The LIT* type-discipline gate counts mutable-collection use the ruff pass does
not see. Annotate the reads with read-only Sequence/Mapping views, compare
response_format against a frozen MappingProxyType sentinel, and mark the
genuinely mutable litellm message and deployment-list boundaries with
mutable-ok reasons, so LIT001 and LIT002 stay under their limits.
- register the continuation deployment filter as a process-global singleton
instead of tracking it per router, so discarding one router no longer removes
the type-deduplicated filter that other live routers still depend on
- disqualify reasoning_content: it reaches the caller as visible reasoning that
a text-only prefill cannot carry, so a reasoning-capable fallback would
re-derive it and produce an incoherent response
- move MID_STREAM_CONTINUATION_KWARG to constants and lazy-import the filter
class in the router, breaking the module-level import cycle CodeQL flagged
- trim the added docstrings to the repository comment policy
Address review of the mid-stream continuation:
- keep response_format={"type": "text"} eligible; only json_object / json_schema
and other structured formats decline, since "text" is the unconstrained default
- fold a new partial into an existing trailing assistant prefill instead of
appending a second one, so a nested mid-stream break stays a single prefill
turn even on providers that do not merge consecutive assistant messages
- split the deployment prefill-capability lookup into two readable steps
- cover the merge_reasoning decline branch, the text response_format pass, and
the prefill-folding path with tests
A chat-completions stream that breaks after content has been delivered was
re-raised and the fallback deployment never ran (#40404), while the
Responses-API path already continues via a prefilled assistant turn. Add the
same for chat completions behind `enable_mid_stream_fallback_continuation`
(opt-in, async-only), so a post-content break re-enters the fallback chain with
the partial text as an assistant prefill instead of surfacing the error.
Continuation is only attempted when it is safe: the stream emitted plain
assistant text (no tool/function calls, thinking blocks, reasoning items,
audio or images), the request is not constrained output (response_format or a
forced tool_choice) and not merge-reasoning mode, and the fallback target's
model supports assistant prefill. The target check runs as a deployment
pre-call filter, so a chain with no prefill-capable deployment empties and the
original error is surfaced rather than a duplicated or rejected request being
sent. Plain reasoning_content is treated as out-of-band and does not block a
continuation, matching the Responses-API path; Anthropic thinking is excluded
because its signed thinking blocks cannot ride a text-only prefill.
CustomStreamWrapper now latches whether a disqualifying delta was streamed and
carries it on MidStreamFallbackError so the router can decide without
re-scanning chunks. Default behavior is unchanged.
CredentialLiteLLMParams omitted tenant_id, client_id, client_secret,
azure_scope, azure_username and azure_password, so the strict dump used
by credential reuse and Azure client init dropped them and the reused
credential ended with no auth at all
Co-authored-by: yassin <yassin@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
The e2e harness exists to prove product features end to end against a live
proxy. The prior Hard Rule carved out an exception for "tests that cover the
harness itself" and pointed at coverage_registry/test_collector.py, which in
practice invited unit tests of harness helpers to be staged alongside e2e
work. That is the wrong tool: harness logic that is worth locking down does
not need a mock-driven unit test living under tests/e2e.
Drop the carve-out. The Hard Rule now reads that no unit tests of any kind
belong under tests/e2e, and the passing mention of unmarked harness coverage
in the transport section is removed so the doc no longer contradicts itself.
coverage_registry/test_collector.py still exists on disk and is left in place
for now; whether to relocate or remove it is a separate decision.
Keeps the base's rule that a non-admin id lookup matching no spend-log row answers 403, so the detail route never consults cold storage without an owner row
* fix(proxy): bound tool and guardrail index create_many by the spend-log statement budgets
One flush drains up to MAX_LOGS_PER_INTERVAL source transactions or logs, but a
transaction fans out to one LiteLLM_SpendLogToolIndex row per tool and a log
to one LiteLLM_SpendLogGuardrailIndex row per guardrail, so the index
create_many payload was unbounded. Both index writes now go through
spend_log_write_batches(SPEND_LOG_WRITE_BATCH_MAX_BYTES, SPEND_LOG_WRITE_BATCH_MAX_ROWS).
The tool index write moves out of the rollup batch_() so the split reduces
the query-engine payload; replayed index rows are no-ops under
skip_duplicates, and the daily rollup upserts stay in one transaction
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(proxy): pin the row budget in the index fan-out tests
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: yucheng <yucheng@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
The bridge probe asked `responses_api_bridge_check` with the summary read straight off
the Responses object, but `litellm.completion` reads it from `optional_params` via
`peek_reasoning_summary_aliases`, which the bridged request never populated. So gpt-5,
gpt-5.1 and azure/gpt-5 answered "bridging" to the probe and "not bridging" for real,
and the object still landed on Chat Completions, which only takes a string
`reasoning_effort` is now always the effort string, and `summary` rides the
`reasoning_summary` alias that main.py already reassembles into `{effort, summary}` on
the bridged path. The alias is emitted only when the probe says the model bridges, so
no chat provider ever sees it, and the probe is now asked with the exact params this
transform emits
The debug call built its message with an f-string, which
test_logging_calls_do_not_build_their_message_eagerly rejects. Pass the exception
as a %-style argument so the message is only built when the log is emitted.
The bridge probe called responses_api_bridge_check without api_base, so it
resolved the OpenAI base from globals and environment rather than from the
request, while litellm.completion runs the same check with the caller's value.
Today the two cannot disagree: this path always supplies a reasoning_effort,
which short-circuits the endpoint term in the only arm that reads it. Passing it
anyway keeps the probe a faithful mirror of the definitive check rather than one
that happens to agree.
The Responses API takes reasoning as an object, {effort, summary}. Chat
Completions takes reasoning_effort as a string enum and has no equivalent of
summary, but the completion bridge forwarded the whole object whenever summary
was set, which agentic clients set on every request.
Bedrock Converse guards its mapping with isinstance(value, str) and has no else
branch, so the object fell through, thinking was never enabled, and the caller
was billed for a non-thinking turn with nothing in the response to explain it.
The object is still forwarded for the one caller that can consume it: a model
whose cost-map mode is responses, which litellm.completion bridges back onto the
Responses API and reassembles {effort, summary} there. That decision is delegated
to responses_api_bridge_check, the same check litellm.completion runs, rather
than a second copy of the rule that could drift from it. An object carrying no
effort now yields no reasoning_effort at all.
/model/info fills a deployment's missing pricing in from the model cost map so the
Admin UI has a rate to display. Clients echo that whole model_info blob back on save,
and update_db_model merged it into the row, so editing an unrelated setting turned
that day's catalog price into a real per-deployment override. After that the
deployment ignored the cost map and Reload Price Data could no longer move it,
because the reload replays each deployment's stored pricing over the fresh catalog.
Drop the derived pricing from incoming model_info on the two write paths. The
drop-set is read off the same objects the read path uses, CustomPricingLiteLLMParams
plus the tiered *_above_N_tokens pattern that get_model_info passes through and no
model declares, so it cannot drift as new rates are added. output_vector_size is
exempt: it lives on the pricing model but is an embedding dimension, not a rate.
A deployment's own pricing still rides litellm_params, which is untouched, as is the
explicit-null clear, which reads the incoming model rather than the filtered dict.
The filter sits in the endpoint bodies rather than _add_model_to_db, which master-key
rotation reuses to re-serialize every stored deployment.
