Resolve conflicts in the e2e Responses suite: main replaced the bespoke
EndpointsClient with provider SDK clients (76bf0cd579), so the four added
Vertex and Azure OpenAI /v1/responses tests are ported onto the OpenAI SDK
and tool_choice=required now comes from the SDK instead of a custom
request model.
Register s3_access_key_id, s3_secret_access_key and s3_encryption_key_id as
LiteLLM-owned batch params so they are no longer forwarded to Bedrock as
additionalModelRequestFields (which 400s ordinary chat on a batch-configured
deployment), keep them on CredentialLiteLLMParams so the batch/file paths
still receive them, and redact the S3 credential key names in debug logs.
Resolves LIT-8290
Co-authored-by: yucheng <yucheng@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Register s3_access_key_id, s3_secret_access_key and s3_encryption_key_id as
LiteLLM-owned batch params so they are no longer forwarded to Bedrock as
additionalModelRequestFields (which 400s ordinary chat on a batch-configured
deployment), keep them on CredentialLiteLLMParams so the batch/file paths
still receive them, and redact the S3 credential key names in debug logs.
Resolves LIT-8290
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
The deleted wrapper put cache: {"no-cache": true} on every request body, so the
gateway's response cache never answered a re-sent prompt. The SDKs send nothing
of the sort, and the mid-conversation prompt-cache priming loop re-sends an
identical body until the provider reports a warm cache, which a cached reply
never does. NO_PROXY_CACHE in sdk_clients.py restores the field as extra_body
on every messages, responses, completions, and embeddings call.
The wrapper also returned a 4xx as a value where the SDKs raise. The Bedrock
safety_identifier test judges the captured Converse body, and Claude on Bedrock
rejects the forwarded field with a 400, so the /v1/responses leg now suppresses
openai.BadRequestError the way the chat leg carries the same 400 as a Result.
Resolves the six conflicts main accumulated against the SDK migration.
endpoints_client.py stays deleted (main only added the safety_identifier
kwarg to it), and its two new importers move onto the SDK clients: the
Bedrock safety_identifier forwarding test in test_responses_e2e.py now
sends its /v1/responses leg through the OpenAI SDK and its chat leg
through ProxyClient.chat, and test_messages_azure_foundry_e2e.py drops
the x-litellm-rust header check together with EXPECT_RUST, which main
removed from e2e_config in a84f68b6e3. proxy_client.py keeps both the
rerank and router-settings model imports. uv.lock is main's lock plus
the anthropic e2e-dev entry, and the pricing JSONs are taken verbatim
from main since this PR never meant to change them.
The Buildkite ephemeral stack runs the gateway in another pod, so it cannot reach the pytest host's provider edge. The GitHub changed-e2e lane runs gateways on the runner and sets E2E_PROVIDER_EDGE_HOST_REACHABLE
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
AGENTS.md allows a vendor-owned literal only when its source and date are cited
next to it. A live realtime test cannot avoid naming a model, so record how the
id was checked, and record that a retired id fails as a hang rather than an
error so the next reader does not start by suspecting litellm.
AWS retired amazon.nova-sonic-v1:0. GetFoundationModel now answers
ResourceNotFoundException "This model version has reached the end of its life",
and opening a bidirectional stream against it fails with ValidationException
"The provided model identifier is invalid". amazon.nova-2-sonic-v1:0 is the
active replacement.
The test passed on builds 219 (2026-09-16) and 246 (2026-09-17) and has failed
every run since, three attempts per build, with no litellm change to the
realtime path in between. The symptom was a clean websocket close: Bedrock ends
the stream rather than erroring, the forwarder treats a None receive as a normal
stream end and closes the client socket, so the client sees ConnectionClosedOK
and the test fails waiting for response.done.
litellm already carries both models in the cost map, with
"deprecation_date": "2026-09-14" on the old one, and the promptStart
transformation already sends the audioOutputConfiguration that nova-2-sonic
requires; only the test constant was left behind.
Verified against live Bedrock with the promptStart shape the transformation
builds: amazon.nova-sonic-v1:0 raises "The provided model identifier is
invalid", amazon.nova-2-sonic-v1:0 opens a session and returns a usageEvent.
The mocked handler and provider-cache tests keep the old id: it is only a label
there, no call reaches AWS.
The check spins up a hypercorn TLS peer and drives the SDK's own httpx
handlers at it, so it needs litellm importable, hypercorn installed and a
loopback socket. It lived under tests/e2e, whose Buildkite runner image
installs neither litellm nor hypercorn by design (the suite drives a
remote proxy over HTTP), so every scheduled e2e build since #230 failed
to import the module and pytest reported it as a collection error. The
unit tree bans sockets, so it does not belong there either
tests/integration is the CircleCI tier built for real TCP against local
protocol peers. This adds an sdk shard to it for cases that exercise the
SDK's clients with no gateway in the path, registers the two HTTP/2
nodes in the contracts manifest, and adds the shard to the CircleCI
matrix. The test now flips the feature through LITELLM_HTTP2 (the user
surface) instead of patching module attributes, and asserts the version
the peer observed on the wire next to the one the client reports
Each route now has litellm/rust_bridge/<route>/{entrypoints,callbacks}.py and a
public dispatch module (litellm/chat_completions/dispatch.py,
litellm/responses/dispatch.py, litellm/messages/dispatch.py) that binds the
public call to the legacy Python signature, builds a frozen request, and asks
the runtime to pick Rust or Python from the catalog. The legacy implementations
stay in litellm/main.py, litellm/responses/main.py and the anthropic messages
handler, and litellm/__init__.py re-exports the dispatch names over them the
same way it already does for ocr
The per-handler shims in rust_bridge/chat_completions/native.py and
rust_bridge/messages/native.py are removed along with their call sites in the
anthropic and bedrock chat handlers and the http handler. The exception
mapping that every callbacks module repeated moves to rust_bridge/failures.py
and the signature binding helpers to rust_bridge/public_call.py
The key normalizes a unique marker so two builds match, which is the whole
point, but it makes this test's identity collide with an earlier run's: it mints
a fresh receipt, sends it through a tool result, and asserts the model echoes it
back verbatim, so a stale recording matched and answered with the old receipt.
Build 223 is where that surfaced, once the corpus was full enough for the first
call to hit. A test that asserts a provider echoed this run's own unique value
belongs on the live path.
The first cache-enabled litellm-e2e build (211) showed three gaps in the shared provider cache:
Every OpenAI response carries Cloudflare bot-management Set-Cookie headers, and the capture rejected any response with Set-Cookie, so no OpenAI response was ever recorded (179 of 372 misses rejected). The edge already withholds Set-Cookie from the proxy, so drop it before validating and storing instead of rejecting.
The provider prompt-caching tests need fresh provider state: a replayed priming response reports cache creation rather than a cache read, and the TPM test then trips the key limit. Mark both modules provider_live.
TestApiBaseSeam::test_live_mode_returns_none ran inside the cache-enabled runner and saw the shared edge; isolate it from E2E_PROVIDER_CACHE.
Re-migrates the e2e tests that main extended through endpoints_client
since this branch was opened (Azure Foundry, mid-conversation system
messages, Bedrock web search, google native streaming) onto the
provider SDK clients, so no endpoints_client reference remains
* fix(mcp): honor an explicit null on toolset update, cover MCP lifecycle e2e
PUT /v1/mcp/toolset dumped its payload with exclude_none, so a field sent as
null looked exactly like one the caller left out and the stored value
survived. An admin could not clear a toolset's description: the save reported
success and the old text came straight back. It now dumps with exclude_unset,
so absent keeps and null clears, which is what PUT /v1/mcp/server already did.
A null tools list clears the selection to empty, and a null toolset_name is
ignored because a toolset always has a name.
Adds create, read, partial-update, clear and delete e2e coverage for MCP
servers and toolsets, with every read-back polled on every replica so an edit
that lands on one replica and not another fails the test, plus an enforcement
test proving a key granted a toolset lists exactly that toolset's tools
against the real Datadog upstream.
* fix(e2e): refuse a read-back that no replica serves
A read-back over an empty replica mapping satisfied every predicate and
returned as if it had converged, so it would have asserted nothing and
passed. No wiring can produce that today, since the replica list always
falls back to at least one URL, but a helper whose whole job is proving a
write reached every replica should not have a shape that passes vacuously.
* fix(mcp): keep a null tools list a no-op on toolset update
Treating a null tools list as a clear meant an existing client that sends
tools=null during a partial update, meaning "leave the selection alone",
silently lost every tool the toolset grants. That is a permission surface,
so the quiet version of it is the worst version.
A toolset always has a tool list, the same way it always has a name, so a
null on either is now a no-op. Emptying the selection is an explicit [],
which cannot be confused with a field the caller left out, and which is
what the dashboard already sends.
* fix(e2e): keep MCP admin routes on the data plane
/v1/mcp/* is a lazily mounted feature, so a gateway registers it on the first
matching request, which happens after the startup route trim that drops
management endpoints. Routing it to the control plane therefore sent every MCP
call to the one backend process: the new lifecycle read-backs proved a single
process rather than every replica, and mcp_client's await_registered barrier
waited on a registry that does not serve the tools/list call it guards, so the
existing MCP suites polled a gateway that had not synced yet until poll_timeout
Verified against a two-gateway split stack (backend on 4001, gateways on 4010
and 4011, one postgres): both gateways answer /v1/mcp/server and /v1/mcp/toolset,
and each served 6 server reads and 7 toolset reads over the run
* fix(e2e): grant the toolset by the tool's own name, not the wire name
tools/list serves a tool as <prefix><tool_name>, but a toolset grants by the
tool's own name: resolve_toolset_permissions reads toolset.tools[].tool_name
straight through, and the prefix is added on the way out. The test built the
toolset from the names tools/list reported, so the grant matched nothing, the
scoped key listed no tools, and await_tools ran out its whole poll_timeout
before failing
Measure the prefix off search_datadog_logs, whose own name is known, rather than
guessing it from the alias, since the proxy can be configured to prefix with a
short server id instead. The expectation compared against tools/list stays in
wire names; only what the toolset stores crosses back
* test(mcp): build immutable lifecycle updates and replica results
* test: validate opaque stream IDs and hide log-reader credentials
* test: isolate auto-router scenarios and clean partial setup
* test: honor Datadog search rate-limit reset headers
* test: share the Datadog read-back deadline across retries
* test: preserve captured MCP toolset update fields
The Anthropic and Together AI /v1/messages streaming tests required at
least two content_block_delta events. How many deltas a reply is split into
is the provider's choice, and Haiku answers a short count in one or two, so
the assertion failed on provider variance with no change in the proxy: four
of the day's full runs on the PR e2e gate went red on it on 2026-09-05.
The harness now stamps when each SSE event reached the client
(StreamingResponse.stream_event_arrivals, index-aligned with stream_events,
with the clock injectable so the reader has a unit test). Both tests ask for
a reply long enough to take seconds to generate and require the first
content delta to land at least STREAM_MIN_LEAD_SECONDS before message_stop.
A relayed stream shows a lead of about two seconds. A proxy that buffered
the response delivers every event in one burst and fails every time, which
a whole-response buffering relay in front of a live proxy confirmed. The
event-grammar assertions are unchanged.
Replay hands the proxy its recorded chunks back to back, so timing says
nothing there. The assertion is gated on provider_paces_stream() and replay
proves the grammar only, which tests/e2e/CLAUDE.md now says.
Anthropic now returns the 64-token 'count to 20' reply in one to three content_block_delta events, measured directly against api.anthropic.com and through proxies at 7672399 and 49a1145 alike, so the incrementality assertion (at least two deltas) failed in litellm-e2e builds 125, 130 and the 278 rerun with no proxy change behind it. A 'count to 100' reply at max_tokens 400 arrived in five to fifty deltas across every measured run
Four registry cells that had no e2e test. The cache_control suite gains a direct
Anthropic case (the same cache_control prefix the Bedrock and Vertex rows send)
and an OpenAI case, where caching is automatic so the prefix goes out as a plain
system string with a prompt_cache_key; both assert the second identical call
reports cache-read tokens. The shared second-call helper now takes the send
callable so the OpenAI shape fits without a second copy of the retry loop.
The embeddings suite gains a cohere/embed-v4.0 deployment that must return a
non-zero vector, and the passthrough suite gains an OpenAI-format chat through
the raw /openai/v1/chat/completions prefix that must relay a real completion and
log a costed pass_through_endpoint row whose token counts match the usage the
caller was served.
Adds TestAnthropicChatCompletions to the chat completions regression suite,
registering a claude-haiku-4-5 deployment via /model/new and asserting the
streamed call delivers real content deltas and a tool-forced call returns a
well-formed get_weather tool_call on both the non-streamed and streamed paths.
Covers three P0 registry cells that had no e2e test.
/v1/batches now rejects a missing input_file_id with a 400 through
raise_if_required_body_param_missing, so the contract negative that was
skipped for "500s instead of 400" passes as written. Verified against a
live proxy.
The three Datadog MCP tests were skipped because each one sent a
`telemetry` argument that search_datadog_logs rejects with "unexpected
additional properties". That argument was never a documented Datadog
parameter and no assertion reads it, so it is dropped and the tests run
again unchanged otherwise.
The two vision tests pointed at a Wikipedia-hosted cat photo, so every run
depended on upload.wikimedia.org staying up and unthrottled. It throttled,
and the 429 surfaced as a bedrock APIConnectionError, which reads as a
gateway failure rather than what it was.
The image is now a fixture in the repo, passed as a data URL. That also puts
the two providers on the same bytes: litellm downloads the image itself for
bedrock, while openai is handed the link and fetches it from its own servers,
so the hosted URL quietly meant the two tests were not testing the same thing.
The image was generated for this repo rather than borrowed, so nothing here
carries a third-party license. Also drops a stale comment about openai prompt
caching that sat above the vision helper; no caching test uses it.
The /v1/messages validator checked a tool_use block's name and id but not its
input, so a block whose location came back empty or wrong still passed, while
the chat side rejected the same damage. That gap predates this branch; it is
worth closing here because the point of the change is that every parallel call
is checked rather than counted.
AnthropicContentBlock now declares input as a typed field. It already survived
on extra="allow", but reaching it from a test needs a real field to keep the
e2e basedpyright gate at zero. Serialization is unchanged: bodies are dumped
with exclude_none, so a block without an input still replays exactly as before.
The together backend is picked as the cheapest chat row that supports both
tools and reasoning, which currently resolves to together_ai/openai/gpt-oss-120b.
That row is marked supports_parallel_function_calling, so one weather prompt
can legitimately come back as several get_weather calls. Both tool tests
asserted exactly one call, so a parallel answer failed them even though the
gateway handled it correctly.
They now check every returned call instead of counting them: each one has to
be a get_weather naming Paris, with an id a tool result can answer. Dropping,
misnaming, or mangling a call is still red; only the count is the model's
business. The round trips answer every call rather than just the first, which
is also what the Anthropic Messages spec asks for.
