/v1/messages and other litellm_metadata endpoints store proxy metadata,
including x-litellm-tags header tags, under litellm_metadata instead of
metadata. The pre-routing hook read request tags with a hardcoded
metadata bucket, so it never saw the tags that selected the marker and
cleared the consumed-tags stamp, and tag filtering then 401'd the routed
tier. Resolve the bucket from the request kwargs instead, matching how
the stamp write and the tag-filter read already resolve it.
The outer wrap_sse_stream_with_keepalive_pings layer duplicated the
keepalive engine that PR #34423 already runs inside async_data_generator
for chat completions and responses streams, and it kept pinging
deployments whose operator set keepalive_seconds: 0 as a hard disable.
sse_keepalive_ping_interval_seconds is now the global fallback inside
_resolve_keepalive_seconds, so deployment and request values keep
precedence, an explicit 0 still disables, the [1, 300]s clamp applies,
and router-less proxies arm the wrap when the global default is set.
The Rust messages bridge logs a parsed Anthropic response without an
httpx_response, so the fallback transform dropped the request speed and
billed fast-mode calls at the standard rate. Thread optional_params
speed into transform_parsed_response and add a regression test for the
parsed-response branch.
A stream can terminate with a response.failed event that still reports
consumed tokens; those were rebuilt as None and logged at zero spend.
Parse response.failed alongside completed and incomplete, matching the
buffered path, which prices any terminal response that reports usage.
The Anthropic batch create never persisted the creating key's hashed token or its
request tags on the managed object, so when CheckBatchCost billed the batch hours
later there was nothing to attribute it to. Key spend, key budgets and tag spend
never moved for batch usage.
Persist both from the create, the way the Vertex passthrough already does, and
register the batch only from the collection route. An id-scoped route cannot
rebuild the unified object id, because it embeds the model and the model comes
from the create's request body, so it could only claim a row it did not create or
fail the model_object_id unique constraint.
The shared metadata helpers, the route predicate and the registration-result
logging now live in batch_attribution instead of being copied per provider. The
Anthropic write previously logged success unconditionally, before the
fire-and-forget task had run.
Resolves LIT-5288
LiteLLMCompletionStreamingIterator overrides __init__ without calling
super().__init__(), so completed_response was only set once the stream
reached RESPONSE_COMPLETED. On a mid-stream provider error the router's
_extract_partial_responses_usage read source_iterator.completed_response
during fallback recovery and raised AttributeError, masking the real
provider error (e.g. Anthropic 529) and bypassing configured retries and
fallbacks. Initialize the attribute to None so recovery degrades to no
partial usage instead of crashing.
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Streams that terminate with response.incomplete (e.g. max_output_tokens
reached) carry real usage in the terminal event but were rebuilt as None
and logged at zero spend, letting callers bypass budget enforcement.
Parse response.incomplete alongside response.completed when
reconstructing the streamed response.
Addresses review feedback: the ResponseCompletedEvent SSE parsing now lives
in OpenAIResponsesAPIConfig next to the other Responses stream event handling,
and the proxy logging handler calls it. Adds coverage for streams that end
without a response.completed event.
Register remote MCP servers without awaiting OAuth metadata, warm discovery in the background, and share bounded request-time retries with per-server cooldowns.
Preserve the existing discovered-tool boundary for explicit server calls.
Co-Authored-By: Codex
AWS no longer serves `anthropic.claude-3-sonnet-20240229-v1:0`. The streaming
path returns a plain 404, "Model with the provided id
anthropic.claude-3-sonnet-20240229-v1:0 is not found", and the non-streaming
path answers 500 for the same reason. Our own cost map has carried a
2026-07-30 deprecation date for it since #36538
That accounts for 20 failures across local_testing_part1, local_testing_part2
and llm_translation_testing. litellm maps both statuses correctly, so the
tests are what went stale, not the client
Replacement is `us.anthropic.claude-sonnet-4-5-20250929-v1:0`: a like-for-like
Sonnet, and the newest Bedrock Sonnet this repo exercises against the real API
in tests/e2e. Newer ids exist in the cost map, but nothing in the repo calls
them live, so picking one would be an unverified guess about model access on
the CI account
Scope is limited to the tests that actually issue a request. The occurrences
that assert on the model string itself, or that feed mocked transformations,
keep the old id so their assertions stay meaningful
* feat(router): make routing groups callable as virtual models and list them in /v1/models
* fix(router): traffic-scoped cooldown exemption, live model_names on delete, group-info cache invalidation
* fix(router): share one recognized-model predicate across proxy gates, resolve aliases in group cooldown, read metadata via the dual-bucket owner
* fix(router): close the gate and cache families for callable groups, strip member access_groups from group rows, prove cooldown wiring end to end
* refactor(router): cache materialized group rows under the model-group cache owner and drop the redundant wiring test
* fix(router): warn-and-shadow on group name collisions, name-level test coverage for group helpers, faithful router doubles in a2a and cursor tests
* test(router): pin group cooldown metadata across the retry path
Gemini image usage carries prompt_tokens and friends as extra fields on
ResponseAPIUsage, which collided with the bridge's explicit kwargs and
raised TypeError. Exclude keys the bridge already sets explicitly.
`supports_native_structured_output` was set only on the bare `deepseek.v3.2`
and `zai.glm-5` entries, so the cross-region inference profiles and the
region-pinned ids resolved to None. The flag gates the native
`outputConfig.textFormat` branch in BedrockConverseConfig, so callers
addressing the same model as `us.deepseek.v3.2` or
`bedrock/us-west-2/deepseek.v3.2` silently fell back to synthetic tool
injection. `us.` is the form Bedrock steers callers toward, so the most
common way to reach these models was the one missing the capability.
Adds the flag to the 12 affected ids and keeps the packaged backup in sync.
test_get_model_info_bedrock_models already caught the region-pinned ids, but
it filters on `litellm_provider == "bedrock"` and the cross-region profiles
carry `bedrock_converse`, so reverting just `us.deepseek.v3.2` and
`eu.deepseek.v3.2` left it green. The new parity test covers the prefixed
profiles and fails on exactly that mutation.
* test(e2e): cover google-native generateContent framing and prometheus queue time
Adds live coverage for three shipped regressions that had none, all reached
through surfaces a customer drives from Google SDKs and operator dashboards.
The managed google-native route (`/v1beta/models/{model}:generateContent`) had
no harness support at all, so EndpointsClient gains generate_content and
stream_generate_content plus the request body models, and a new suite asserts
the two contracts that broke there: the response carries
x-litellm-response-cost so SDK traffic reconciles against spend (LIT-4076), and
the stream relays single-prefixed SSE frames with no OpenAI [DONE] terminator.
A doubled `data:` prefix, a leaked bytes literal, or the [DONE] sentinel each
fail the stream test; [DONE] absence is only asserted once real content has
arrived, because a first-chunk upstream error legitimately falls back to the
OpenAI error shape and does emit it.
The prometheus test pins litellm_request_queue_time_seconds to an actual
observation on our own key's series rather than to the family merely existing,
which is the distinction the original regression turned on: the histogram stayed
registered while nothing was ever written to it (LIT-2034).
Each assertion was mutation-checked against the live proxy; inverting the
[DONE] expectation, the cost-header expectation, or the metric name fails the
corresponding test.
* refactor(e2e): simplify google native coverage
* test(e2e): add failing reproducers for two open gateway bugs
Both tests assert the behavior a customer expects and both are red today. They
are reproducers, not regressions: the product is wrong, not the tests.
Native passthrough returns almost none of the operational headers the managed
route does. A /gemini/ generateContent call comes back with three x-litellm-*
headers and no x-ratelimit-* at all, against sixteen and four on
/v1beta/models/{m}:generateContent for the same prompt, and critically it omits
x-litellm-response-cost. Customers front provider-native traffic through this
route and read those headers to reconcile spend and pace themselves, so native
traffic is currently invisible to the tooling that covers every other route.
/budget/update rejects any model_max_budget with a 500. The reported symptom was
model ids containing dots, and that reproduces (prisma raises "Unexpected
`-5.2[FloatValue]` Expected `:`" because the key is interpolated into a GraphQL
query unquoted, so glm-5.2 lexes as an identifier followed by a float), but the
plain name gpt4o fails too, on a separate "model_max_budget should be of any of
the following types: Json" type mismatch at budget_management_endpoints.py:173.
Omitting the field returns 200. The test drives both names so the failure says
whether per-model budgets are broken outright or only for punctuated ids; today
it stops on the plain name, which is the wider bug.
* test(e2e): add reproducer for unenforced end-user per-model rate limits
model_max_budget accepts an rpm_limit alongside the spend cap, and /budget/new
stores it: the create response echoes {"gemini-2.5-flash": {"rpm_limit": 1,
"max_budget": 100.0, "budget_duration": "1d"}}. Attach that budget to an end
user, drive three calls as that user, and all three return 200. The limit is
accepted, persisted, and then ignored.
The same shape already works when the budget hangs off a key, which is what
makes this quietly dangerous: the API gives every indication the cap is in
force. A customer using it to hold one end user to a slow rate on a shared key
gets no throttling at all.
Harness additions this needs: ModelBudgetEntry carries the rpm_limit/tpm_limit
the route already accepts, BudgetNewBody and create_budget carry
model_max_budget, and create_customer can attach an existing budget_id rather
than only an inline max_budget.
Red today, for the reason in the assertion message.
* test(e2e): tighten model_max_budget reproducers and drop in-loop closure
Trim the reproducer docstrings to the contract they assert, keeping the
failure messages that document each red-by-design bug. Replace the nested
per-model closure in the /budget/update test with a module-level predicate
and a per-model helper so nothing closes over a loop variable, and fix the
import order the merge left unsorted.
* test(e2e): skip the three reproducers while their gateway bugs stay open
The passthrough header contract, /budget/update model_max_budget, and
end-user per-model rpm enforcement reproducers all still fail against
staging by design. Skip each with the product gap named so the combined
suite can gate merges on green while the collector keeps reporting the
cells as uncovered.
* test(e2e): validate model budget response contracts
* refactor(e2e): unify model budget schema
* refactor(e2e): reuse shared model budget type
* test(e2e): cover vendor strategy gaps for chat contract, image edits, auth, team activity
Resolves the first slice of LIT-4778 (vendor API testing strategy): image edits happy path, chat multi-turn + validation + sanitization, LLM-route auth header matrix, and /team/daily/activity structure
* test(e2e): expand vendor API strategy coverage across endpoints
Adds validation cases on existing endpoint suites, plus vector stores, search,
bedrock native, realtime HTTP secrets/calls, responses retrieve, files/batches
contract, and chat stream SSE. Registers coverage cells for LIT-4778
* test(e2e): finish vendor strategy open items
Audio transcription negatives, vector-store file attach/poll/search,
OpenAI moderation category matrix across chat/messages/responses, and
smoke model matrix for chat (LIT-4778)
* test(e2e): harden vendor strategy suite against live env edges
Fix stream [DONE] tracking, XSS no-crash contract, realtime model routing,
vector store list/search models, responses validation, and provider-denied
Bedrock paths so the suite is stable against a live proxy
* test(e2e): rename suites, drop vendor_contract, fix greptile gaps
Move shared status helpers into e2e_http, rename chat auth headers and
chat security suites, remove vendor_contract and dev_config files_settings,
and tighten transcription validation plus vector-store search assertions
* test(e2e): route bedrock stream disconnects through e2e_http
Catch mid-stream RequestException in the shared harness so bedrock native
tests do not import requests directly
Every TypedDict field must carry a ReadOnly[...] qualifier (PEP 705),
nesting freely with Required/NotRequired/Annotated. Detection covers the
class form (including same-module transitive subclasses) and the
functional form. The 4519 existing violations across litellm/ are
grandfathered via type-discipline-budget.json; suppress deliberate
writable keys with # writable-ok: <reason>.
The otel trace tests asserted that a streamed call produces exactly one gen-AI
span. The proxy opens one gen-AI span per upstream attempt, so a call the
router retried carries an error span for every failed attempt beside the one
that answered, and the assertion fails on a request that succeeded.
Select the served attempt instead: drop spans whose otel.status_code is ERROR,
require exactly one survivor, and run the TTFT and streaming-flag assertions
against it. That keeps what these assertions exist for, a split trace or a
stream logged as two served spans, while tolerating a retry.
Only the failed attempt lacks TTFT, so the old code also had a second failure
mode: when the first span happened to be the error one, the test reported the
attribute as missing rather than as belonging to a different attempt.
test_span_selection.py covers the selection itself against Jaeger-shaped
payloads and carries no e2e marker, since reproducing a first-attempt failure
live is not something a test can arrange.
Drop the transform overrides that swapped response.usage to the chat
shape, which broke the /v1/responses client contract. Provider extras
like server_side_tool_usage_details already survive validation via
ResponseAPIUsage extra fields, so the shared usage bridge now carries
them onto the bridged chat Usage generically. The web_search_call
output gate also reads dict output items, since items that fail SDK
validation stay plain dicts, and the chat path gains billing tests.
* fix(deps): ship boto3 with the base SDK so bedrock works out of the box
* keep boto3 listed in the proxy extra as well
* scrub ambient AWS env vars in the base SDK bedrock smoke check
* fix(alerting): dedupe scheduled Slack spend reports across pods
Every pod ran its own weekly/monthly spend report jobs, prometheus
fallback stats cron, and daily report loop, so deployments with
multiple replicas or uvicorn workers received one copy per pod.
Gate each scheduled send behind the shared PodLockManager redis lock.
The lock is never released: its TTL (the full reporting window for the
weekly interval job, whose per-pod anchors drift by boot time and
jitter) doubles as a sent-this-window marker. acquire_lock returning
None (no redis wired) proceeds, preserving single-pod behavior.
Also generalize the pod lock could-not-acquire log line, which claimed
to be about spend tracking for every consumer.
Fixes#14809
* fix(alerting): harden spend report locks after adversarial review
Weekly lock TTL gets an hour haircut: with ttl equal to the interval,
the winner re-fires just before its own key expires, reacquires without
a TTL refresh, and the key then lapses in time for a trailing pod to
re-send. Job/lock ids move to litellm/constants.py per convention, and
spend_report_frequency now rejects non-positive day counts, which
previously coerced to an every-second schedule and would now compute a
negative lock TTL that silently never sends.
Adds the missing test coverage the review flagged: startup_event's
pod_lock_manager wiring (identity-asserted), the prometheus closure's
positive path, and the ungated immediate prometheus send pinned to
exactly one await.
* test(alerting): consolidate spend_report_frequency validator coverage
Drops a duplicate non-positive-days test and parametrizes the survivor
over the suffix half of the validator too
* fix(alerting): route the startup prometheus fallback send through the pod lock
Greptile caught that the boot-time send still ran once per pod when
PROMETHEUS_URL is set, the same duplication class this PR removes
* fix(alerting): make report lock acquisition non-reentrant
Greptile caught that a pod booting within an hour of the fallback stats
cron sent twice: the startup send takes the lock, then the cron fire
hits acquire_lock's reacquire branch, which returns True for the
holder. Window-marker gates now pass allow_reentrant=False so a live
lock blocks everyone including its holder; leader-election consumers
keep the reentrant default
* test(proxy): give spec'd ProxyLogging mocks a db_spend_update_writer
_initialize_slack_alerting_jobs now reads it for the pod lock manager,
and spec=ProxyLogging blocks instance-only attributes
