`AnthropicCacheControlHook` spends the configured injection points on the first
message list it is shown and drops the message points that matched nothing. That is
right when the messages it sees are the ones going upstream. It is wrong for
/v1/responses: the system prompt lives in `instructions`, which only becomes a system
message once the chat-completion bridge builds one, so a role-targeted point matched
nothing and was thrown away before the message it wanted existed. Injection silently
did nothing across the whole surface.
Hand those points back instead, stamped as judged, when the caller says its message
list is provisional. The stamp is what makes carrying them safe: without it the next
pass re-judges the points against messages this pass has already marked and stands the
whole configuration down. Callers holding the final messages -- /chat/completions and
/v1/messages -- do not raise the signal and keep dropping unmatched points as before.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
An auto_router deployment is a marker, not something a probe can contact, so
`_run_model_health_check` returns `{}` for it and it lands healthy whatever is
behind it. This derives its verdict from the models it actually resolves.
Rules and owners:
- `strategy_router_dependencies` is the single answer to "what does this router
call": tier, default, classifier and embedding names per router kind, aligned
with what init and the request path actually use.
- `_health_check_eligible` is the single probe-eligibility gate, applied to the
requested set and to the pool a router's dependencies are drawn from alike, so
an opted-out deployment cannot re-enter through a router that depends on it.
- `_resolved_deployment_ids` resolves names through `get_model_list`, the same
composition of alias, routing-group and wildcard channels a request uses.
- A dependency reds its router only when *every* deployment behind the name is
known unhealthy. A replica this run never judged, hidden from the caller or
opted out of health checks, can still serve what the dead one drops, so
partial evidence leaves the verdict green. Absent information never reds.
- Verdicts settle over rounds, because a marker never fails a probe of its own
and a parent whose tier is a red router must inherit that fault. Both sweeps
are bounded loops, so a router cycle terminates green.
- Dependency probes are added only on the targeted `/health?model_id=` path the
dashboard uses per deployment, and are dropped from the response.
Resolves LIT-6073
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Vertex Gemini 3.x models route through cost_per_character (the cost_router
token-path gate only matches gemini-2), and its token fallbacks dropped
service_tier, so ON_DEMAND_FLEX responses were still billed at the standard
rate. Pass the tier through the call site and all four fallbacks.
The GA vertex model had no cost map entry, and the realtime cost handler
accepted the router's price-less auto-registered deployment entry for the
session.created model at zero-defaulted rates, so sessions billed 0.0 even
when base_model pointed at the priced preview key. Adds the GA entry at its
published rates and makes the handler fall through zero-defaulted candidates
unless their cost map entry explicitly declares pricing.
get_fireworks_session_id fell back to litellm_trace_id when no session id was
given. That id is generated per request (uuid4 when absent), so x-session-affinity
carried a different value every time and Fireworks prompt caching never hit;
cached_tokens stayed 0 across identical prompts.
The None path the original change described was effectively unreachable because
of it. Drop the fallback so affinity comes only from an id the caller actually
supplied: litellm_session_id, session_id, or metadata.session_id.
Callers who were relying on a trace id for affinity can pass litellm_session_id
instead, which is stable across the requests they want grouped.
Co-authored-by: mubashir1osmani <mubashir.osmani777@gmail.com>
Gemini API Maps-grounded prompts were billed as web search and Vertex AI Maps-grounded prompts were not billed at all. Classify grounding metadata per candidate into web search vs Maps requests, carry a distinct google_maps_grounding_requests usage counter through non-streaming and streaming paths, and price it via the new google_maps_grounding_cost_per_query cost map key with per-query and per-prompt defaults keyed off web_search_billing_unit. Fixes#35906
The image_edit health probe sent a 512x512 solid-gray PNG with the generic
chat prompt "test from litellm", an ambiguous pair OpenAI's gpt-image-1
output moderation sometimes rejects as moderation_blocked, which reported a
working deployment as unhealthy. The probe now sends a blue circle on a
white background with a descriptive edit prompt, and a provider moderation
verdict (ContentPolicyViolationError or a moderation_blocked error body) is
treated as proof the endpoint works rather than as an unhealthy deployment.
Live probes against every gemini_native_audio model on both providers show
setup accepts a valid prebuilt voice and 1007s only unknown voice names, so
the strip predicate rested on a false premise and silently discarded the
client's voice on AI Studio native-audio sessions
reload_search_tools_from_db is a read-modify-write of the shared llm_router
global: it reads the whole table, merges the config tools in, and replaces
router.search_tools wholesale. Two of those interleaving lets the older
snapshot's assignment land last and put back a tool the newer one deleted, so a
revoked tool keeps serving on the provider key it carried until the next reload.
Take MODEL_RECONCILE_LOCK, which add_deployment already uses to serialize the
same shape of work on the same global. It has to go on this entry point rather
than in _init_search_tools_in_db, because _init_non_llm_objects_in_db calls that
while already holding the lock and asyncio.Lock is not reentrant.
A separate search-tools-only lock would not close the race: the periodic
reconcile reaches _init_search_tools_in_db under MODEL_RECONCILE_LOCK, so only
that same lock orders an endpoint refresh against a cron tick.
Ordering across workers is unchanged and still reconciles on the next tick.
