indexing doesn't update connectors if the index attempt is the future embedding model
+                # TODO: figure out why this doesn't pause connectors during swap
+                cc_pair.status = (
+                    ConnectorCredentialPairStatus.PAUSED
+                    if cc_pair.is_user_file
+                    else ConnectorCredentialPairStatus.ACTIVE
+                )
                 db_session.commit()
 
             # Clear repeated error state on success

backend/onyx/background/indexing/run_docfetching.py

@@ -271,7 +271,7 @@ def _check_failure_threshold(
 
 # NOTE: this is the old run_indexing function that the new decoupled approach
 # is based on. Leaving this for comparison purposes, but if you see this comment
-# has been here for >1 month, please delete this function.
+# has been here for >2 month, please delete this function.
 def _run_indexing(
     db_session: Session,
     index_attempt_id: int,

backend/onyx/connectors/file/connector.py

@@ -24,6 +24,7 @@ from onyx.file_processing.image_utils import store_image_and_create_section
 from onyx.file_store.file_store import get_default_file_store
 from onyx.utils.logger import setup_logger
 
+
 logger = setup_logger()
 
 
@@ -229,21 +230,18 @@ class LocalFileConnector(LoadConnector):
 
     # Note: file_names is a required parameter, but should not break backwards compatibility.
     # If add_file_names migration is not run, old file connector configs will not have file_names.
-    # This is fine because the configs are not re-used to instantiate the connector.
     # file_names is only used for display purposes in the UI and file_locations is used as a fallback.
     def __init__(
         self,
         file_locations: list[Path | str],
-        file_names: list[
-            str
-        ],  # Must accept this parameter as connector_specific_config is unpacked as args
-        zip_metadata: dict[str, Any],
+        file_names: list[str] | None = None,
+        zip_metadata: dict[str, Any] | None = None,
         batch_size: int = INDEX_BATCH_SIZE,
     ) -> None:
         self.file_locations = [str(loc) for loc in file_locations]
         self.batch_size = batch_size
         self.pdf_pass: str | None = None
-        self.zip_metadata = zip_metadata
+        self.zip_metadata = zip_metadata or {}
 
     def load_credentials(self, credentials: dict[str, Any]) -> dict[str, Any] | None:
         self.pdf_pass = credentials.get("pdf_password")

backend/onyx/llm/chat_llm.py

@@ -24,6 +24,7 @@ from langchain_core.messages import SystemMessageChunk
 from langchain_core.messages.tool import ToolCallChunk
 from langchain_core.messages.tool import ToolMessage
 from langchain_core.prompt_values import PromptValue
+from litellm.utils import get_supported_openai_params
 
 from onyx.configs.app_configs import LOG_DANSWER_MODEL_INTERACTIONS
 from onyx.configs.app_configs import MOCK_LLM_RESPONSE
@@ -52,6 +53,8 @@ litellm.telemetry = False
 _LLM_PROMPT_LONG_TERM_LOG_CATEGORY = "llm_prompt"
 VERTEX_CREDENTIALS_FILE_KWARG = "vertex_credentials"
 VERTEX_LOCATION_KWARG = "vertex_location"
+LEGACY_MAX_TOKENS_KWARG = "max_tokens"
+STANDARD_MAX_TOKENS_KWARG = "max_completion_tokens"
 
 
 class LLMTimeoutError(Exception):
@@ -313,6 +316,14 @@ class DefaultMultiLLM(LLM):
 
         self._model_kwargs = model_kwargs
 
+        self._max_token_param = LEGACY_MAX_TOKENS_KWARG
+        try:
+            params = get_supported_openai_params(model_name, model_provider)
+            if STANDARD_MAX_TOKENS_KWARG in (params or []):
+                self._max_token_param = STANDARD_MAX_TOKENS_KWARG
+        except Exception as e:
+            logger.warning(f"Error getting supported openai params: {e}")
+
     def _safe_model_config(self) -> dict:
         dump = self.config.model_dump()
         dump["api_key"] = mask_string(dump.get("api_key", ""))
@@ -393,7 +404,6 @@ class DefaultMultiLLM(LLM):
                 messages=processed_prompt,
                 tools=tools,
                 tool_choice=tool_choice if tools else None,
-                max_tokens=max_tokens,
                 # streaming choice
                 stream=stream,
                 # model params
@@ -426,6 +436,7 @@ class DefaultMultiLLM(LLM):
                     if structured_response_format
                     else {}
                 ),
+                **({self._max_token_param: max_tokens} if max_tokens else {}),
                 **self._model_kwargs,
             )
         except Exception as e:

backend/tests/unit/onyx/llm/test_chat_llm.py

@@ -148,7 +148,6 @@ def test_multiple_tool_calls(default_multi_llm: DefaultMultiLLM) -> None:
             ],
             tools=tools,
             tool_choice=None,
-            max_tokens=None,
             stream=False,
             temperature=0.0,  # Default value from GEN_AI_TEMPERATURE
             timeout=30,
@@ -294,7 +293,6 @@ def test_multiple_tool_calls_streaming(default_multi_llm: DefaultMultiLLM) -> No
             ],
             tools=tools,
             tool_choice=None,
-            max_tokens=None,
             stream=True,
             temperature=0.0,  # Default value from GEN_AI_TEMPERATURE
             timeout=30,

deployment/helm/charts/onyx/Chart.yaml

@@ -5,7 +5,7 @@ home: https://www.onyx.app/
 sources:
   - "https://github.com/onyx-dot-app/onyx"
 type: application
-version: 0.2.5
+version: 0.2.6
 appVersion: latest
 annotations:
   category: Productivity

deployment/helm/charts/onyx/templates/celery-worker-user-files-indexing.yaml

@@ -43,7 +43,7 @@ spec:
             [
               "celery",
               "-A",
-              "onyx.background.celery.versioned_apps.docprocessing",
+              "onyx.background.celery.versioned_apps.docfetching",
               "worker",
               "--loglevel=INFO",
               "--hostname=user-files-indexing@%n",

web/src/app/admin/configuration/search/UpgradingPage.tsx

@@ -154,6 +154,9 @@ export default function UpgradingPage({
                     re-indexed successfully, the new model will be used for all
                     search queries. Until then, we will use the old model so
                     that no downtime is necessary during this transition.
+                    <br />
+                    Note: User file re-indexing progress is not shown. You will
+                    see this page until all user files are re-indexed!
                   </Text>
 
                   {sortedReindexingProgress ? (