mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-10 08:28:25 +00:00
4786dd861b
* test(proxy): cover /health scoping for wildcard keys and team model allowlists * fix(proxy): scope /health with the model access check auth runs on a request * fix(proxy): scope /health to the model group an allowed name routes to A key alias, team alias, or router model_group_alias lets auth admit a requested name by checking its target, so /health also kept the deployment under the requested name, one the key's requests never reach. Keep only the served model group each admitted name dispatches to, through the same alias layers /v1/models mirrors, and let alias_map read its own read-only output. * test(proxy): cover /health skipping a deployment that has no model name for a restricted key * fix(health): show deployments reached through wildcard and alias allowlists, and by id when an alias shadows the name * chore(router): drop should_include_deployment from the outside-use allowlist now that /health no longer calls it * fix(health): never list a default-fallback deployment for an unserved grant, and keep access-group keys to their group's deployments * fix(health): resolve ?model= through the caller's aliases the way chat does * fix(proxy): apply user and agent model limits to /health and yield between checks /health now runs the same user and agent model checks chat runs, so a personal key whose owner may call only some models, or an agent key limited by its access group, no longer sees or probes other deployments. Each per-name check yields to the event loop, and ?model= reaches the wildcard deployment serving an allowed name * fix(proxy): check owner and agent grants for wildcard routes in /health and fail on non-access auth errors * fix(proxy): answer /health auth lookup failures the way chat does and record ?model= checks under the requested name * test(proxy): pass recorded_model_name in the direct health save tests * fix(proxy): count alias names as /health candidates so aliased names a key may call get probed * fix(health): narrow wildcard-served deployments to the key's access groups * fix(health): narrow every deployment serving a name to the key's access groups, team-owned ones included * fix(health): narrow a team deployment's internal name to the key's access groups like its public name * fix(health): keep a team deployment's internal name narrow when a limit holds only that name An agent or key limit that holds one team deployment's internal name no longer reaches every deployment behind the team's public name. The internal name maps to its public name only when the public name is also admitted. * test(health): cover all-proxy-models next to * for a key that also holds an access group * refactor(auth): keep the _as_proxy_exception name so main's tests still find it * fix(health): keep an internal team name a wildcard grant admits as its own target --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|---|---|---|
| .. | ||
| html_forms | ||
| admin_ui_utils.py | ||
| auth_cache_invalidation_pubsub.py | ||
| banner.py | ||
| cache_aware_routing.py | ||
| cache_coordinator.py | ||
| cache_pydantic_utils.py | ||
| callback_config_validation.py | ||
| callback_utils.py | ||
| claude_code_marketplace_auth.py | ||
| codex_base_instructions.md | ||
| codex_bundled_models_0.159.3.json | ||
| codex_model_catalog.py | ||
| config_includes.py | ||
| config_sync_pubsub.py | ||
| credential_hydration.py | ||
| custom_openapi_spec.py | ||
| debug_utils.py | ||
| discoverable_model_filter.py | ||
| encrypt_decrypt_utils.py | ||
| error_body_call_id.py | ||
| expired_ui_session_key_cleanup_manager.py | ||
| get_routes.py | ||
| healthy_model_filter.py | ||
| http_parsing_utils.py | ||
| json_merge_patch.py | ||
| key_rotation_manager.py | ||
| load_config_utils.py | ||
| model_deprecation.py | ||
| model_listing_utils.py | ||
| openai_endpoint_utils.py | ||
| openai_error_payload.py | ||
| openapi_route.py | ||
| openapi_schema_compat.py | ||
| path_utils.py | ||
| periodic_reload_schedule.py | ||
| prompt_cache_prediction.py | ||
| prompt_cache_pricing.py | ||
| proxy_rate_limit_error.py | ||
| proxy_state.py | ||
| rbac_utils.py | ||
| realtime_utils.py | ||
| registry_read_through.py | ||
| request_tag_mappings.py | ||
| reset_budget_job.py | ||
| resource_ownership.py | ||
| responses_stream_errors.py | ||
| scheduled_job_stagger.py | ||
| semantic_text_index.py | ||
| slow_gc_logger.py | ||
| sse_keepalive.py | ||
| static_asset_utils.py | ||
| swagger_utils.py | ||
| timezone_utils.py | ||
| token_count_log.py | ||
| token_count_settings.py | ||
| user_api_key_cache.py | ||
| validation_error_body.py | ||