The gemini 3.6 flash batch rates landed at half the standard rates in
94a29e0708, so the hardcoded standard-rate expectation started failing
on staging and red-lit misc / Run tests on every PR.
Resync registry misses with single-row DB fetches (guardrail by unique
name, agent by unique id or name, model by name then id) instead of
full-table loads, and bound them with a global budget of 20 resyncs per
5s window per registry that fails closed without negative-caching the
key.
Access group create/update now trust the reconcile outcome snapshot
captured under the reload lock instead of a post-lock router read, so a
concurrent reconcile can no longer surface a false degraded-serving 500.
Router.upsert_deployment restores the previously served deployment when
the replacement add fails under ignore_invalid_deployments, so a bad
update no longer silently drops a healthy deployment from serving.
The strict-denial pin only asserted a 401, so any unrelated 401 (a bad key, a deleted key) would have kept it green while tag routing silently broke. The harness now keeps the 401 response body, the way it already does for 429s, and the pin asserts the tag-routing denial message.
* fix(ui): highlight the first member search match so Enter picks it
Moving the modal off antd Form swapped antd Select for Base UI Combobox, which
highlights nothing until an arrow key moves the cursor. Typing an email and
pressing Enter therefore selected no one, and the form submitted anyway, so
/team/member_add went out with member.user_email undefined.
autoHighlight="always" restores the behaviour antd had. The Enter key still
does not submit the form, which the existing test continues to cover.
* fix(ui): keep the always-highlight prop past the type check
Combobox.Root re-declares autoHighlight as boolean while the AriaCombobox it
wraps types it as boolean | "always". Only "always" highlights a list this
component filters server-side, since the plain flag highlights on Base UI's own
filtering pass, which filter={null} turns off.
Verified with next build, the same type check that failed in CI.
The earlier sweep only caught the conformance suite in tests/llm_translation.
Groq retired llama-3.1-8b-instant alongside llama-3.3-70b-versatile, and four
tests under tests/local_testing still call them for real, so litellm_router_testing
and both local_testing shards 404 with model_not_found.
Only the sites that leave the process move. The chunk fixtures in
test_stream_chunk_builder, and the cost and routing tests that never open a
socket, keep the old ids because the string is data there, not a request.
* feat(otel): attribute Prisma database spans to PostgreSQL instead of localhost
Prisma reaches PostgreSQL through a query engine on loopback, so transport
instrumentation attributes database waits to localhost and operators cannot
tell the work is PostgreSQL or correlate it with the database's own metrics.
Datastore service spans now carry db.system.name, db.system, db.operation.name
and, for PostgreSQL, server.address, server.port and db.namespace derived from
DATABASE_URL, and are emitted as CLIENT spans. Only host, port, database and
schema are read, so no credential reaches an exporter. Endpoint attributes are
omitted when a read replica is configured, because routing is decided per Prisma
call underneath the span.
* fix(otel): reject a mis-split DSN authority and name socket-only databases
An unencoded '/' in the password truncates the URL authority, so urlparse
reports the username as the host and the password tail as the database, which
put credential material in db.namespace. Postgres drivers reject that DSN
outright, so the only safe reading is no endpoint at all.
A hostless 'postgresql:///litellm' is a valid local-socket DSN that Prisma
accepts, and it now yields db.namespace with no server address rather than
nothing. The default schema is matched case-insensitively, since an unquoted
PostgreSQL identifier folds and one deployment must yield one namespace.
* fix(otel): keep a non-default schema in db.namespace
Prisma quotes the schema name, so a DSN with ?schema=PUBLIC provisions a
second schema alongside public rather than reusing it. Observed on a live
proxy: the PUBLIC schema came up with its own 70 tables next to public's 78,
and a key created under one was not visible under the other.
Case-folding the two into a single namespace therefore reported two different
schemas as the same database, which is the misattribution this feature exists
to remove. Match the default literally.
* fix(otel): reject any DSN whose userinfo fell outside the authority
An unencoded '#' or '?' in the password sends the tail to the fragment or
query, leaving the path empty, so the marker check on the database segment
never fired and urlparse's hostname (the database username) was exported as
server.address.
The stranded userinfo '@' is the general tell for every mis-split, so guard on
that instead of enumerating the characters that cause it.
* fix(otel): allow an at-sign inside a well-formed DSN query
The previous guard rejected any DSN whose userinfo at-sign fell outside the
netloc, which also caught libpq parameters that legitimately carry one, so
?application_name=svc@prod and ?user=admin@company.com lost their endpoint
attributes.
Discriminate instead: a PostgreSQL DSN never has a fragment, its database name
cannot hold an unencoded at-sign or slash, and an at-sign in the query is only
suspicious when the query did not parse as parameters.
* fix(otel): resolve the database endpoint per span instead of once per process
The endpoint was cached for the process lifetime on the premise that
DATABASE_URL is deployment-static. It is not. The RDS IAM refresh rebuilds the
URL from DATABASE_HOST/PORT/NAME/SCHEMA on every rotation, the reconnect path
re-reads DATABASE_URL, and the DB-backed environment_variables config overlay
sets arbitrary keys post-startup with no blocklist covering DATABASE_*. A
process that had genuinely failed over kept exporting the old server.
Read the environment per span, which is also what Prisma connects with, so the
span can no longer name a different server than the one serving the query;
get_secret_str consulted a secret manager first and could diverge from it. Only
the parse is memoized, keyed on the URL.
* fix(otel): reject a question-mark mis-split whose tail parses as parameters
A '?' in a password strands the rest of the authority in the query, and that
tail can still parse as key=value, so testing only for an unparseable query let
the login through as server.address. One spelling hijacked the host= parameter
and put the password suffix there directly.
A legitimate at-sign in a query always follows a database path, and a
'?'-mis-split never leaves one, so require both.
* refactor(otel): drop the DSN parse cache that retained rotated credentials
The cache was keyed on the full DATABASE_URL, so up to eight complete DSNs,
each carrying a password or a retired IAM token, stayed referenced for the
process lifetime and outlived the rotation that replaced them. Nothing reached a
span, but a heap dump or crash report would have surfaced them.
Parsing costs about four microseconds against a span emission that costs orders
of magnitude more, so the cache bought nothing worth that.
* fix(otel): avoid a set construction the tightened LIT002 budget rejects
* fix(otel): refuse an ambiguous DSN authority instead of guessing at it
A password holding both an unencoded slash and a query-like tail defeated all
three shape checks: the slash left a clean path carrying the password
remainder, the query still parsed as parameters, and no fragment survived. The
login went out as server.address, the password's leading digits as server.port
and the rest as db.namespace.
A DSN whose at-sign sits in a query parameter is indistinguishable from that
mis-split by any property of the parse; both leave no userinfo, a host, a port
and a path. Guessing wrong publishes a credential fragment, so the ambiguity
now resolves to refusing the endpoint. Such a DSN loses server.address and
db.namespace and keeps the rest of the span; percent-encoding the at-sign
restores them.
Also honour port= alongside host=, which libpq pairs and this read ignored.
* docs(otel): fix a spliced sentence and a stale cache claim in db_endpoint
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* refactor(ui): extract the create-key payload builder out of create_key_button
handleCreate built the POST /key/generate body inline by mutating the object antd
handed it, across roughly 170 lines mixing form values, twelve pieces of React
state and five sources that all fold into object_permission. Nothing could assert
that shape without rendering the whole modal.
The construction now lives in createKeyPayload.ts as a pure function returning a
tagged union, with 46 unit tests that run in single digit milliseconds and pin
whole payloads with toStrictEqual, following the createServerPayload pattern the
dashboard CLAUDE.md documents.
Behaviour is unchanged. The extraction was checked against the pre-extraction
handler with a differential harness over 50 input combinations, comparing the
built object, its key order and its serialised bytes, and eight mutations of the
new module were each confirmed to fail the committed tests.
* refactor(ui): align the key payload builder to the programme's entry-condition spec
Folds the duplicate-alias guard and the endpoint choice into the builder, so
KeyPayloadResult now carries three variants and handleCreate holds no payload
decision of its own. The two failure branches keep their original position
around toast.info and setIsModalVisible, and the builder checks the alias before
the agent selection, so the observable order is unchanged.
Pins the serialised wire shape as well as the object. The closed form registers
team_id at null through initialValue while organization_id has no initialValue
and stays undefined, so an untouched create sends seven of its eight keys.
Opening Optional Settings takes the object to 23 keys and the wire to nine. Both
directions of the definedness contract are now covered: undefined must not
become null, and null must not be dropped.
The earlier fixture fed a team_id the closed form cannot produce and pinned a
six-key wire as a result.
* refactor(ui): fold the service-account metadata write into its only caller
Removes assignServiceAccountId as a separate helper so there is no mutating
function available for reuse, which was the substance of the review finding. The
write now sits two lines below the JSON.parse that produced the value, so it is
visibly local and cannot reach a caller-owned object.
The write itself stays. Metadata has no validation rules, so a user can submit a
JSON body that parses to a primitive or an array. On a primitive the property
write raises a TypeError and the existing catch surfaces an error toast, and on
an array it leaves the array intact through JSON.stringify. A spread coerces
both to plain objects instead, silently creating a key from input the form
rejects today. Two tests pin those cases and go red against the spread.
gemini/imagen-4.0-generate-001 now 404s with "no longer available, please
update your code to use models/gemini-3.1-flash-image", and
groq/llama-3.3-70b-versatile reached its deprecation_date of 2026-08-16, which
is the day the llm_translation and image_gen jobs went red.
Only the two live call sites move. The remaining references to the old ids sit
in offline transformation tests, where the string is just a routing key and no
request leaves the process.
A file uploaded through Router.acreate_file lands in the account of the
deployment that stored it, so a cross-group fallback silently stores the
file with the wrong provider and every later batch or fine-tuning call
against the returned id permanently fails. Extend the provider-scoped
fallback pin that already covers input_file_id and training_file to file
creation, so the original provider error surfaces instead.
litellm.completion(thinking=True) crashed pre-network in is_thinking_enabled
with a retryable APIConnectionError ('bool' object has no attribute 'get'),
so the router burned retries on a deterministic failure and proxy clients got
a traceback instead of a usable response.
validate_and_fix_thinking_param now coerces thinking=True to the enabled dict
with the default medium budget and drops thinking=False, and the remaining
dict-assuming thinking accessors (base config, bedrock converse, deepseek)
guard with isinstance so raw bools can never crash a transform.
* test(ui): repoint the e2e locators at the post-antd form controls
Nine Playwright specs went red after the tremor and antd removals, none of
them because the product broke. The specs selected on markup those libraries
owned: tremor's TextInput stamped data-testid="base-input", the antd toast
facade rendered .ant-notification, and the team member modal's email field was
an .ant-select. Removing the libraries deleted those hooks silently.
Repoint each onto a user-facing locator that survives the next migration:
getByLabel for the key name, the MCP tool argument and the two cache pricing
fields, getByRole("combobox") for the team member email search, and the toast
container for the add-model success message.
The pricing fields needed a source change to be reachable at all. antd's
Form.Item used to assign the field name as the input id and tie the label to
it; the react-hook-form rewrite renders FieldLabel with no htmlFor and lets
FormField generate an opaque control id, so both cache inputs lost their
accessible name and could only be told apart by placeholder, which they share.
Pin the id back to the field name and point the label at it.
* fix(ui): let FormField own the pricing field label instead of hand-rendering one
The previous commit gave the cache cost inputs an accessible name by rendering
a FieldLabel with htmlFor next to the FormField. FieldLabel forwards Label's
props, and Label only accepts children, so next build failed type checking.
FormField already renders a label wired to the control id it generates, and
FormField.test.tsx covers that association, so passing label through is both
type-safe and less markup. The read-only branch keeps its plain FieldLabel,
which has no control to point at.
* refactor(ui): migrate the key edit form off Ant Design onto react-hook-form
The key editor was the last antd Form in the key flow. antd's store decided
the saved payload implicitly: onFinish reported whatever mounted Form.Items
happened to be registered, so a control could stop feeding the request
without anything failing.
The form now runs on react-hook-form with the dashboard's own field
primitives, and the payload is projected explicitly in keyEditFormValues so
every saved key is written out by name. That reproduces the old payload
exactly, including the keys antd sent holding undefined and the two
role-gated keys it dropped entirely when the field was not mounted.
The existing suite is kept as the contract and passes unedited apart from
the selects, whose queries moved from Ant Design class selectors to roles
and labels. MultiSelect now forwards a per-option disabled flag, which the
Models select needs to keep greying out individual models once the
all-proxy-models sentinel is picked.
* fix(ui): keep the key edit prompts control gated behind premium
The antd control carried disabled={!premiumUser} and a tooltip saying
prompts by key are premium. The port kept the premium placeholder but
dropped the gate, so a non-premium admin could type a prompt and have the
whole save rejected by the endpoint.
TagsInput had no disabled prop at all, which is why the gate could not
survive the port; it now takes one and passes it to the combobox. An audit
of every disabled expression against the antd original shows this was the
only gate lost.
Also pins the two payload keys that are assembled in the submit handler
from React state rather than bound fields, budget_fallbacks and
tag_rpm_limit. A field-driven suite cannot see them, and tag_rpm_limit's
only previous appearance was an empty map that reads the same whether the
assignment works or is deleted. The cases were written against the antd
implementation by another lane and are added unchanged.
The create key suite mocked antd wholesale and swapped in a fake form
store whose onFinish spread every value it had ever seen, which is the
inverse of what rc-field-form does: real antd reports only the mounted
registered fields, so anything inside a closed collapsible is absent
from the body rather than present and null. Fourteen children were
stubbed to () => null on top of that, so the suite passed no matter
what the form would actually submit and could not protect a port off
antd.
The integration test now renders the real tree and stubs only the
network boundary, and a submit payload contract block pins the exact
body with toStrictEqual so null, undefined and absent stay distinct.
It covers all sections closed, Optional Settings alone, all open, and
each nested section opened by itself, plus the collapse and re-expand
path that antd's store survives and a shouldUnregister port would not.
The genuinely unit-level logic, fetchTeamModels and fetchUserModels,
moves to a plain unit test that mocks modelAvailableCall alone.
getOpenAPISchema stays reachable through a stubbed global fetch rather
than the networking module mock, because networking imports jsonFields
from check_openapi_schema and that module imports getOpenAPISchema back,
so importOriginal binds the real export through the cycle.
`mcp_server_edit` builds its update payload from `validateFields()`, so the
request body is whatever antd had mounted at the moment Save was pressed.
Nothing asserted that body, and the existing suite has no `toStrictEqual`
anywhere, so a key appearing, disappearing, or arriving as null instead of
undefined was invisible.
This adds a characterization net that drives the real component tree and
captures the exact object handed to `updateMCPServer`. Every case is pinned
with `toStrictEqual`, which is the only matcher that separates the three
states a key can be in: absent, present-as-undefined, and present-with-value.
That distinction is the whole point here, since a gated-off field and a
mounted-but-empty field currently differ, and `entra_obo` versus plain
`token_exchange` is exactly that difference in the wild.
19 cases cover the transports, all ten auth types, the credential subsets each
one contributes, and the tool allowlist gate. Values were captured from the
component as it behaves today rather than written from reading the source, so
the file records current behaviour instead of intended behaviour
A six-mutant battery run against the payload builder kills 6/6, each run
executing all 19 tests: leaking an audit field, deleting a binding, removing a
gate, turning a flat key into a lodash path, swapping undefined for null, and
swapping an empty-array default for null
Two explicit assignments in the payload turn out to be dead. `alias` and
`mcp_access_groups` are both already supplied by the `...restValues` spread
above them, so removing either line changes nothing. Left in place here and
noted for the payload extraction that follows
The pass-through method contract asserted that every route accepts GET, POST,
PUT, DELETE and PATCH. That holds for routes fronting a REST API, where
narrowing the set turns a request the upstream would have served into a 405,
but it does not hold for Amazon Comprehend Medical: it speaks AWS JSON 1.1,
which is POST-only, and its handler signs a POST upstream and requires a JSON
object body. Registering the other four verbs there would only produce routes
that always fail body parsing, so the route is POST-only on purpose and the
blanket assertion is the part that is wrong.
Carry a narrow exemption list instead of dropping the check. Exempted routes
are still pinned to an exact method set, so widening or narrowing one fails,
and a companion test asserts every exempted path is still registered so the
list cannot outlive the routes it covers.
Two suites broke because they stood in for production objects with stand-ins
that no longer answer the same way.
The redaction test faked a ResponsesAPIResponse and then reassigned
builtins.isinstance so the fake would pass the type check. Redaction now gates
on a tuple of accepted types, and the patched isinstance only recognised the
bare class, so the fake fell through to the generic branch and the assertions
ran against a plain dict. Building a real ResponsesAPIResponse drops the
builtins patch entirely and exercises the same type gate production takes.
The batch rate limiter tests constructed _PROXY_BatchRateLimiter with
parallel_request_limiter=None even though the parameter is not optional. That
stayed harmless until the output-token estimate started reading the limiter,
which turned it into an AttributeError. Inject the limiter the proxy injects,
sharing one InternalUsageCache the way _add_proxy_hooks does.
The aggregated daily-activity queries build their WHERE clause as raw SQL, and
an empty entity list rendered as `"team_id" IN ()`, which Postgres rejects with
`syntax error at or near ")"`. Callers reach that state normally: a caller
without admin view and no explicit team_ids has its scope resolved to the teams
it belongs to, so anyone belonging to no teams, an org admin for instance, sent
an empty list and got a 500 back from /team/daily/activity/aggregated.
The paginated endpoint hands the same empty list to Prisma, which renders
`in: []` and matches nothing, so it kept returning 200 with an empty result set.
Emit FALSE for the empty case so the raw-SQL path lands on the same answer,
mirroring what the api_key filter a few lines below already does.
The fix covers both aggregated queries at once because they share one WHERE
clause builder.
* refactor(ui): codemod the antd Tooltips outside form files onto the shadcn atom
The atoms/Tooltip wrapper gains an optional side and renders its children without a popup when there is no content, which covers antd's placement prop and its title={undefined} escape hatch. A TypeScript-AST codemod then rewrites every antd Tooltip that does not sit in a file still using antd Form, since those files get their primitives swapped as part of the react-hook-form migration.
* refactor(ui): fold the atoms Tooltip into the ui tooltip primitive as SimpleTooltip
Threading a TypeVar through the helper makes the fallback attribute write
unprovable, and routing it through setattr to quiet that only trades one
diagnostic for a bugbear violation. Leave the seam as it was
team/LoggingSettings.tsx carried 34 hardcoded palette classes and
common_components/PremiumLoggingSettings.tsx another 9, so both render
light-only regardless of theme. Map the neutrals onto foreground,
muted-foreground, muted and border, the red affordances onto destructive,
and swap the hand-rolled chips for the shadcn Badge primitive.
The three event-type options carried decorative green, red and blue dots.
The design system has no success or info token, so the dots are dropped
and the option labels, which already say "Success Only", "Failure Only"
and "Success & Failure", carry the meaning on their own.
This is groundwork, not a visible change: nothing in the dashboard ever
applies the .dark class today, so the dark palette is unreachable. The
files no longer hardcode colour and will follow the theme once one exists.
Typing the managed-responses call kwargs as dict[str, object] forced an
isinstance filter on the popped model and previous_response_id, which turned a
malformed client value from a loud downstream failure into a silent fallback to
the connection's model. Keep those two seams and the metadata mapping as they
were so the frame still fails the way it always did
true_passthrough and oauth_delegate forward the caller's own bearer and mint
nothing, so their sessions consume no discovered OAuth endpoint. The discovery
completeness gate still failed them closed with a 503 raised before the upstream
was ever contacted, which the tools/list fan-out swallowed into HTTP 200 with an
empty tool list. Any upstream that publishes no RFC 9728 metadata, including
every OpenAPI-backed internal API, was permanently unusable.
A failed discovery is now fatal only to servers whose OAuth flow the gateway
runs itself. Discovery still runs for the forwarding modes, so /authorize,
/token and /register keep resolving their endpoints, and each keeps its own 400
when they are missing.
MCPServer.is_client_forwarded_token now owns the mode pair that five call sites
spelled inline, one of which had already named it is_client_forwarded_mode.
enterprise/ and litellm-proxy-extras/ both changed between main and staging, so each gets a PATCH bump. The 1.98.0 line already graduated with v1.98.0-rc.1, so this promotion opens the 1.99.0 line and litellm takes its MINOR bump.
uv.lock re-resolved against the three new versions; the exclude-newer timestamp moves because the lock uses a rolling P3D window
Replace implicit and explicit Any with real types across the highest-density
reportAny/reportExplicitAny files: module-private TypedDicts for dict payloads,
Protocols for duck-typed collaborators, and existing litellm/types models where
they already describe the shape
No new cast(), no # type: ignore, no # pyright: ignore, no # noqa, and no
new suppressions. Diagnostics that could not be resolved without one were left in
place rather than hidden
* fix(logging): close three secret-leak paths in verbose logging
The AWS credential pattern was the only key-name matcher in secret_redaction
that skipped optional quotes, so quoted dict-repr values leaked. Fold the three
AWS key names into the shared key-name alternation instead.
SecretRedactionFilter only scrubs str record attributes, so a dict/list/set
passed through extra={...} reached the formatter unredacted. Redact at the
formatter boundary so no value shape can bypass it.
log_raw_request_response wrote the request curl command to metadata["raw_request"]
unredacted, returned an unmasked raw_request_api_base, and fell back to dumping
model_call_details whenever api_base was empty.
* Update litellm/_logging.py
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* fix(logging): redact JSON log values without breaking the document
JsonFormatter redacted the serialized JSON, so a secret-named member
collapsed from '"api_key": "sk-..."' to a bare REDACTED token and the
line stopped parsing as JSON.
Redact before serialization instead: safe_dumps takes an optional
value_transform hook (default None, so all other callers are unchanged)
and redact_structured_value collapses only the value, leaving the key
and surrounding structure intact.
JsonFormatter now emits "api_key": "REDACTED" where the formatter unit
test expected the already-masked "sk**********". That test bypasses
SecretRedactionFilter, which in production collapses the pair before any
formatter runs, so the assertion is updated to match real behavior.
---------
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* fix(vector_stores): stop leaking stored credentials in direct search debug logs
Direct vector store providers (RESP datastores like Valkey) have no HTTP
request to echo, so both search handlers called `logging_obj.pre_call` with
no `api_base`. The logging helper treats an empty `api_base` as "nothing to
render" and falls back to `str(self.model_call_details)`, which carries the
resolved `litellm_params`: the stored `valkey_password` and the embedding
config's `api_key` among them.
The stdout logger's regex redaction hid this, but `pre_call` also writes the
same string to `litellm_params["metadata"]["raw_request"]`, which ships
unredacted to every logging callback (Langfuse, OTel, etc.).
Pass a synthetic `<provider>://<vector_store_id>` endpoint plus an explicit
`request_str` so the debug output describes the call instead of dumping call
details, and fold the duplicated sync/async blocks into one helper so the
sanitized descriptor cannot drift between them.
* fix(vector_stores): type direct search query as Sequence[str]
The new helper's list[str] annotation pushed LIT001 over its
type-discipline ceiling. Sequence is the read-only shape the helper
actually needs, and list[str] still satisfies it at both call sites.
