Remove Depends(get_session) from POST /query endpoint to prevent database connections from being held during embedding API calls (1-5+ seconds).
The Memories.get_memories_by_user_id() function manages its own short-lived session internally, releasing the connection before the slow EMBEDDING_FUNCTION() call begins.
Remove Depends(get_session) from POST /create endpoint to prevent database connections from being held during embedding API calls (1-5+ seconds).
The has_permission() and Knowledges.insert_new_knowledge() functions manage their own short-lived sessions internally, releasing connections before the slow embed_knowledge_base_metadata() call begins.
Remove Depends(get_session) from POST /{id}/update endpoint to prevent database connections from being held during embedding API calls (1-5+ seconds).
All database operations (get_knowledge_by_id, has_access, has_permission, update_knowledge_by_id, get_file_metadatas_by_id) manage their own short-lived sessions internally, releasing connections before and after the slow embed_knowledge_base_metadata() call.
Remove Depends(get_session) from the /v1/completions endpoint to prevent database connections from being held during the entire duration of LLM calls.
Previously, the database session was acquired at request start and held until the response completed. Under concurrent load, this exhausted the connection pool, causing QueuePool timeout errors.
The fix allows Models.get_model_by_id() and has_access() to manage their own short-lived sessions internally, releasing the connection immediately after authorization checks complete.
Remove Depends(get_session) from the /v1/chat/completions endpoint to prevent database connections from being held during the entire duration of LLM calls.
Previously, the database session was acquired at request start and held until the streaming response completed. Under concurrent load, this exhausted the connection pool, causing QueuePool timeout errors.
The fix allows Models.get_model_by_id() and has_access() to manage their own short-lived sessions internally, releasing the connection immediately after authorization checks complete.
Remove Depends(get_session) from the /api/chat endpoint to prevent database connections from being held during the entire duration of LLM calls (30-60+ seconds for streaming responses).
Previously, the database session was acquired at request start and held until the streaming response completed. Under concurrent load, this exhausted the connection pool, causing QueuePool timeout errors for other database operations.
The fix allows Models.get_model_by_id() and has_access() to manage their own short-lived sessions internally, releasing the connection immediately after the quick authorization checks complete - before the slow external LLM API call begins.
Remove Depends(get_session) from the /chat/completions endpoint to prevent database connections from being held during the entire duration of LLM calls (30-60+ seconds for streaming responses).
Previously, the database session was acquired at request start and held until the streaming response completed. Under concurrent load, this exhausted the connection pool, causing QueuePool timeout errors for other database operations.
The fix allows Models.get_model_by_id() and has_access() to manage their own short-lived sessions internally, releasing the connection immediately after the quick authorization checks complete - before the slow external LLM API call begins.
Remove Depends(get_session) from POST /add endpoint to prevent database connections from being held during embedding API calls (1-5+ seconds).
The Memories.insert_new_memory() function manages its own short-lived session internally, releasing the connection before the slow EMBEDDING_FUNCTION() call begins.
Remove Depends(get_session) from POST /metadata/reindex endpoint to prevent database connections from being held during N embedding API calls.
This endpoint is CRITICAL as it loops through ALL knowledge bases and calls embed_knowledge_base_metadata() for each one. With the original code, a single connection would be held for the entire duration (potentially minutes for large deployments), completely exhausting the pool.
The Knowledges.get_knowledge_bases() function manages its own short-lived session, releasing the connection before the embedding loop begins.
Remove Depends(get_session) from POST /process/files/batch endpoint to prevent database connections from being held during batch embedding API calls (5-60+ seconds for large batches).
The save_docs_to_vector_db() function makes external embedding API calls. Post-embedding file updates (Files.update_file_by_id) manage their own short-lived sessions internally, releasing connections promptly.
Remove Depends(get_session) from POST /reset to prevent catastrophic connection pool exhaustion.
This endpoint was holding a SINGLE database connection while executing N PARALLEL embedding API calls via asyncio.gather(). For a user with 100 memories, this meant one connection blocked for potentially MINUTES (100 calls * 1-5 seconds each, even in parallel due to rate limits).
A single user triggering /reset could completely starve the connection pool, causing QueuePool timeout errors across the entire application.
The Memories.get_memories_by_user_id() function now manages its own short-lived session, releasing the connection immediately before the massive parallel embedding operation begins.
fix: release database connections immediately after auth instead of holding during LLM calls
Authentication was using Depends(get_session) which holds a database connection
for the entire request lifecycle. For chat completions, this meant connections
were held for 30-60 seconds while waiting for LLM responses, despite only needing
the connection for ~50ms of actual database work.
With a default pool of 15 connections, this limited concurrent chat users to ~15
before pool exhaustion and timeout errors:
sqlalchemy.exc.TimeoutError: QueuePool limit of size 5 overflow 10 reached,
connection timed out, timeout 30.00
The fix removes Depends(get_session) from get_current_user. Each database
operation now manages its own short-lived session internally:
BEFORE: One session held for entire request
──────────────────────────────────────────────────
│ auth │ queries │ LLM wait (30s) │ save │
│ CONNECTION HELD ENTIRE TIME │
──────────────────────────────────────────────────
AFTER: Short-lived sessions, released immediately
┌──────┐ ┌───────┐ ┌──────┐
│ auth │ │ query │ LLM (30s) │ save │
│ 10ms │ │ 20ms │ NO CONNECTION │ 20ms │
└──────┘ └───────┘ └──────┘
This is safe because:
- User model has no lazy-loaded relationships (all simple columns)
- Pydantic conversion (UserModel.model_validate) happens while session is open
- Returned object is pure Pydantic with no SQLAlchemy ties
Combined with the telemetry efficiency fix, this resolves connection pool
exhaustion for high-concurrency deployments, particularly on network-attached
databases like AWS Aurora where connection hold time is more impactful.
fix: use efficient COUNT queries in telemetry metrics to prevent connection pool exhaustion
This fixes database connection pool exhaustion issues reported after v0.7.0,
particularly affecting PostgreSQL deployments on high-latency networks (e.g., AWS Aurora).
## The Problem
The telemetry metrics callbacks (running every 10 seconds via OpenTelemetry's
PeriodicExportingMetricReader) were using inefficient queries that loaded entire
database tables into memory just to count records:
len(Users.get_users()["users"]) # Loads ALL user records to count them
On high-latency network-attached databases like AWS Aurora, this would:
1. Hold database connections for hundreds of milliseconds while transferring data
2. Deserialize all records into Python objects
3. Only then count the list length
Under concurrent load, these long-held connections would stack up and drain the
connection pool, resulting in:
sqlalchemy.exc.TimeoutError: QueuePool limit of size 5 overflow 10 reached,
connection timed out, timeout 30.00
## The Fix
Replace inefficient full-table loads with efficient COUNT(*) queries using
methods that already exist in the codebase:
- `len(Users.get_users()["users"])` → `Users.get_num_users()`
- Similar changes for other telemetry callbacks as needed
COUNT(*) queries use database indexes and return a single integer, completing in
~5-10ms even on Aurora, versus potentially 500ms+ for loading all records.
## Why v0.7.1's Session Sharing Disable "Helped"
The v0.7.1 change to disable DATABASE_ENABLE_SESSION_SHARING by default appeared
to fix the issue, but it was masking the root cause. Disabling session sharing
causes connections to be returned to the pool faster (more connection churn),
which reduced the window for pool exhaustion but didn't address the underlying
inefficient queries.
With this fix, session sharing can be safely re-enabled for deployments that
benefit from it (especially PostgreSQL), as telemetry will no longer hold
connections for extended periods.
## Impact
- Telemetry connection usage drops from potentially seconds to ~30ms total per
collection cycle
- Connection pool pressure from telemetry becomes negligible (~0.3% utilization)
- Enterprise PostgreSQL deployments (Aurora, RDS, etc.) should no longer
experience pool exhaustion under normal load
Refactored the file processing status streaming endpoint to avoid holding
a database connection for the entire stream duration (up to 2 hours).
Changes:
- Each status poll now creates its own short-lived database session instead
of capturing the request's session in the generator closure
- Increased poll interval from 0.5s to 1s, halving database queries with
negligible UX impact
This prevents a single file status stream from blocking a connection pool
slot for hours, which could contribute to pool exhaustion under load.
* feat: add ENABLE_USER_STATUS toggle for admin-controlled user status visibility
feat: add ENABLE_USER_STATUS toggle for admin-controlled user status visibility
Add a new admin panel toggle (Admin > Settings > General) called "User Status" that allows administrators to globally enable or disable user status functionality.
When disabled:
- User status API endpoints return 403 Forbidden
- Status emoji, message, and "Update your status" button are hidden from the user menu
The setting:
- Defaults to True (enabled)
- Can be overridden via ENABLE_USER_STATUS environment variable
- Persists across restarts using PersistentConfig
Files modified:
- backend/open_webui/config.py - Added ENABLE_USER_STATUS PersistentConfig
- backend/open_webui/main.py - App state init and features dict
- backend/open_webui/routers/auths.py - AdminConfig model and endpoints
- backend/open_webui/routers/users.py - 403 guards on status endpoints
- src/lib/components/admin/Settings/General.svelte - Toggle UI
- src/lib/components/layout/Sidebar/UserMenu.svelte - Conditional status display
* Update UserMenu.svelte
feat: add ENABLE_USER_STATUS toggle for admin-controlled user status visibility
Add a new admin panel toggle (Admin > Settings > General) called "User Status" that allows administrators to globally enable or disable user status functionality.
When disabled:
- User status API endpoints return 403 Forbidden
- Active/Away indicator with blinking dot is hidden from the user menu
- Status emoji, message, and "Update your status" button are hidden from the user menu
The setting:
- Defaults to True (enabled)
- Can be overridden via ENABLE_USER_STATUS environment variable
- Persists across restarts using PersistentConfig
Files modified:
- backend/open_webui/config.py - Added ENABLE_USER_STATUS PersistentConfig
- backend/open_webui/main.py - App state init and features dict
- backend/open_webui/routers/auths.py - AdminConfig model and endpoints
- backend/open_webui/routers/users.py - 403 guards on status endpoints
- src/lib/components/admin/Settings/General.svelte - Toggle UI
- src/lib/components/layout/Sidebar/UserMenu.svelte - Conditional status display
* nuke the indicator
* fix
Replaced per-message user lookup with batch fetch using SQL IN clause.
Changes:
- Fetch all message user_ids in a single pass
- Use Users.get_users_by_user_ids() for batch lookup
- Build user mapping to avoid DB calls in loop
- Add early return for empty message lists
Performance: Reduces N+1 queries to 2 queries (messages + users)
Replaced per-message user lookup with batch fetch using SQL IN clause.
Changes:
- Fetch all message user_ids in a single pass
- Use Users.get_users_by_user_ids() for batch lookup
- Build user mapping to avoid DB calls in loop
- Add early return for empty message lists
Performance: Reduces N+1 queries to 2 queries (messages + users)
## Summary
Fixed N+1 query pattern in the `/api/v1/users` endpoint where groups were being fetched for each user individually.
### Problem
The `GET /api/v1/users` endpoint called `Groups.get_groups_by_member_id()` for each user, resulting in:
- 1 query for users
- N queries for groups (one per user)
### Solution
Added a new `Groups.get_groups_by_member_ids()` method that fetches groups for multiple users in a single query using SQL `IN` clause and `JOIN`.
### Changes
- **[groups.py](open_webui/models/groups.py)**: Added `get_groups_by_member_ids()` method
- **[users.py](open_webui/routers/users.py)**: Updated endpoint to use bulk method
### Result
- Before: 1 + N queries
- After: 2 queries total (1 for users, 1 for all groups)
* feat: Add read-only access support for Tools
- Backend: Add write_access field to ToolAccessResponse
- Backend: Update /tools/list to return tools with write_access
- Frontend: Display Read Only badge in Tools list
- Frontend: Disable inputs and save button when no write access
- Frontend: Add readOnly prop to CodeEditor component
* Update Tools.svelte
* fix: Return write_access from getToolById endpoint
fix: Return write_access from getToolById endpoint
- Use ToolAccessResponse instead of raw dict
- Remove inefficient getToolList call in edit page
* refactor: Rename write_access to disabled in ToolkitEditor
- Rename prop from write_access to disabled
- Invert logic where needed
- Update edit page to pass disabled instead of write_access
* rem
* Update +page.svelte
* fix
* Update ToolkitEditor.svelte
* Update CodeEditor.svelte
* Update ToolkitEditor.svelte
* feat: Add read-only access support for Models
- Backend: Add write_access field to ModelAccessResponse
- Backend: Update /models/list to return ModelAccessListResponse
- Frontend: Display Read Only badge in Models list
- Frontend: Disable inputs and save button when no write access
- Frontend: Hide action buttons for read-only models
* fix: Handle ModelAccessListResponse format in getModels API
- Backend returns {items, total} instead of {data}
- Update getModels API to handle both formats for backward compatibility
* fix: Show read-only shared models in workspace list
- Backend: Change search_models permission from 'write' to 'read' to include shared models
- Backend: Keep user_id filter to only show owned/shared models (not all public)
- Frontend: Handle ModelAccessListResponse format in getModels API
* fix: Align Read Only badge inline with model name
* fix: Correct badge placement and fix syntax error
* fix: Resolve badge truncation in Models list
- Add w-full to flex container for proper spacing
- Wrap Badge in div to prevent truncation
- Match Knowledge.svelte badge pattern
* fix: Align Read Only badge with Knowledge.svelte pattern
- Match Knowledge.svelte structure for badge placement
- Actions only show when write_access or admin
- Remove w-full from container to prevent right-overflow
* fix: Return write_access from getModelById endpoint
fix: Return write_access from getModelById endpoint
- Use ModelAccessResponse instead of raw dict
- Remove inefficient getModels call in edit page
* revert
* fix
* fix
* fix
* feat: Add read-only access support for Prompts
- Backend: Add write_access field to PromptAccessResponse
- Backend: Update /prompts/list to return prompts with write_access
- Frontend: Display Read Only badge in Prompts list
- Frontend: Disable inputs and save button when no write access
* feat: Add read-only visual indicators for Prompts workspace
* fix: Return write_access from getPromptByCommand endpoint
- Backend returns write_access directly in response
- Frontend extracts write_access from getPromptByCommand response
- Remove inefficient getPromptList call in edit page
* fix: Align Read Only badge to right in Prompts.svelte
- Title and command stay on left
- Badge pushed to right by justify-between
* fix: Use PromptAccessResponse in get_prompt_by_command endpoint
fix: Use PromptAccessResponse in get_prompt_by_command endpoint
- Return PromptAccessResponse Pydantic model instead of raw dict
- Properly type the response with response_model
Add null check before calling .map() on the API response in getChatList(). When the fetch fails silently or returns null, the function now gracefully returns an empty array instead of crashing with 'Cannot read property map of null'.
Use optional chaining and nullish coalescing when accessing res.OPENAI_API_BASE_URLS and res.OPENAI_API_KEYS. Returns empty array instead of crashing with 'Cannot read property of null'.
Add null check after Chats.get_chat_by_id() in both admin and user code paths. When chat doesn't exist, now raises HTTP 404 instead of crashing with AttributeError when accessing chat.meta.
Add null check in list comprehension before accessing session['id']. When a session_id exists in the room but has been removed from SESSION_POOL, the function now skips it instead of crashing with TypeError.
Add null check after db.get(Chat, chat_id) before accessing chat.share_id. Returns None instead of crashing with AttributeError when chat doesn't exist.
Change 'return Exception(...)' to 'raise Exception(...)' in chat_completed() and chat_action() functions. Returning an exception object instead of raising it causes errors to be silently swallowed, breaking error propagation.
Add null check after Files.get_file_by_id() before accessing file.hash. Raises HTTP 404 instead of crashing with AttributeError when file doesn't exist.
Add null checks for ui settings in update_user_settings_by_session_user() and for user in get_notes(). Prevents AttributeError when ui dict is None or when note's user has been deleted.
Fix implicit None returns in get_model_by_id, get_knowledge_by_id, get_tools_by_id, and get_prompt_by_command. Now properly returns 401 for access denied and 404 for not found instead of silently returning None.
On Windows, clipboard text uses CRLF (\r\n) line endings which can cause newlines to be lost or displayed incorrectly when pasted via the {{CLIPBOARD}} variable.
Fixed by converting CRLF to LF before inserting clipboard text.
Fixes#19370
The Config type incorrectly defined 'pending_user_overlay_description'
but the backend returns 'pending_user_overlay_content' and the
AccountPending.svelte component correctly accesses that property.
This fixes the type mismatch to align with the actual backend response.
Fixes#20284
* chore(deps): update and synchronize backend dependencies
- Updated dependencies in requirements files and pyproject.toml to latest versions.
- Preserved pinned versions for av, pinecone, and pyarrow.
- Added missing dependencies to pyproject.toml to match requirements.txt.
- Ensured all dependency versions are synchronized across files.
* Update pyproject.toml
* sequential
* zero default
* fix
* fix: preserve absolute paths in sqlite+sqlcipher URLs
Previously, the connection logic incorrectly stripped the leading slash
from `sqlite+sqlcipher` paths, forcibly converting absolute paths
(e.g., `sqlite+sqlcipher:////app/data.db`) into relative paths
(which became `app/data.db`). This caused database initialization failures
when using absolute paths, such as with Docker volume mounts.
This change removes the slash-stripping logic, ensuring that absolute
path conventions (starting with `/`) are respected while maintaining
support for relative paths (which do not start with `/`).
* fix: MCP OAuth 2.1 token exchange and multi-node propagation
Fix two MCP OAuth 2.1 bugs affecting tool server authentication:
1. Token exchange failing with duplicate credentials (#19823)
- Removed explicit client_id/client_secret passing in handle_callback()
- Authlib already has credentials configured during add_client(),
passing them again caused concatenation (e.g., "ID1,ID1") and 401 errors
- Added token validation to detect missing access_token and provide
clear error messages instead of cryptic database constraint errors
2. OAuth clients not propagating across multi-node setups (#19901)
- Updated get_client() and get_client_info() to auto-lazy-load
OAuth clients from the Redis-synced TOOL_SERVER_CONNECTIONS config
- Clients are now instantiated on-demand on any node that needs them
Fixes#19823, #19901
* Update db.py
* Update wrappers.py
* sequential
* zero default
* fix
* fix: preserve absolute paths in sqlite+sqlcipher URLs
Previously, the connection logic incorrectly stripped the leading slash
from `sqlite+sqlcipher` paths, forcibly converting absolute paths
(e.g., `sqlite+sqlcipher:////app/data.db`) into relative paths
(which became `app/data.db`). This caused database initialization failures
when using absolute paths, such as with Docker volume mounts.
This change removes the slash-stripping logic, ensuring that absolute
path conventions (starting with `/`) are respected while maintaining
support for relative paths (which do not start with `/`).
* fix (#99)
Co-authored-by: Tim Baek <tim@openwebui.com>
Co-authored-by: Claude <noreply@anthropic.com>
* Update auths.py
* unified logic
* PUSH
* remove getattr
* rem getattr
* whitespace
* Update oauth.py
* trusted header group sync
Added default group re-application after trusted header group sync
* not apply after syncs
* .
* rem
---------
Co-authored-by: Tim Baek <tim@openwebui.com>
Co-authored-by: Claude <noreply@anthropic.com>
Classic298