Swap the JSON encoder/decoder used across the backend from stdlib json to
orjson when ENABLE_ORJSON is set — HTTP request bodies, JSONResponse
bodies, upstream provider responses, SSE chunks, and socket.io/Redis
payloads.
The flag defaults to off, in which case the app uses stdlib json and
engineio's codec verbatim, so default behaviour is unchanged.
- json_codec exports JSONCodec (stdlib json or the orjson codec) and
SOCKETIO_JSON (engineio's codec or the orjson codec); call sites import
JSONCodec and stay implementation-agnostic
- apply_orjson_http_json() is a no-op when the flag is off, leaving
starlette's Request.json / JSONResponse.render untouched
- the orjson codec falls back to the stdlib for inputs orjson rejects
(non-str dict keys, ints beyond 64 bits, NaN literals)
- orjson is imported only when the flag is on
- FastAPI(default_response_class=...) is deliberately not used: an
explicit default disables the Pydantic direct-to-bytes fast path for
response_model routes
build_matcher compiled a caller-supplied pattern with Python's backtracking re and ran it over every line of every reachable file, with no timeout, no thread offload and no length caps. is_regex_pattern promotes any pattern containing a metacharacter, and a bare pipe counts, so no explicit regex flag is needed to reach the compiler. The search loop is synchronous inside an async handler, and UVICORN_WORKERS defaults to 1, so the cost lands on every other user of the instance. MAX_GREP_RESULTS bounds how many matches are reported, not how much work is done.
Backtracking cost is exponential in the length of the text being matched, so capping the pattern or the line does not bound it: the subject in the measurements below is 30 characters. `(x|x)*y` against a line of 30 x took 80 seconds, `(a+)+$` against 32 a took 169 seconds, and the same subject with a literal pattern took 0.6 microseconds.
Matching now runs on the regex module, which accepts a per-search timeout that re has no equivalent for. The timeout is the actual bound: regex resolves many classic catastrophic patterns instantly, but not all of them, and `(a|aa)+$` and `(?:a|a)*$` still need it. The budget covers a whole tool call rather than a single search, because a pipeline builds one matcher per segment and a per-search budget would multiply by segment count, and because a per-line timeout would allow timeout multiplied by line count. It is carried in a context variable so one command shares it without threading a parameter through every handler, and it is charged only for time spent inside search(), so database round-trips and other coroutines cannot consume it. Exhausting it raises, and both entry points already render that as an error for the model to read.
Note for anyone tracking search behaviour: re and the regex module define \w, \W and \b differently on non-ASCII text. re follows str.isalnum(), the regex module follows UTS#18, so \w no longer matches superscripts and fractions such as the ones in Nd-adjacent categories, and now does match combining marks. POSIX classes like [[:alpha:]] are interpreted rather than read as a literal set, and \p{...} compiles instead of erroring. Results on ASCII content are unchanged.
regex was already installed as a transitive dependency of nltk, tiktoken and transformers. It is now declared directly, pinned in pyproject.toml and requirements.txt to the version the lockfile already resolves.
aiohttp resolves every hostname with ThreadedResolver unless the aiodns package is importable, and ThreadedResolver runs socket.getaddrinfo on asyncio's default ThreadPoolExecutor. That executor is capped at min(32, cpu_count + 4) threads and is shared with every other piece of blocking work posted to it, so DNS is currently a bounded blocking resource sitting in front of every model call, every web search fetch, every RAG page load and every tool call. In plain terms: once that pool is busy, requests wait on name lookups that should never have occupied a thread at all.
This is a dependency-only change. aiohttp sets `DefaultResolver = AsyncResolver` as soon as aiodns is importable (aiohttp/resolver.py), so resolution moves onto the event loop via c-ares with zero application code touched. That is deliberate rather than lazy: there are 50 `aiohttp.ClientSession(...)` construction sites in the backend, most building a fresh default connector per call, and the alternative of passing `resolver=aiohttp.AsyncResolver()` explicitly would mean touching all of them and re-touching every future one. The shared pool in `utils/session_pool.py` does set `ttl_dns_cache`, but that only helps the shared pool. Every per-request session, including `SafeWebBaseLoader._fetch()` which builds a new session per URL, starts with a cold DNS cache and resolves from scratch.
It also unbreaks a code path that is dead today. `backend/open_webui/retrieval/loaders/mistral.py:480` constructs `aiohttp.AsyncResolver()` unconditionally, and `AsyncResolver.__init__` raises `RuntimeError("Resolver requires aiodns library")` when aiodns is absent, so the Mistral OCR content extraction engine fails on a stock install. This supplies the dependency that line already assumes. Once it is present that kwarg is redundant, since it now names the default, and dropping it is a reasonable follow-up. Reproduced by blocking the aiodns import:
```
aiodns importable: False
DefaultResolver: ThreadedResolver
AsyncResolver(): RuntimeError: Resolver requires aiodns library
```
## Benchmarks
Both sides run the real aiohttp resolver classes. The DNS wire time is replaced by an identical fixed 50ms delay on both sides, so the only variable measured is where that delay is spent. 24 cores, so the default executor holds 28 threads. `exec_max` is the worst latency an unrelated `run_in_executor` job suffered while the lookups were in flight.
Concurrent lookups, wall time:
| concurrent lookups | ThreadedResolver | AsyncResolver | speedup | exec_max before | exec_max after |
|---|---|---|---|---|---|
| 16 | 54.3ms | 41.0ms | 1.3x | 2.1ms | 1.9ms |
| 32 | 101.9ms | 50.6ms | 2.0x | 36.6ms | 1.8ms |
| 64 | 152.5ms | 43.2ms | 3.5x | 88.4ms | 2.0ms |
| 128 | 254.9ms | 44.5ms | 5.7x | 190.3ms | 2.2ms |
| 256 | 508.2ms | 50.7ms | 10.0x | 443.8ms | 2.4ms |
| 512 | 965.2ms | 47.8ms | 20.2x | 900.2ms | 2.6ms |
ThreadedResolver scales linearly with concurrency because it can only run 28 lookups at a time. AsyncResolver stays flat at roughly the cost of one lookup.
The reverse direction is worse and is not hypothetical. Open WebUI already posts long blocking jobs to that same executor (`retrieval/vector/dbs/pinecone.py:323` batch upserts, `retrieval/loaders/youtube.py:156` transcript loads). With 28 such jobs holding the pool, a single DNS lookup waits for them to finish:
| | one DNS lookup |
|---|---|
| ThreadedResolver | 1989.7ms |
| AsyncResolver | 58.9ms |
A Pinecone bulk upsert currently stalls name resolution for every other user on the instance. After this change it cannot.
At low concurrency on a real network the two are equivalent, as expected: 8 concurrent lookups against disjoint cold hostname sets landed within noise of each other in both directions.
## Behaviour verification
Checked against Open WebUI's own code, not in isolation:
- c-ares reads the system hosts file. Verified against a machine whose hosts file maps `adobe.io` to `0.0.0.0`, an address real DNS never returns for that name: c-ares returned `0.0.0.0`. `host.docker.internal`, compose `extra_hosts` and Kubernetes `hostAliases` keep working.
- `_SSRFSafeResolver` subclasses `aiohttp.resolver.DefaultResolver`, so this change swaps its base class from ThreadedResolver to AsyncResolver at runtime. It still resolves public hosts, still returns entries with the `host`/`port` keys the SSRF check reads, and still raises on a private address: resolving `localhost` raised `ValueError: The URL you provided is invalid.`
- A real fetch through `get_ssrf_safe_session()` returned 200.
- NXDOMAIN still surfaces as `aiohttp.ClientError` (`ClientConnectorDNSError`), not a c-ares specific exception, so existing error handling is unaffected.
Known limit: c-ares reads `/etc/resolv.conf` and the hosts file but not the rest of `nsswitch.conf`. Names served only by an NSS module, such as `.local` via avahi/mDNS, NIS/LDAP backends or Windows NBNS, will resolve differently or not at all. On a multi-homed test machine the local hostname returned two addresses through the system resolver and one through c-ares. Deployments pointing Open WebUI at an mDNS or NetBIOS hostname are the group affected. Resolver failures also arrive as plain `OSError` rather than `socket.gaierror`, which no code in this repo catches today.
Every page pulled in by web search and web RAG is parsed with BeautifulSoup's `html.parser`, a pure-Python parser. It is the slowest option bs4 offers, and it is being handed 300 KiB to 1.5 MiB documents, several per query. `SafeWebBaseLoader` inherits `default_parser = "html.parser"` from langchain's `WebBaseLoader` and never overrides it, so this is an upstream default carried by accident, not a decision anyone made for Open WebUI.
`default_parser` is the single chokepoint for both the sync `_scrape()` path and the async `ascrape_all()` path, so one `setdefault` covers everything and an explicit caller override still wins.
lxml is already in the tree as a transitive hard dependency of ddgs, python-pptx and unstructured, so nothing new enters the image and `uv.lock` already resolves it at 6.1.1. The pin makes it explicit and closes a latent failure: bs4's `"xml"` feature, already used for `.xml` URLs in `_unpack_fetch_results()`, requires lxml and would raise `FeatureNotFound` the day that transitive dependency moves.
## Benchmarks
37 real pages, 13.8 MiB of HTML, median of 5 runs each. The timed operation is `BeautifulSoup(html, parser)` plus `get_text()` plus `extract_metadata()`, which is exactly what the loader does per page. bs4 4.14.3, lxml 6.1.1, CPython 3.12.
| | html.parser | lxml | |
|---|---|---|---|
| 37 pages, 13.8 MiB total | 1611.0ms | 1151.3ms | 1.4x faster, 460ms saved |
Largest pages:
| page | size | html.parser | lxml | speedup |
|---|---|---|---|---|
| pypi.org/project/aiohttp/ | 1259 KiB | 243.75ms | 180.51ms | 1.4x |
| gnu.org/software/bash/manual/bash.html | 1017 KiB | 257.97ms | 178.99ms | 1.4x |
| rfc-editor.org/rfc/rfc9110.html | 1157 KiB | 205.94ms | 154.87ms | 1.3x |
| docs.aiohttp.org/en/stable/client_reference.html | 403 KiB | 108.62ms | 84.93ms | 1.3x |
| ollama.com/library | 779 KiB | 117.55ms | 73.64ms | 1.6x |
| theregister.com | 1052 KiB | 88.32ms | 60.12ms | 1.5x |
| kubernetes.io/docs/concepts/services-networking/service/ | 563 KiB | 72.18ms | 43.43ms | 1.7x |
| docs.python.org/3/library/socket.html | 301 KiB | 71.88ms | 49.04ms | 1.5x |
Ranges from 1.1x to 1.7x, and the win grows with page size. A ten result web search sheds roughly 125ms of parsing. Because the async path builds its soups inline in `_unpack_fetch_results()`, that is 125ms the event loop spends parsing HTML instead of serving other users' streams. Pages under about 10 KiB are marginally slower under lxml due to fixed setup cost, which is worth nothing either way.
## Output verification
The risk in changing parser is silently different extracted text, so that was measured rather than assumed. Across all 37 real pages:
- **Zero characters of text were lost.** Every diff opcode against html.parser output was an insertion. Not one page dropped content under lxml.
- 659 characters were added, all on one page (docs.docker.com), where an inline Alpine.js `@click` handler containing a regex confuses libxml2's attribute handling and leaks a 73-character JS fragment into the text nine times. That is 659 characters of script noise in 27,206 characters of extracted text, with no content affected.
- Metadata (`title`, `description`, `language`) was identical on 35 of 37 pages. The two exceptions are 141-byte Wikipedia bot-block stubs with no `<html>` element, where lxml's fragment auto-wrapping adds `language: "No language found."`. Both parsers extract the same text from them.
Large documents were checked separately because libxml2 carries internal size caps. A 12 MiB single text node, 12 MiB spread across 400k nodes, a 3 MiB attribute value and 50k sibling elements with a trailing marker all produced byte-identical text under both parsers, with no truncation.
Malformed markup was checked too. lxml and html.parser diverge on unterminated comments, bare CDATA and duplicated `<html>` elements, all cases where both parsers are guessing and neither is correct. None of those shapes appeared in the 37 page corpus.
`backend/open_webui/env.py:184` also uses `html.parser`, on the local CHANGELOG at import time. That is trivial input on a startup path and is deliberately left alone.
The migration to joserfc completed the job but left the old dependency pinned. `python-jose` now has zero imports anywhere in the backend: the only `jose` references left are `joserfc` in `utils/oauth.py`, and a repo-wide search for `from jose`, `import jose` or `python_jose` returns nothing outside the three pin files.
Removing it also removes `ecdsa` and `rsa` from the image, which were pulled in only by python-jose. `uv lock` confirms that: it drops exactly those three packages and nothing else, because google-auth 2.55 depends on cryptography and pyasn1-modules rather than rsa. That is worth having beyond the size saving, since `ecdsa` ships a documented Minerva-style timing side-channel in its P-256 signing path that upstream has declined to fix, so keeping it in the image means shipping a flagged crypto library that nothing calls.
Verified by blocking the `jose` module at import time and importing the backend anyway:
```
PASS import open_webui.utils.auth
PASS import open_webui.utils.oauth
PASS import open_webui.main
jose in sys.modules: False
PASS create_token/decode_token round trip
```
One user-visible consequence worth stating: Tools and Functions run in the same interpreter, so a third-party plugin that imports `jose` directly stops working after this. Nothing in Open WebUI itself does, and PyJWT remains a dependency, but a plugin relying on a library the application never declared for that purpose is the only thing this can break.
`uv.lock` was edited surgically rather than regenerated, to avoid the unrelated whole-file churn a newer uv version introduces. The result was diffed against real `uv lock` output and matches it exactly apart from that version's cosmetic fields.
Uvicorn's `--ws auto` selected its `websockets_impl` protocol on 0.41.0, which is built on `websockets.legacy`. That module raises `AssertionError` in `_drain_helper` during keepalive pings and kills the websocket connection. Each crash runs the Socket.IO `disconnect` handler and drops the session from `SESSION_POOL`, so every subsequent server-to-browser call fails. The most visible symptom is the Pyodide code execution tool, which reaches the browser through `sio.call('events', ...)` and returns `{"stderr": "Client session disconnected."}` on every run.
Uvicorn 0.50.0 changed `--ws auto` to select the sans-io implementation whenever websockets is installed, and deprecated the legacy one. Bumping the pin therefore fixes this on every launch path at once, without adding a `--ws` flag to the startup scripts. Doing nothing is not stable either: websockets is unpinned apart from uvicorn's own `>=13.0` floor, and `websockets.legacy` is removed outright in websockets 17, which turns the current AssertionError into an ImportError on a fresh install.
Bumping to 0.51.0 rather than the minimum 0.50.0 also picks up the sans-io keepalive pings added in 0.44.0, so raw websocket endpoints keep the idle-timeout behaviour they have today behind a reverse proxy. Uvicorn 0.51.0 drops colorama from its `standard` extra and raises the httptools floor to 0.8.0, which the lockfile already satisfies.
Verified on the bumped pin: the backend boots, `/health` returns 200, `--ws auto` resolves to `WebSocketsSansIOProtocol`, a Socket.IO client completes a websocket handshake against the running app, and a bidirectional `sio.call` round trip succeeds. The unit test suite reports an identical 2273 passed / 7 failed on 0.41.0 and 0.51.0, with the 7 failures unrelated to uvicorn.
Fixes#27550
* perf: cut per-instance CPU cost of shared socket.io Redis pub/sub channel
Profiling a multi-instance deployment (py-spy --gil) showed ~44% of worker
CPU in the socket.io pub/sub listener. Two causes, two fixes:
- Add hiredis so redis-py parses the RESP protocol in C instead of pure
Python (redis/_parsers/resp3.py alone accounted for ~28% of GIL samples;
redis-py auto-selects the hiredis parser when importable).
- Subclass AsyncRedisManager to drop emits whose target room has no local
participants before upstream _handle_emit re-encodes the full packet.
Every instance receives every emit published on the shared channel, so
with N instances all but the hosting one were paying full packet
re-serialization per message just to deliver it to nobody. Broadcasts
(room=None) are unaffected.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9CQ9qnp3sZGYQQwztsCJT
* fix: restrict pub/sub emit early-out to string rooms
Adversarial review against python-socketio 5.16.2 found one divergence
from upstream: for a degenerate empty-sequence room (emit to room=[]) on
an instance whose namespace has no local clients, the filter's
get_participants probe raises IndexError from room[0] where upstream
returns silently at the namespace guard and still publishes to Redis.
Open WebUI only ever emits to scalar string rooms or room=None, so the
case is unreachable today; guard on isinstance(room, str) anyway so any
non-string room shape passes through to upstream behavior unchanged.
Every open-webui emit uses a string room, so the fast path still covers
all real traffic.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9CQ9qnp3sZGYQQwztsCJT
* Update requirements.txt
* Update pyproject.toml
* Update requirements-min.txt
---------
Co-authored-by: Claude <noreply@anthropic.com>
* chore: bump Python backend dependencies, drop unused peewee
Minor/patch + reviewed major bumps across requirements.txt,
requirements-min.txt, pyproject.toml and uv.lock; playwright image bumped in
docker-compose.playwright.yaml. peewee/peewee-migrate removed (zero imports).
Security-relevant: cryptography 46->48, authlib 1.6.10->1.7.2, PyJWT 2.11->2.13,
requests 2.33.1->2.34.2, RestrictedPython 8.1->8.2, pillow 12.1.1->12.2.0.
Reviewed majors: redis 7->8, pymilvus ->2.6.14, azure-search-documents 11->12,
chardet 5->7, unstructured 0.18->0.22, pycrdt 0.12->0.13.
Testing:
- Resolution: `uv lock` resolves the full bumped set with no conflicts; uv.lock
regenerated to match (peewee dropped, every pin including
azure-search-documents==12.0.0 resolves).
- Per-dependency contract tests (external tests repo, unit/deps/): 105 files,
2205 passed / 6 skipped, ruff-clean. One file per dependency pins the symbols,
signatures and behaviour the backend actually uses, so an API removal/rename in
a bumped version fails loudly instead of at runtime. Offline/deterministic.
- End-to-end embed->retrieve test driving transformers + sentence-transformers +
chromadb together through Open WebUI's real RAG path (cached model, in-memory
chroma, semantic retrieval asserted).
- Install/startup/health resolution gate added to the dep-bump workflow and the
integration suite (uv/pip resolve + uvicorn /health + Playwright dev visibility).
- Bugs surfaced while testing each got an isolated fix branch + regression test:
Mistral OCR aiohttp FilePayload (#25779), chroma has_collection (#25780),
aiocache per-user model-cache key (security), otel semconv deprecation,
pydub/audioop <3.13 note.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Bump python-multipart 0.0.22 -> 0.0.27 (CVE-2026-42561, CVE-2026-40347)
0.0.22 is affected by two DoS CVEs in the multipart parser that
Starlette/FastAPI run for every multipart/form-data request, so any
authenticated user hitting an upload endpoint can trigger them:
- CVE-2026-42561: unbounded part-header count/size -> CPU exhaustion (fixed 0.0.27)
- CVE-2026-40347: large multipart preamble/epilogue DoS (fixed 0.0.26)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat: add support for Valkey vector database
Signed-off-by: Riley Des <riley.desserre@improving.com>
* feat: add CLIENT SETNAME to Valkey vector store connections
Set client_name on GlideClientConfiguration for both the main client
and batch client so connections are identifiable in CLIENT LIST,
monitoring dashboards, and CloudWatch metrics.
Signed-off-by: Riley Des <riley.desserre@improving.com>
---------
Signed-off-by: Riley Des <riley.desserre@improving.com>
* 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
