revenium-python-sdk 0.12.0__tar.gz → 0.13.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/PKG-INFO +76 -6
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/README.md +75 -5
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/pyproject.toml +1 -1
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/cache_tokens.py +9 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/enforcement.py +165 -26
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/log_sanitize.py +6 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/metering_buffer.py +1 -5
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/proxy/guardrail.py +301 -35
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/proxy/middleware.py +17 -7
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_python_sdk.egg-info/PKG-INFO +76 -6
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/LICENSE +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/call_context.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/call_ownership.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/closable_scope.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/config.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/context.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/decorators.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/delivery_circuit.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/error_text.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/exceptions.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/failed_calls.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/json_safe.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/langchain_settings.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/load_diagnostics.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/metering.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/metering_pool.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/metering_status.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/metering_submission.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/outcomes.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/package_versions.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/parse_failures.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/patch_registry.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/perplexity_cost.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/perplexity_usage_metadata.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/prompt_extraction.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/responses_status.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/secret_shapes.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/selective_metering.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/shutdown_signals.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/sse_body_tap.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/subscriber.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/tool_event_transport.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/usage_iterations.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/vendor_model_aliases.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/wire_guards.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/LICENSE +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_base_client.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_client.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_compat.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_constants.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_exceptions.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_files.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_models.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_qs.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_resource.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_response.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_streaming.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_types.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_logs.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_proxy.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_reflection.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_resources_proxy.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_streams.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_sync.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_transform.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_typing.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_utils/_utils.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/_version.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/context.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/decorator.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/py.typed +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/resources/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/resources/ai.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/resources/apis.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/resources/events.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/ai_create_audio_params.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/ai_create_completion_params.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/ai_create_image_params.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/ai_create_video_params.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/api_meter_request_params.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/api_meter_response_params.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/event_create_params.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_metering/types/metering_response_resource.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/agentic_outcomes.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/bedrock_adapter.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/bedrock_model_names.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/bedrock_transport.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/config.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/inference_profiles.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/provider.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/raw_response.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/stream_create.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/anthropic/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/fal/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/fal/_call.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/fal/_metering.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/fal/_queue.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/fal/config.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/fal/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/fal/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/common/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/common/exceptions.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/common/protocols.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/common/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/common/types.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/common/utils.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/config.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/google_ai/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/google_ai/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/google_ai/provider.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/vertex_ai/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/vertex_ai/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/google/vertex_ai/provider.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/_metadata.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/_retry.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/anthropic_driver.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/budget_errors.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/litellm_driver.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/ollama_driver.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/openai_driver.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/openai_embedding_driver.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/griptape/universal_driver.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/job_context.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/job_history.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/job_type_economics.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/config.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/context.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/decorators.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/hooks.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/integrations/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/integrations/crewai.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/client/validation.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/proxy/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/litellm/proxy/_metering_owner.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/ollama/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/ollama/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/ollama/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/audio.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/azure_config.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/azure_model_resolver.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/config.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/exceptions.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/_abandoned_runs.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/_model_call.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/_perplexity_cost_capture.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/_prompt_capture.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/_text_completion_usage.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/_utils.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/langchain/unified_handler.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/provider.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/raw_response.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/openai/xai_usage.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/_metering_scope.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/middleware.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/patching.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/perplexity_sdk.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/provider.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/raw_response.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/streaming.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/perplexity/trace_fields.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/webhooks/__init__.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/webhooks/_verify.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_python_sdk.egg-info/SOURCES.txt +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_python_sdk.egg-info/dependency_links.txt +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_python_sdk.egg-info/requires.txt +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_python_sdk.egg-info/top_level.txt +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/setup.cfg +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/tests/test_metering.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/tests/test_metering_dispatch_timestamps.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/tests/test_patch_coverage.py +0 -0
- {revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/tests/test_release_verify_ticket.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: revenium-python-sdk
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.13.0
|
|
4
4
|
Summary: The official Revenium Python SDK — unified AI metering middleware for OpenAI, Anthropic, Google, Ollama, LiteLLM, Perplexity, and fal.ai.
|
|
5
5
|
Author-email: Revenium <support@revenium.io>
|
|
6
6
|
License: MIT
|
|
@@ -192,7 +192,7 @@ Some billable provider calls are not metered at all: see [Calls the SDK does not
|
|
|
192
192
|
The calls below are billed by the provider (some, such as batch and fine-tuning jobs, when the job runs rather than when you call them), but the SDK records no usage for them: nothing reaches Revenium and nothing is logged. On every build a test compares this list with the SDK's inventory of the public calls of the OpenAI, Anthropic, Perplexity, Gemini, Ollama, LiteLLM and fal clients, boto3's Bedrock Runtime client, and the Vertex AI model, session, batch prediction, tuning and context-cache classes, on the provider versions the SDK is tested on. Calls reached another way, for example LangChain classes that call their own vendor SDK, are outside that check, except LangChain embeddings, listed by hand below; wrapping one with `wrap()` logs a warning. Sync and async forms are folded into one name, and the `with_raw_response` and `with_streaming_response` forms of a listed call are not metered either. To record any of these calls, wrap it with `@meter_tool` (see [Tool Metering](#tool-metering)).
|
|
193
193
|
|
|
194
194
|
<!-- not-metered:start -->
|
|
195
|
-
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.create`, `beta.agents.sessions.create`, `beta.agents.sessions.events.create`, `beta.agents.sessions.events.stream`, `beta.agents.sessions.stream`, `beta.agents.sessions.update`, `beta.chatkit.sessions.create`, `beta.realtime.connect`, `beta.responses.compact`, `beta.responses.connect`, `beta.responses.create`, `beta.threads.create_and_run`, `beta.threads.create_and_run_poll`, `beta.threads.create_and_run_stream`, `beta.threads.runs.create`, `beta.threads.runs.create_and_poll`, `beta.threads.runs.create_and_stream`, `beta.threads.runs.stream`, `beta.threads.runs.submit_tool_outputs`, `beta.threads.runs.submit_tool_outputs_and_poll`, `beta.threads.runs.submit_tool_outputs_stream`, `completions.create`, `decisions.create`, `evals.runs.create`, `fine_tuning.alpha.graders.run`, `fine_tuning.jobs.create`, `images.create_variation`, `images.edit`, `images.generate`, `live.connect`, `live.create`, `live.forks.connect`, `live.sessions.accept`, `live.sessions.fork`, `live.sideband.connect`, `realtime.calls.accept`, `realtime.calls.create`, `realtime.connect`, `realtime.translations.connect`, `responses.compact`, `responses.connect`, `videos.create`, `videos.create_and_poll`, `videos.create_character`, `videos.edit`, `videos.extend`, `videos.remix`
|
|
195
|
+
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.create`, `beta.agents.automations.create`, `beta.agents.automations.invocations.create`, `beta.agents.sessions.create`, `beta.agents.sessions.events.create`, `beta.agents.sessions.events.stream`, `beta.agents.sessions.stream`, `beta.agents.sessions.update`, `beta.chatkit.sessions.create`, `beta.realtime.connect`, `beta.responses.compact`, `beta.responses.connect`, `beta.responses.create`, `beta.threads.create_and_run`, `beta.threads.create_and_run_poll`, `beta.threads.create_and_run_stream`, `beta.threads.runs.create`, `beta.threads.runs.create_and_poll`, `beta.threads.runs.create_and_stream`, `beta.threads.runs.stream`, `beta.threads.runs.submit_tool_outputs`, `beta.threads.runs.submit_tool_outputs_and_poll`, `beta.threads.runs.submit_tool_outputs_stream`, `completions.create`, `decisions.create`, `evals.runs.create`, `fine_tuning.alpha.graders.run`, `fine_tuning.jobs.create`, `images.create_variation`, `images.edit`, `images.generate`, `live.connect`, `live.create`, `live.forks.connect`, `live.sessions.accept`, `live.sessions.fork`, `live.sideband.connect`, `realtime.calls.accept`, `realtime.calls.create`, `realtime.connect`, `realtime.translations.connect`, `responses.compact`, `responses.connect`, `videos.create`, `videos.create_and_poll`, `videos.create_character`, `videos.edit`, `videos.extend`, `videos.remix`
|
|
196
196
|
- **Anthropic** (`Anthropic` and `AsyncAnthropic` clients): `beta.deployments.run`, `beta.dreams.create`, `beta.environments.work.poller`, `beta.environments.work.worker`, `beta.messages.batches.create`, `beta.sessions.create`, `beta.sessions.events.send`, `beta.sessions.events.stream`, `beta.sessions.events.tool_runner`, `beta.sessions.threads.events.stream`, `messages.batches.create`
|
|
197
197
|
- **Google Gemini** (`google.genai` `Client` and `Client.aio`): `batches.create`, `batches.create_embeddings`, `caches.create`, `caches.update`, `file_search_stores.import_file`, `file_search_stores.upload_to_file_search_store`, `interactions.create`, `live.connect`, `live.music.connect`, `models.edit_image`, `models.recontext_image`, `models.segment_image`, `models.upscale_image`, `triggers.run`, `tunings.tune`
|
|
198
198
|
- **Google Vertex AI** (`vertexai` classes): `batch_prediction.BatchPredictionJob.submit`, `caching.CachedContent.create`, `caching.CachedContent.update`, `language_models.ChatModel.tune_model`, `language_models.ChatModel.tune_model_rlhf`, `language_models.ChatSession.send_message`, `language_models.ChatSession.send_message_async`, `language_models.ChatSession.send_message_streaming`, `language_models.ChatSession.send_message_streaming_async`, `language_models.CodeChatModel.tune_model`, `language_models.CodeChatSession.send_message`, `language_models.CodeChatSession.send_message_async`, `language_models.CodeChatSession.send_message_streaming`, `language_models.CodeChatSession.send_message_streaming_async`, `language_models.CodeGenerationModel.batch_predict`, `language_models.CodeGenerationModel.predict`, `language_models.CodeGenerationModel.predict_async`, `language_models.CodeGenerationModel.predict_streaming`, `language_models.CodeGenerationModel.predict_streaming_async`, `language_models.CodeGenerationModel.tune_model`, `language_models.TextEmbeddingModel.batch_predict`, `language_models.TextEmbeddingModel.deploy_tuned_model`, `language_models.TextEmbeddingModel.tune_model`, `language_models.TextGenerationModel.batch_predict`, `language_models.TextGenerationModel.predict`, `language_models.TextGenerationModel.predict_async`, `language_models.TextGenerationModel.predict_streaming`, `language_models.TextGenerationModel.predict_streaming_async`, `language_models.TextGenerationModel.tune_model`, `language_models.TextGenerationModel.tune_model_rlhf`, `preview.caching.CachedContent.create`, `preview.caching.CachedContent.update`, `preview.vision_models.ImageCaptioningModel.get_captions`, `preview.vision_models.ImageGenerationModel.upscale_image`, `preview.vision_models.ImageQnAModel.ask_question`, `preview.vision_models.ImageSegmentationModel.segment_image`, `preview.vision_models.ImageTextModel.ask_question`, `preview.vision_models.ImageTextModel.get_captions`, `preview.vision_models.MultiModalEmbeddingModel.get_embeddings`, `preview.vision_models.WatermarkVerificationModel.verify_image`, `tuning.sft.train`, `vision_models.ImageCaptioningModel.get_captions`, `vision_models.ImageGenerationModel.upscale_image`, `vision_models.ImageQnAModel.ask_question`, `vision_models.ImageTextModel.ask_question`, `vision_models.ImageTextModel.get_captions`, `vision_models.MultiModalEmbeddingModel.get_embeddings`
|
|
@@ -849,7 +849,7 @@ async def main():
|
|
|
849
849
|
asyncio.run(main())
|
|
850
850
|
```
|
|
851
851
|
|
|
852
|
-
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse` and each
|
|
852
|
+
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse`, each step of `tool_runner` and, on the Anthropic client, each turn of a `thread`), sync and async, including their `with_raw_response` and `with_streaming_response` forms, on the Anthropic, Bedrock and Vertex AI clients. `tool_runner` and `thread` do not accept `usage_metadata=`; set metadata for their calls with `@revenium_metadata` (see [Decorator Support](#decorator-support)). Calls listed under [Calls the SDK does not meter](#calls-the-sdk-does-not-meter), message batches among them, are not metered.
|
|
853
853
|
|
|
854
854
|
A call the provider refuses sends one record with stop reason `ERROR`, zero tokens and the error's class and HTTP status as its error reason (plus the provider's redacted error message when both `REVENIUM_CAPTURE_ERROR_MESSAGES` and `REVENIUM_CAPTURE_PROMPTS` are `true`; a provider's error can quote the request, so it is never sent with prompt capture off; at most 255 characters, keeping its "(HTTP <status>)" ending), as does a `create(stream=True)` stream that breaks off part-way. The HTTP status is also sent as the record's error code; a call that got no HTTP error answer (a connection error, a timeout, or a stream that broke off) has none. Your code still gets the original exception. Calls served through the Bedrock fast path, and calls a spending cap refuses, send no error record.
|
|
855
855
|
|
|
@@ -1197,6 +1197,24 @@ the others to be recorded. A malformed header value, such as a quality score of
|
|
|
1197
1197
|
`high` or an effort of `very high`, is left off the record with a WARNING in the
|
|
1198
1198
|
gateway log; the call is still recorded.
|
|
1199
1199
|
|
|
1200
|
+
**LiteLLM identities.** Below those sources, the guardrail also reads what
|
|
1201
|
+
LiteLLM itself resolved for the call:
|
|
1202
|
+
|
|
1203
|
+
| Revenium field | LiteLLM source | Used when |
|
|
1204
|
+
|---|---|---|
|
|
1205
|
+
| `subscriber.id` | The end user: the request's `user`, or the header named by LiteLLM's `user_header_name` | Nothing else named a subscriber id or email. A key with an owner keeps attributing to the owner, and the end user is never paired with the owner's email. |
|
|
1206
|
+
| `organizationName` | The alias of the LiteLLM organization the key belongs to | No header, team alias or `revenium_organization_name` key metadata named one |
|
|
1207
|
+
| `productName` | The alias of the LiteLLM project the key belongs to | No `x-revenium-product-name` header or `product_name:` tag |
|
|
1208
|
+
|
|
1209
|
+
LiteLLM organizations and projects need a LiteLLM enterprise license; without
|
|
1210
|
+
one these fallbacks never apply. Every record also carries `reasoningTokenCount`
|
|
1211
|
+
from LiteLLM's normalized usage (it is already inside the output count, which is
|
|
1212
|
+
unchanged) and `modelHost`, the LiteLLM provider that served the call (`openai`,
|
|
1213
|
+
`azure`, `bedrock`, ...; LiteLLM's `vertex_ai` is sent as `vertex` and `azure_ai` as
|
|
1214
|
+
`foundry`, the names coding-assistant records use). `provider` stays `LITELLM`, so
|
|
1215
|
+
provider-scoped spending caps and existing breakdowns are unaffected. The
|
|
1216
|
+
deprecated callback reads none of these.
|
|
1217
|
+
|
|
1200
1218
|
**Trace id.** The guardrail takes a record's trace id from `x-revenium-trace-id`.
|
|
1201
1219
|
Without that header it takes the LiteLLM session id the caller sent: the
|
|
1202
1220
|
`x-litellm-session-id` or `x-litellm-trace-id` header, or `litellm_session_id` or
|
|
@@ -1220,6 +1238,58 @@ never named after its LiteLLM session. A name over 256 characters is cut to 256.
|
|
|
1220
1238
|
with no LiteLLM session and no name has no trace name. The deprecated callback
|
|
1221
1239
|
reads only the header and `REVENIUM_TRACE_NAME`.
|
|
1222
1240
|
|
|
1241
|
+
**Attribution from LiteLLM tags.** Six fields can also come from LiteLLM's own tags,
|
|
1242
|
+
so a gateway caller needs no Revenium header to set them. A tag `<field>:<value>`
|
|
1243
|
+
fills the field when its `x-revenium-*` header is absent or blank:
|
|
1244
|
+
|
|
1245
|
+
| Tag | Fills | Instead of |
|
|
1246
|
+
|-----|-------|------------|
|
|
1247
|
+
| `trace_type:<value>` | trace type | `x-revenium-trace-type` |
|
|
1248
|
+
| `task_type:<value>` | task type | `x-revenium-task-type` |
|
|
1249
|
+
| `agent:<value>` | agent | `x-revenium-agent` |
|
|
1250
|
+
| `product_name:<value>` | product name | `x-revenium-product-name` |
|
|
1251
|
+
| `environment:<value>` | environment | `x-revenium-environment` |
|
|
1252
|
+
| `transaction_name:<value>` | transaction name | `x-revenium-transaction-name` |
|
|
1253
|
+
|
|
1254
|
+
The guardrail reads the tags LiteLLM has merged by the time it runs: the caller's
|
|
1255
|
+
(body `metadata.tags` or root `tags`, or the `x-litellm-tags` header) and those of
|
|
1256
|
+
the calling virtual key, team and project. For each field the header wins, then the
|
|
1257
|
+
caller's tag, then a key, team or project tag, then the gateway setting such as
|
|
1258
|
+
`REVENIUM_TRACE_TYPE`. Two different values at one level are ignored and the next
|
|
1259
|
+
level decides; a WARNING names the field once per request. Values are validated as
|
|
1260
|
+
the header's are. A key or team tag is a per-customer default; a request tag
|
|
1261
|
+
describes one call:
|
|
1262
|
+
|
|
1263
|
+
```python
|
|
1264
|
+
client.chat.completions.create(
|
|
1265
|
+
model="gpt-4o-mini",
|
|
1266
|
+
messages=[...],
|
|
1267
|
+
extra_body={"metadata": {"tags": ["trace_type:research-brief", "agent:research-planner"]}},
|
|
1268
|
+
)
|
|
1269
|
+
```
|
|
1270
|
+
|
|
1271
|
+
A tag carries the same weight as its header. Agent, task type and product also
|
|
1272
|
+
decide which spending caps apply, so a `product_name:`, `agent:` or `task_type:` tag
|
|
1273
|
+
can move a call into or out of a cap filtered on that field, and a caller's request
|
|
1274
|
+
tag overrides a key's or team's. On upgrade, a key or team that already carries tags
|
|
1275
|
+
of this form starts setting these fields, for the record and the cap check alike;
|
|
1276
|
+
review existing tags before upgrading if you filter caps on agent, task type or
|
|
1277
|
+
product.
|
|
1278
|
+
|
|
1279
|
+
Tags reach the pre-call cap check only when the guardrail writes the record itself
|
|
1280
|
+
(`default_on: true` with `post_call` in its mode), so a call is never checked under a
|
|
1281
|
+
value its record lacks. The deprecated callback reads headers only.
|
|
1282
|
+
|
|
1283
|
+
LiteLLM records which tags the request itself sent (`caller_tags`), so a caller's tag
|
|
1284
|
+
stays the caller's even when the key or team carries the same string. Body
|
|
1285
|
+
`metadata.tags` on `/chat/completions` is the exception: LiteLLM leaves it out of
|
|
1286
|
+
that record, so there a tag that is also a key or team tag counts as the key's or
|
|
1287
|
+
team's.
|
|
1288
|
+
|
|
1289
|
+
Tags never set the organization, subscriber, subscription, agentic job or trace id:
|
|
1290
|
+
those are billing identity and trace grouping, and keep coming from headers and the
|
|
1291
|
+
virtual key.
|
|
1292
|
+
|
|
1223
1293
|
**Organization caps and the organization header.** By default the
|
|
1224
1294
|
`x-revenium-organization-name` header (and the retired `x-revenium-organization-id`)
|
|
1225
1295
|
wins over the calling virtual key's team alias, for the budget check and the record
|
|
@@ -2241,7 +2311,7 @@ pip install 'revenium-python-sdk[openai]'
|
|
|
2241
2311
|
REVENIUM_CIRCUIT_BREAKER_ENABLED=true
|
|
2242
2312
|
REVENIUM_METERING_API_KEY=hak_your_key_here
|
|
2243
2313
|
REVENIUM_TEAM_ID=your_hashed_team_id
|
|
2244
|
-
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
2314
|
+
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional: this is the default
|
|
2245
2315
|
```
|
|
2246
2316
|
|
|
2247
2317
|
### Environment Variables
|
|
@@ -2251,7 +2321,7 @@ REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
|
2251
2321
|
| `REVENIUM_CIRCUIT_BREAKER_ENABLED` | `false` | Master switch. `true` / `1` / `yes` / `on` to enable. |
|
|
2252
2322
|
| `REVENIUM_BYPASS` | `false` | When `true`, every `check_enforcement` call short-circuits to a no-op. Useful for incident response. |
|
|
2253
2323
|
| `REVENIUM_TEAM_ID` | — | Hashed team ID. Path component on rule fetches; required when the breaker is enabled. |
|
|
2254
|
-
| `REVENIUM_ENFORCEMENT_BASE_URL` |
|
|
2324
|
+
| `REVENIUM_ENFORCEMENT_BASE_URL` | `https://api.revenium.ai/profitstream` | Base URL for the enforcement API, taken as given. When unset it is the address outcomes and jobs use (`REVENIUM_PROFITSTREAM_BASE_URL`, else the Revenium address `REVENIUM_METERING_BASE_URL` or `initialize_metering(base_url=...)` sits under, without its `/meter` segment) plus `/profitstream`. A rules fetch that answers 404 is retried once at the same address with `/profitstream` removed, or added if it was missing, and the address that answers with rules is used from then on, so a self-hosted server that serves the API at the root, or a value set without `/profitstream`, still reaches the rules. |
|
|
2255
2325
|
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `60` | Background poll interval for rule refreshes. |
|
|
2256
2326
|
| `REVENIUM_CB_FAIL_MODE` | `open` | `open` (default) lets calls through when no cache exists; `closed` raises `BudgetExceededError` until rules are loaded. |
|
|
2257
2327
|
| `REVENIUM_CACHE_DIR` | unset | When set, the rule cache is mirrored to `<dir>/revenium_enforcement_rules.<scope>.json` so a restarted process doesn't fail-closed on the very first call. `<scope>` is a short hash of the enforcement base URL and `REVENIUM_TEAM_ID`, so services for different teams can share one folder and each loads only its own rules. Nothing is saved without `REVENIUM_TEAM_ID`. |
|
|
@@ -2287,7 +2357,7 @@ When the circuit breaker is disabled, it is a no-op. When enabled:
|
|
|
2287
2357
|
1. A daemon thread (`revenium-enforcement-poll`) starts on first use.
|
|
2288
2358
|
2. It polls `GET {REVENIUM_ENFORCEMENT_BASE_URL}/v2/api/ai/enforcement-rules/{REVENIUM_TEAM_ID}` every `REVENIUM_CB_POLL_INTERVAL_SECONDS` with the `x-api-key` header.
|
|
2289
2359
|
3. Rules are cached in-process. The pre-call check only reads that cache: it never makes a network request and never sleeps, so it is safe on an event loop (LiteLLM proxy, `AsyncOpenAI`). When the cache is more than 120 s old the check wakes the poller to refresh it early and carries on with the rules it has. Only one refresh runs at a time per process.
|
|
2290
|
-
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS`, however much traffic arrives meanwhile.
|
|
2360
|
+
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS`, however much traffic arrives meanwhile. A 4xx answer the SDK does not retry (any but 408 and 429) is logged at WARNING with the URL and the status, so a wrong address (404) or a wrong key or team id (401, 403) is visible without debug logging.
|
|
2291
2361
|
5. `204 No Content` is treated as "no rules configured" — the cache is cleared.
|
|
2292
2362
|
|
|
2293
2363
|
A rule with filters (a cap scoped to an agent, model, provider, subscriber, credential, task type, product or organization) blocks or warns only the calls that match all of them; a filter the SDK cannot evaluate, such as the model of a LiteLLM proxy request, keeps the rule applying.
|
|
@@ -76,7 +76,7 @@ Some billable provider calls are not metered at all: see [Calls the SDK does not
|
|
|
76
76
|
The calls below are billed by the provider (some, such as batch and fine-tuning jobs, when the job runs rather than when you call them), but the SDK records no usage for them: nothing reaches Revenium and nothing is logged. On every build a test compares this list with the SDK's inventory of the public calls of the OpenAI, Anthropic, Perplexity, Gemini, Ollama, LiteLLM and fal clients, boto3's Bedrock Runtime client, and the Vertex AI model, session, batch prediction, tuning and context-cache classes, on the provider versions the SDK is tested on. Calls reached another way, for example LangChain classes that call their own vendor SDK, are outside that check, except LangChain embeddings, listed by hand below; wrapping one with `wrap()` logs a warning. Sync and async forms are folded into one name, and the `with_raw_response` and `with_streaming_response` forms of a listed call are not metered either. To record any of these calls, wrap it with `@meter_tool` (see [Tool Metering](#tool-metering)).
|
|
77
77
|
|
|
78
78
|
<!-- not-metered:start -->
|
|
79
|
-
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.create`, `beta.agents.sessions.create`, `beta.agents.sessions.events.create`, `beta.agents.sessions.events.stream`, `beta.agents.sessions.stream`, `beta.agents.sessions.update`, `beta.chatkit.sessions.create`, `beta.realtime.connect`, `beta.responses.compact`, `beta.responses.connect`, `beta.responses.create`, `beta.threads.create_and_run`, `beta.threads.create_and_run_poll`, `beta.threads.create_and_run_stream`, `beta.threads.runs.create`, `beta.threads.runs.create_and_poll`, `beta.threads.runs.create_and_stream`, `beta.threads.runs.stream`, `beta.threads.runs.submit_tool_outputs`, `beta.threads.runs.submit_tool_outputs_and_poll`, `beta.threads.runs.submit_tool_outputs_stream`, `completions.create`, `decisions.create`, `evals.runs.create`, `fine_tuning.alpha.graders.run`, `fine_tuning.jobs.create`, `images.create_variation`, `images.edit`, `images.generate`, `live.connect`, `live.create`, `live.forks.connect`, `live.sessions.accept`, `live.sessions.fork`, `live.sideband.connect`, `realtime.calls.accept`, `realtime.calls.create`, `realtime.connect`, `realtime.translations.connect`, `responses.compact`, `responses.connect`, `videos.create`, `videos.create_and_poll`, `videos.create_character`, `videos.edit`, `videos.extend`, `videos.remix`
|
|
79
|
+
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.create`, `beta.agents.automations.create`, `beta.agents.automations.invocations.create`, `beta.agents.sessions.create`, `beta.agents.sessions.events.create`, `beta.agents.sessions.events.stream`, `beta.agents.sessions.stream`, `beta.agents.sessions.update`, `beta.chatkit.sessions.create`, `beta.realtime.connect`, `beta.responses.compact`, `beta.responses.connect`, `beta.responses.create`, `beta.threads.create_and_run`, `beta.threads.create_and_run_poll`, `beta.threads.create_and_run_stream`, `beta.threads.runs.create`, `beta.threads.runs.create_and_poll`, `beta.threads.runs.create_and_stream`, `beta.threads.runs.stream`, `beta.threads.runs.submit_tool_outputs`, `beta.threads.runs.submit_tool_outputs_and_poll`, `beta.threads.runs.submit_tool_outputs_stream`, `completions.create`, `decisions.create`, `evals.runs.create`, `fine_tuning.alpha.graders.run`, `fine_tuning.jobs.create`, `images.create_variation`, `images.edit`, `images.generate`, `live.connect`, `live.create`, `live.forks.connect`, `live.sessions.accept`, `live.sessions.fork`, `live.sideband.connect`, `realtime.calls.accept`, `realtime.calls.create`, `realtime.connect`, `realtime.translations.connect`, `responses.compact`, `responses.connect`, `videos.create`, `videos.create_and_poll`, `videos.create_character`, `videos.edit`, `videos.extend`, `videos.remix`
|
|
80
80
|
- **Anthropic** (`Anthropic` and `AsyncAnthropic` clients): `beta.deployments.run`, `beta.dreams.create`, `beta.environments.work.poller`, `beta.environments.work.worker`, `beta.messages.batches.create`, `beta.sessions.create`, `beta.sessions.events.send`, `beta.sessions.events.stream`, `beta.sessions.events.tool_runner`, `beta.sessions.threads.events.stream`, `messages.batches.create`
|
|
81
81
|
- **Google Gemini** (`google.genai` `Client` and `Client.aio`): `batches.create`, `batches.create_embeddings`, `caches.create`, `caches.update`, `file_search_stores.import_file`, `file_search_stores.upload_to_file_search_store`, `interactions.create`, `live.connect`, `live.music.connect`, `models.edit_image`, `models.recontext_image`, `models.segment_image`, `models.upscale_image`, `triggers.run`, `tunings.tune`
|
|
82
82
|
- **Google Vertex AI** (`vertexai` classes): `batch_prediction.BatchPredictionJob.submit`, `caching.CachedContent.create`, `caching.CachedContent.update`, `language_models.ChatModel.tune_model`, `language_models.ChatModel.tune_model_rlhf`, `language_models.ChatSession.send_message`, `language_models.ChatSession.send_message_async`, `language_models.ChatSession.send_message_streaming`, `language_models.ChatSession.send_message_streaming_async`, `language_models.CodeChatModel.tune_model`, `language_models.CodeChatSession.send_message`, `language_models.CodeChatSession.send_message_async`, `language_models.CodeChatSession.send_message_streaming`, `language_models.CodeChatSession.send_message_streaming_async`, `language_models.CodeGenerationModel.batch_predict`, `language_models.CodeGenerationModel.predict`, `language_models.CodeGenerationModel.predict_async`, `language_models.CodeGenerationModel.predict_streaming`, `language_models.CodeGenerationModel.predict_streaming_async`, `language_models.CodeGenerationModel.tune_model`, `language_models.TextEmbeddingModel.batch_predict`, `language_models.TextEmbeddingModel.deploy_tuned_model`, `language_models.TextEmbeddingModel.tune_model`, `language_models.TextGenerationModel.batch_predict`, `language_models.TextGenerationModel.predict`, `language_models.TextGenerationModel.predict_async`, `language_models.TextGenerationModel.predict_streaming`, `language_models.TextGenerationModel.predict_streaming_async`, `language_models.TextGenerationModel.tune_model`, `language_models.TextGenerationModel.tune_model_rlhf`, `preview.caching.CachedContent.create`, `preview.caching.CachedContent.update`, `preview.vision_models.ImageCaptioningModel.get_captions`, `preview.vision_models.ImageGenerationModel.upscale_image`, `preview.vision_models.ImageQnAModel.ask_question`, `preview.vision_models.ImageSegmentationModel.segment_image`, `preview.vision_models.ImageTextModel.ask_question`, `preview.vision_models.ImageTextModel.get_captions`, `preview.vision_models.MultiModalEmbeddingModel.get_embeddings`, `preview.vision_models.WatermarkVerificationModel.verify_image`, `tuning.sft.train`, `vision_models.ImageCaptioningModel.get_captions`, `vision_models.ImageGenerationModel.upscale_image`, `vision_models.ImageQnAModel.ask_question`, `vision_models.ImageTextModel.ask_question`, `vision_models.ImageTextModel.get_captions`, `vision_models.MultiModalEmbeddingModel.get_embeddings`
|
|
@@ -733,7 +733,7 @@ async def main():
|
|
|
733
733
|
asyncio.run(main())
|
|
734
734
|
```
|
|
735
735
|
|
|
736
|
-
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse` and each
|
|
736
|
+
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse`, each step of `tool_runner` and, on the Anthropic client, each turn of a `thread`), sync and async, including their `with_raw_response` and `with_streaming_response` forms, on the Anthropic, Bedrock and Vertex AI clients. `tool_runner` and `thread` do not accept `usage_metadata=`; set metadata for their calls with `@revenium_metadata` (see [Decorator Support](#decorator-support)). Calls listed under [Calls the SDK does not meter](#calls-the-sdk-does-not-meter), message batches among them, are not metered.
|
|
737
737
|
|
|
738
738
|
A call the provider refuses sends one record with stop reason `ERROR`, zero tokens and the error's class and HTTP status as its error reason (plus the provider's redacted error message when both `REVENIUM_CAPTURE_ERROR_MESSAGES` and `REVENIUM_CAPTURE_PROMPTS` are `true`; a provider's error can quote the request, so it is never sent with prompt capture off; at most 255 characters, keeping its "(HTTP <status>)" ending), as does a `create(stream=True)` stream that breaks off part-way. The HTTP status is also sent as the record's error code; a call that got no HTTP error answer (a connection error, a timeout, or a stream that broke off) has none. Your code still gets the original exception. Calls served through the Bedrock fast path, and calls a spending cap refuses, send no error record.
|
|
739
739
|
|
|
@@ -1081,6 +1081,24 @@ the others to be recorded. A malformed header value, such as a quality score of
|
|
|
1081
1081
|
`high` or an effort of `very high`, is left off the record with a WARNING in the
|
|
1082
1082
|
gateway log; the call is still recorded.
|
|
1083
1083
|
|
|
1084
|
+
**LiteLLM identities.** Below those sources, the guardrail also reads what
|
|
1085
|
+
LiteLLM itself resolved for the call:
|
|
1086
|
+
|
|
1087
|
+
| Revenium field | LiteLLM source | Used when |
|
|
1088
|
+
|---|---|---|
|
|
1089
|
+
| `subscriber.id` | The end user: the request's `user`, or the header named by LiteLLM's `user_header_name` | Nothing else named a subscriber id or email. A key with an owner keeps attributing to the owner, and the end user is never paired with the owner's email. |
|
|
1090
|
+
| `organizationName` | The alias of the LiteLLM organization the key belongs to | No header, team alias or `revenium_organization_name` key metadata named one |
|
|
1091
|
+
| `productName` | The alias of the LiteLLM project the key belongs to | No `x-revenium-product-name` header or `product_name:` tag |
|
|
1092
|
+
|
|
1093
|
+
LiteLLM organizations and projects need a LiteLLM enterprise license; without
|
|
1094
|
+
one these fallbacks never apply. Every record also carries `reasoningTokenCount`
|
|
1095
|
+
from LiteLLM's normalized usage (it is already inside the output count, which is
|
|
1096
|
+
unchanged) and `modelHost`, the LiteLLM provider that served the call (`openai`,
|
|
1097
|
+
`azure`, `bedrock`, ...; LiteLLM's `vertex_ai` is sent as `vertex` and `azure_ai` as
|
|
1098
|
+
`foundry`, the names coding-assistant records use). `provider` stays `LITELLM`, so
|
|
1099
|
+
provider-scoped spending caps and existing breakdowns are unaffected. The
|
|
1100
|
+
deprecated callback reads none of these.
|
|
1101
|
+
|
|
1084
1102
|
**Trace id.** The guardrail takes a record's trace id from `x-revenium-trace-id`.
|
|
1085
1103
|
Without that header it takes the LiteLLM session id the caller sent: the
|
|
1086
1104
|
`x-litellm-session-id` or `x-litellm-trace-id` header, or `litellm_session_id` or
|
|
@@ -1104,6 +1122,58 @@ never named after its LiteLLM session. A name over 256 characters is cut to 256.
|
|
|
1104
1122
|
with no LiteLLM session and no name has no trace name. The deprecated callback
|
|
1105
1123
|
reads only the header and `REVENIUM_TRACE_NAME`.
|
|
1106
1124
|
|
|
1125
|
+
**Attribution from LiteLLM tags.** Six fields can also come from LiteLLM's own tags,
|
|
1126
|
+
so a gateway caller needs no Revenium header to set them. A tag `<field>:<value>`
|
|
1127
|
+
fills the field when its `x-revenium-*` header is absent or blank:
|
|
1128
|
+
|
|
1129
|
+
| Tag | Fills | Instead of |
|
|
1130
|
+
|-----|-------|------------|
|
|
1131
|
+
| `trace_type:<value>` | trace type | `x-revenium-trace-type` |
|
|
1132
|
+
| `task_type:<value>` | task type | `x-revenium-task-type` |
|
|
1133
|
+
| `agent:<value>` | agent | `x-revenium-agent` |
|
|
1134
|
+
| `product_name:<value>` | product name | `x-revenium-product-name` |
|
|
1135
|
+
| `environment:<value>` | environment | `x-revenium-environment` |
|
|
1136
|
+
| `transaction_name:<value>` | transaction name | `x-revenium-transaction-name` |
|
|
1137
|
+
|
|
1138
|
+
The guardrail reads the tags LiteLLM has merged by the time it runs: the caller's
|
|
1139
|
+
(body `metadata.tags` or root `tags`, or the `x-litellm-tags` header) and those of
|
|
1140
|
+
the calling virtual key, team and project. For each field the header wins, then the
|
|
1141
|
+
caller's tag, then a key, team or project tag, then the gateway setting such as
|
|
1142
|
+
`REVENIUM_TRACE_TYPE`. Two different values at one level are ignored and the next
|
|
1143
|
+
level decides; a WARNING names the field once per request. Values are validated as
|
|
1144
|
+
the header's are. A key or team tag is a per-customer default; a request tag
|
|
1145
|
+
describes one call:
|
|
1146
|
+
|
|
1147
|
+
```python
|
|
1148
|
+
client.chat.completions.create(
|
|
1149
|
+
model="gpt-4o-mini",
|
|
1150
|
+
messages=[...],
|
|
1151
|
+
extra_body={"metadata": {"tags": ["trace_type:research-brief", "agent:research-planner"]}},
|
|
1152
|
+
)
|
|
1153
|
+
```
|
|
1154
|
+
|
|
1155
|
+
A tag carries the same weight as its header. Agent, task type and product also
|
|
1156
|
+
decide which spending caps apply, so a `product_name:`, `agent:` or `task_type:` tag
|
|
1157
|
+
can move a call into or out of a cap filtered on that field, and a caller's request
|
|
1158
|
+
tag overrides a key's or team's. On upgrade, a key or team that already carries tags
|
|
1159
|
+
of this form starts setting these fields, for the record and the cap check alike;
|
|
1160
|
+
review existing tags before upgrading if you filter caps on agent, task type or
|
|
1161
|
+
product.
|
|
1162
|
+
|
|
1163
|
+
Tags reach the pre-call cap check only when the guardrail writes the record itself
|
|
1164
|
+
(`default_on: true` with `post_call` in its mode), so a call is never checked under a
|
|
1165
|
+
value its record lacks. The deprecated callback reads headers only.
|
|
1166
|
+
|
|
1167
|
+
LiteLLM records which tags the request itself sent (`caller_tags`), so a caller's tag
|
|
1168
|
+
stays the caller's even when the key or team carries the same string. Body
|
|
1169
|
+
`metadata.tags` on `/chat/completions` is the exception: LiteLLM leaves it out of
|
|
1170
|
+
that record, so there a tag that is also a key or team tag counts as the key's or
|
|
1171
|
+
team's.
|
|
1172
|
+
|
|
1173
|
+
Tags never set the organization, subscriber, subscription, agentic job or trace id:
|
|
1174
|
+
those are billing identity and trace grouping, and keep coming from headers and the
|
|
1175
|
+
virtual key.
|
|
1176
|
+
|
|
1107
1177
|
**Organization caps and the organization header.** By default the
|
|
1108
1178
|
`x-revenium-organization-name` header (and the retired `x-revenium-organization-id`)
|
|
1109
1179
|
wins over the calling virtual key's team alias, for the budget check and the record
|
|
@@ -2125,7 +2195,7 @@ pip install 'revenium-python-sdk[openai]'
|
|
|
2125
2195
|
REVENIUM_CIRCUIT_BREAKER_ENABLED=true
|
|
2126
2196
|
REVENIUM_METERING_API_KEY=hak_your_key_here
|
|
2127
2197
|
REVENIUM_TEAM_ID=your_hashed_team_id
|
|
2128
|
-
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
2198
|
+
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional: this is the default
|
|
2129
2199
|
```
|
|
2130
2200
|
|
|
2131
2201
|
### Environment Variables
|
|
@@ -2135,7 +2205,7 @@ REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
|
2135
2205
|
| `REVENIUM_CIRCUIT_BREAKER_ENABLED` | `false` | Master switch. `true` / `1` / `yes` / `on` to enable. |
|
|
2136
2206
|
| `REVENIUM_BYPASS` | `false` | When `true`, every `check_enforcement` call short-circuits to a no-op. Useful for incident response. |
|
|
2137
2207
|
| `REVENIUM_TEAM_ID` | — | Hashed team ID. Path component on rule fetches; required when the breaker is enabled. |
|
|
2138
|
-
| `REVENIUM_ENFORCEMENT_BASE_URL` |
|
|
2208
|
+
| `REVENIUM_ENFORCEMENT_BASE_URL` | `https://api.revenium.ai/profitstream` | Base URL for the enforcement API, taken as given. When unset it is the address outcomes and jobs use (`REVENIUM_PROFITSTREAM_BASE_URL`, else the Revenium address `REVENIUM_METERING_BASE_URL` or `initialize_metering(base_url=...)` sits under, without its `/meter` segment) plus `/profitstream`. A rules fetch that answers 404 is retried once at the same address with `/profitstream` removed, or added if it was missing, and the address that answers with rules is used from then on, so a self-hosted server that serves the API at the root, or a value set without `/profitstream`, still reaches the rules. |
|
|
2139
2209
|
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `60` | Background poll interval for rule refreshes. |
|
|
2140
2210
|
| `REVENIUM_CB_FAIL_MODE` | `open` | `open` (default) lets calls through when no cache exists; `closed` raises `BudgetExceededError` until rules are loaded. |
|
|
2141
2211
|
| `REVENIUM_CACHE_DIR` | unset | When set, the rule cache is mirrored to `<dir>/revenium_enforcement_rules.<scope>.json` so a restarted process doesn't fail-closed on the very first call. `<scope>` is a short hash of the enforcement base URL and `REVENIUM_TEAM_ID`, so services for different teams can share one folder and each loads only its own rules. Nothing is saved without `REVENIUM_TEAM_ID`. |
|
|
@@ -2171,7 +2241,7 @@ When the circuit breaker is disabled, it is a no-op. When enabled:
|
|
|
2171
2241
|
1. A daemon thread (`revenium-enforcement-poll`) starts on first use.
|
|
2172
2242
|
2. It polls `GET {REVENIUM_ENFORCEMENT_BASE_URL}/v2/api/ai/enforcement-rules/{REVENIUM_TEAM_ID}` every `REVENIUM_CB_POLL_INTERVAL_SECONDS` with the `x-api-key` header.
|
|
2173
2243
|
3. Rules are cached in-process. The pre-call check only reads that cache: it never makes a network request and never sleeps, so it is safe on an event loop (LiteLLM proxy, `AsyncOpenAI`). When the cache is more than 120 s old the check wakes the poller to refresh it early and carries on with the rules it has. Only one refresh runs at a time per process.
|
|
2174
|
-
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS`, however much traffic arrives meanwhile.
|
|
2244
|
+
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS`, however much traffic arrives meanwhile. A 4xx answer the SDK does not retry (any but 408 and 429) is logged at WARNING with the URL and the status, so a wrong address (404) or a wrong key or team id (401, 403) is visible without debug logging.
|
|
2175
2245
|
5. `204 No Content` is treated as "no rules configured" — the cache is cleared.
|
|
2176
2246
|
|
|
2177
2247
|
A rule with filters (a cap scoped to an agent, model, provider, subscriber, credential, task type, product or organization) blocks or warns only the calls that match all of them; a filter the SDK cannot evaluate, such as the model of a LiteLLM proxy request, keeps the rule applying.
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "revenium-python-sdk"
|
|
7
|
-
version = "0.
|
|
7
|
+
version = "0.13.0"
|
|
8
8
|
description = "The official Revenium Python SDK — unified AI metering middleware for OpenAI, Anthropic, Google, Ollama, LiteLLM, Perplexity, and fal.ai."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
requires-python = ">=3.8"
|
{revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/cache_tokens.py
RENAMED
|
@@ -238,6 +238,15 @@ def _upstream_provider(sources) -> Optional[str]:
|
|
|
238
238
|
return None
|
|
239
239
|
|
|
240
240
|
|
|
241
|
+
def upstream_provider(*sources) -> Optional[str]:
|
|
242
|
+
"""The LiteLLM provider that served the call, lower-cased, or None.
|
|
243
|
+
|
|
244
|
+
``sources`` are LiteLLM-populated surfaces in lookup order, as for
|
|
245
|
+
``prompt_tokens_fold_cache``; the first that names a provider decides.
|
|
246
|
+
"""
|
|
247
|
+
return _upstream_provider(sources)
|
|
248
|
+
|
|
249
|
+
|
|
241
250
|
def _provider_folds_cache(provider: str, sources) -> bool:
|
|
242
251
|
if provider in _CACHE_FOLDING_PROVIDERS:
|
|
243
252
|
return True
|
{revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/enforcement.py
RENAMED
|
@@ -69,13 +69,14 @@ import threading
|
|
|
69
69
|
import time
|
|
70
70
|
from email.utils import parsedate_to_datetime
|
|
71
71
|
from typing import Dict, List, NamedTuple, Optional, Set, Tuple
|
|
72
|
-
from urllib.parse import quote,
|
|
72
|
+
from urllib.parse import quote, urlsplit, urlunsplit
|
|
73
73
|
|
|
74
74
|
import httpx
|
|
75
75
|
|
|
76
76
|
from .call_ownership import peek_callback_metadata
|
|
77
77
|
from .closable_scope import ClosableScope
|
|
78
|
-
from .config import Config, env_choice, env_flag_enabled
|
|
78
|
+
from .config import Config, env_choice, env_flag_enabled, get_profitstream_base_url
|
|
79
|
+
from .log_sanitize import address_without_credentials
|
|
79
80
|
from .context import merge_metadata, overlay_metadata
|
|
80
81
|
from .exceptions import BudgetExceededError
|
|
81
82
|
from .fields import extract_common_metadata, extract_organization_name, extract_product_name
|
|
@@ -462,21 +463,73 @@ def _process_metering_url() -> Optional[str]:
|
|
|
462
463
|
return _valid_base_url(str(client.base_url)) if client is not None else None
|
|
463
464
|
|
|
464
465
|
|
|
466
|
+
_PROFITSTREAM_PATH = "/profitstream"
|
|
467
|
+
|
|
468
|
+
# Configured enforcement address -> the sibling address (with or without
|
|
469
|
+
# ``/profitstream``) that served the rules after the configured one answered
|
|
470
|
+
# 404. The cache owner and the snapshot scope keep the configured address;
|
|
471
|
+
# only the request URL moves (BACK-5240).
|
|
472
|
+
_recovered_enforcement_addresses: Dict[str, str] = {}
|
|
473
|
+
|
|
474
|
+
|
|
465
475
|
def _get_enforcement_base_url() -> str:
|
|
466
|
-
"""
|
|
476
|
+
"""The enforcement address as configured: the one the cache belongs to.
|
|
467
477
|
|
|
468
|
-
|
|
469
|
-
(``http://localhost:8080/profitstream``) survives intact.
|
|
470
|
-
the
|
|
471
|
-
|
|
478
|
+
``REVENIUM_ENFORCEMENT_BASE_URL`` is taken as given, so a context path
|
|
479
|
+
(``http://localhost:8080/profitstream``) survives intact. Otherwise it is
|
|
480
|
+
the address outcomes and jobs use (``REVENIUM_PROFITSTREAM_BASE_URL``,
|
|
481
|
+
else the Revenium address the metering client's URL sits under, set by
|
|
482
|
+
``initialize_metering()`` or ``REVENIUM_METERING_BASE_URL``) plus
|
|
483
|
+
``/profitstream``, where the hosted API serves the enforcement rules.
|
|
472
484
|
"""
|
|
473
485
|
explicit = os.environ.get(Config.ENV_REVENIUM_ENFORCEMENT_BASE_URL, "")
|
|
474
486
|
if explicit:
|
|
475
487
|
return explicit.rstrip("/")
|
|
476
|
-
|
|
477
|
-
|
|
478
|
-
|
|
479
|
-
|
|
488
|
+
revenium_address = get_profitstream_base_url(metering_base_url=_process_metering_url())
|
|
489
|
+
return revenium_address.rstrip("/") + _PROFITSTREAM_PATH
|
|
490
|
+
|
|
491
|
+
|
|
492
|
+
def _sibling_enforcement_address(address: str) -> str:
|
|
493
|
+
"""The same address with ``/profitstream`` removed if present, added if not."""
|
|
494
|
+
parts = urlsplit(address)
|
|
495
|
+
path = parts.path.rstrip("/")
|
|
496
|
+
if path.lower().endswith(_PROFITSTREAM_PATH):
|
|
497
|
+
path = path[: -len(_PROFITSTREAM_PATH)]
|
|
498
|
+
else:
|
|
499
|
+
path = path + _PROFITSTREAM_PATH
|
|
500
|
+
return urlunsplit(parts._replace(path=path)).rstrip("/")
|
|
501
|
+
|
|
502
|
+
|
|
503
|
+
def _effective_enforcement_address(configured: str) -> str:
|
|
504
|
+
return _recovered_enforcement_addresses.get(configured, configured)
|
|
505
|
+
|
|
506
|
+
|
|
507
|
+
def _loggable_address(address: str) -> str:
|
|
508
|
+
"""An address as the logs may show it: no credentials, query or fragment."""
|
|
509
|
+
try:
|
|
510
|
+
return address_without_credentials(httpx.URL(address)).rstrip("/")
|
|
511
|
+
except Exception: # noqa: BLE001 - an address httpx cannot parse is not worth a traceback
|
|
512
|
+
return "the configured enforcement address"
|
|
513
|
+
|
|
514
|
+
|
|
515
|
+
def _adopt_enforcement_address(configured: str, tried: str, served: str) -> None:
|
|
516
|
+
"""Send every later read for ``configured`` to ``served``, which just proved it serves the rules."""
|
|
517
|
+
if served == configured:
|
|
518
|
+
_recovered_enforcement_addresses.pop(configured, None)
|
|
519
|
+
else:
|
|
520
|
+
_recovered_enforcement_addresses[configured] = served
|
|
521
|
+
logger.warning(
|
|
522
|
+
"Enforcement rules were not found at %s (HTTP 404) but are served at %s; using that "
|
|
523
|
+
"address from now on. Set %s to it to skip this lookup.",
|
|
524
|
+
_loggable_address(tried), _loggable_address(served), Config.ENV_REVENIUM_ENFORCEMENT_BASE_URL,
|
|
525
|
+
)
|
|
526
|
+
|
|
527
|
+
|
|
528
|
+
def _team_rules_url(address: str, team_id: str) -> str:
|
|
529
|
+
# Percent-encode the team_id path segment so a misconfigured value
|
|
530
|
+
# containing '/', '..', or '?' cannot retarget the request to a
|
|
531
|
+
# different endpoint on the same origin.
|
|
532
|
+
return f"{address}/v2/api/ai/enforcement-rules/{quote(team_id, safe='')}"
|
|
480
533
|
|
|
481
534
|
|
|
482
535
|
class _DepartmentBudgets(NamedTuple):
|
|
@@ -969,6 +1022,9 @@ class _FetchTarget(NamedTuple):
|
|
|
969
1022
|
api_key: str
|
|
970
1023
|
team_url: str
|
|
971
1024
|
owner: Tuple[str, str]
|
|
1025
|
+
# The address ``team_url`` was built on: the configured one, or the
|
|
1026
|
+
# sibling an earlier 404 recovered to.
|
|
1027
|
+
address: str
|
|
972
1028
|
|
|
973
1029
|
|
|
974
1030
|
def _fetch_target() -> Optional[_FetchTarget]:
|
|
@@ -1007,19 +1063,26 @@ def _fetch_target() -> Optional[_FetchTarget]:
|
|
|
1007
1063
|
return None
|
|
1008
1064
|
|
|
1009
1065
|
base_url = _get_enforcement_base_url()
|
|
1010
|
-
|
|
1011
|
-
|
|
1012
|
-
# different endpoint on the same origin.
|
|
1013
|
-
safe_team_id = quote(team_id, safe="")
|
|
1014
|
-
return _FetchTarget(api_key, f"{base_url}/v2/api/ai/enforcement-rules/{safe_team_id}",
|
|
1015
|
-
(base_url, team_id))
|
|
1066
|
+
address = _effective_enforcement_address(base_url)
|
|
1067
|
+
return _FetchTarget(api_key, _team_rules_url(address, team_id), (base_url, team_id), address)
|
|
1016
1068
|
|
|
1017
1069
|
|
|
1018
1070
|
def _get_enforcement(url: str, api_key: str, resource: str,
|
|
1019
1071
|
params: Optional[dict], retry_after_budget: float,
|
|
1020
1072
|
owner: Tuple[str, str]) -> Optional["httpx.Response"]:
|
|
1073
|
+
"""``_get_enforcement_spending`` for a read that has no second address to spend the rest on."""
|
|
1074
|
+
return _get_enforcement_spending(url, api_key, resource, params, retry_after_budget, owner)[0]
|
|
1075
|
+
|
|
1076
|
+
|
|
1077
|
+
def _get_enforcement_spending(url: str, api_key: str, resource: str,
|
|
1078
|
+
params: Optional[dict], retry_after_budget: float,
|
|
1079
|
+
owner: Tuple[str, str]) -> Tuple[Optional["httpx.Response"], float]:
|
|
1021
1080
|
"""GET one enforcement resource under the shared retry and Retry-After budget.
|
|
1022
1081
|
|
|
1082
|
+
Returns the response and the Retry-After budget this read left unspent, so
|
|
1083
|
+
a read that goes on to a second address (``_rules_at_sibling_address``)
|
|
1084
|
+
spends one budget across both and never doubles the documented wait.
|
|
1085
|
+
|
|
1023
1086
|
Returns the first response the retry policy does not retry -- the caller
|
|
1024
1087
|
reads its status and its body -- or ``None`` once the fetch has given up,
|
|
1025
1088
|
so every caller fails open on what it already holds. ``resource`` names
|
|
@@ -1051,10 +1114,10 @@ def _get_enforcement(url: str, api_key: str, resource: str,
|
|
|
1051
1114
|
continue
|
|
1052
1115
|
except Exception:
|
|
1053
1116
|
logger.debug("Failed to fetch enforcement %s, falling open", resource, exc_info=True)
|
|
1054
|
-
return None
|
|
1117
|
+
return None, retry_after_budget
|
|
1055
1118
|
|
|
1056
1119
|
if not _is_retryable_status(response.status_code):
|
|
1057
|
-
return response
|
|
1120
|
+
return response, retry_after_budget
|
|
1058
1121
|
|
|
1059
1122
|
plan = _plan_retry(response, backoff, retry_after_budget, budget_total, owner)
|
|
1060
1123
|
retry_after_budget = plan.budget_remaining
|
|
@@ -1062,7 +1125,7 @@ def _get_enforcement(url: str, api_key: str, resource: str,
|
|
|
1062
1125
|
# Give up without waiting and without retrying: the cooldown
|
|
1063
1126
|
# _plan_retry recorded keeps the server's interval, and the caller
|
|
1064
1127
|
# falls open on the reading it already has.
|
|
1065
|
-
return None
|
|
1128
|
+
return None, retry_after_budget
|
|
1066
1129
|
if attempt == _FETCH_MAX_ATTEMPTS - 1:
|
|
1067
1130
|
break
|
|
1068
1131
|
if _sleep(plan.wait_seconds):
|
|
@@ -1075,7 +1138,7 @@ def _get_enforcement(url: str, api_key: str, resource: str,
|
|
|
1075
1138
|
"Enforcement %s fetch exhausted %d attempts; failing open",
|
|
1076
1139
|
resource, _FETCH_MAX_ATTEMPTS,
|
|
1077
1140
|
)
|
|
1078
|
-
return None
|
|
1141
|
+
return None, retry_after_budget
|
|
1079
1142
|
|
|
1080
1143
|
|
|
1081
1144
|
def _rules_query(rule_id: Optional[str], include_advisory: bool) -> Optional[dict]:
|
|
@@ -1121,15 +1184,25 @@ def _fetch_rules(rule_id: Optional[str] = None,
|
|
|
1121
1184
|
if target is None:
|
|
1122
1185
|
return None
|
|
1123
1186
|
|
|
1124
|
-
response =
|
|
1125
|
-
|
|
1187
|
+
response, budget_left = _get_enforcement_spending(target.team_url, target.api_key, "rule",
|
|
1188
|
+
params, retry_after_budget, target.owner)
|
|
1126
1189
|
if response is None:
|
|
1127
1190
|
return None
|
|
1191
|
+
if response.status_code == 404:
|
|
1192
|
+
recovered = _rules_at_sibling_address(target, params, budget_left)
|
|
1193
|
+
if recovered is not None:
|
|
1194
|
+
return recovered.fetched
|
|
1195
|
+
return _rules_from_response(response)
|
|
1196
|
+
|
|
1128
1197
|
|
|
1198
|
+
def _rules_from_response(response: "httpx.Response") -> Optional[_FetchedRules]:
|
|
1199
|
+
"""The rules a rules read answered with, or ``None`` when it answered with anything else."""
|
|
1129
1200
|
try:
|
|
1130
1201
|
# 204 No Content == no rules configured for this team; cache empty
|
|
1131
1202
|
if response.status_code == 204:
|
|
1132
1203
|
return _FetchedRules([], {}, {}, {}, {})
|
|
1204
|
+
if 400 <= response.status_code < 500:
|
|
1205
|
+
_warn_refused_rules_read(response)
|
|
1133
1206
|
response.raise_for_status()
|
|
1134
1207
|
# Server currently returns ``{"rules": [...], "compiledAt": ...,
|
|
1135
1208
|
# "departmentBudgetBlocks": {...}, "departmentBudgetBlockBalances":
|
|
@@ -1141,6 +1214,60 @@ def _fetch_rules(rule_id: Optional[str] = None,
|
|
|
1141
1214
|
return None
|
|
1142
1215
|
|
|
1143
1216
|
|
|
1217
|
+
class _RecoveredRules(NamedTuple):
|
|
1218
|
+
address: str
|
|
1219
|
+
fetched: _FetchedRules
|
|
1220
|
+
budget_left: float
|
|
1221
|
+
|
|
1222
|
+
|
|
1223
|
+
def _rules_at_sibling_address(target: _FetchTarget, params: Optional[dict],
|
|
1224
|
+
retry_after_budget: float) -> Optional[_RecoveredRules]:
|
|
1225
|
+
"""Read the rules once at the sibling of the address that answered 404, and adopt it if it serves them.
|
|
1226
|
+
|
|
1227
|
+
The hosted API serves enforcement under ``/profitstream``; a self-hosted
|
|
1228
|
+
server may serve it at the root, and an operator may have configured
|
|
1229
|
+
either address without that segment. A 404 says the rules are not at this
|
|
1230
|
+
address at all, not that the team has none (that is a 204), so the one
|
|
1231
|
+
other address is tried. It is adopted for every later read only once it
|
|
1232
|
+
has answered with rules (a 204, or a body ``_fetched_from_payload``
|
|
1233
|
+
accepts): a refusal or an unreadable body there leaves the configured
|
|
1234
|
+
address in use, so the next poll reads it again. A refusal at the sibling
|
|
1235
|
+
(any 4xx but 404) is logged by ``_warn_refused_rules_read`` as well, so a
|
|
1236
|
+
wrong key or team id at the working address is not hidden behind the
|
|
1237
|
+
first address's 404.
|
|
1238
|
+
"""
|
|
1239
|
+
configured, team_id = target.owner
|
|
1240
|
+
sibling = _sibling_enforcement_address(target.address)
|
|
1241
|
+
response, budget_left = _get_enforcement_spending(
|
|
1242
|
+
_team_rules_url(sibling, team_id), target.api_key, "rule", params,
|
|
1243
|
+
retry_after_budget, target.owner)
|
|
1244
|
+
if response is None:
|
|
1245
|
+
return None
|
|
1246
|
+
if not (response.is_success or response.status_code == 204):
|
|
1247
|
+
if 400 <= response.status_code < 500 and response.status_code != 404:
|
|
1248
|
+
_warn_refused_rules_read(response)
|
|
1249
|
+
return None
|
|
1250
|
+
fetched = _rules_from_response(response)
|
|
1251
|
+
if fetched is None:
|
|
1252
|
+
return None
|
|
1253
|
+
_adopt_enforcement_address(configured, target.address, sibling)
|
|
1254
|
+
return _RecoveredRules(sibling, fetched, budget_left)
|
|
1255
|
+
|
|
1256
|
+
|
|
1257
|
+
def _warn_refused_rules_read(response: "httpx.Response") -> None:
|
|
1258
|
+
"""Say which URL refused the rules read and why, so a wrong address or key is visible."""
|
|
1259
|
+
status = response.status_code
|
|
1260
|
+
if status == 404:
|
|
1261
|
+
hint = (" The enforcement API is not at this address: set %s to the Revenium address "
|
|
1262
|
+
"that serves /v2/api/ai/enforcement-rules." % Config.ENV_REVENIUM_ENFORCEMENT_BASE_URL)
|
|
1263
|
+
elif status in (401, 403):
|
|
1264
|
+
hint = " Check %s and %s." % (Config.ENV_REVENIUM_API_KEY, Config.ENV_REVENIUM_TEAM_ID)
|
|
1265
|
+
else:
|
|
1266
|
+
hint = ""
|
|
1267
|
+
logger.warning("Enforcement rules could not be read: GET %s answered HTTP %d.%s",
|
|
1268
|
+
address_without_credentials(response.request.url), status, hint)
|
|
1269
|
+
|
|
1270
|
+
|
|
1144
1271
|
def _fetch_roster(rule_id: str, page: int = 0, size: int = 25,
|
|
1145
1272
|
search: Optional[str] = None,
|
|
1146
1273
|
band: Optional[str] = None) -> Optional[dict]:
|
|
@@ -1172,11 +1299,23 @@ def _fetch_roster(rule_id: str, page: int = 0, size: int = 25,
|
|
|
1172
1299
|
if band:
|
|
1173
1300
|
params["band"] = band
|
|
1174
1301
|
|
|
1175
|
-
response =
|
|
1176
|
-
|
|
1177
|
-
|
|
1302
|
+
response, budget_left = _get_enforcement_spending(
|
|
1303
|
+
f"{target.team_url}/roster", target.api_key, "roster", params,
|
|
1304
|
+
_INSPECTION_RETRY_AFTER_GIVE_UP_SECONDS, target.owner)
|
|
1178
1305
|
if response is None:
|
|
1179
1306
|
return None
|
|
1307
|
+
if response.status_code == 404:
|
|
1308
|
+
# Either the rule has no roster or the rules are not at this address.
|
|
1309
|
+
# Only a rules read can tell: if it recovers the address, the roster
|
|
1310
|
+
# is read once more there.
|
|
1311
|
+
recovered = _rules_at_sibling_address(target, None, budget_left)
|
|
1312
|
+
if recovered is None:
|
|
1313
|
+
return None
|
|
1314
|
+
response = _get_enforcement(f"{_team_rules_url(recovered.address, target.owner[1])}/roster",
|
|
1315
|
+
target.api_key, "roster", params, recovered.budget_left,
|
|
1316
|
+
target.owner)
|
|
1317
|
+
if response is None:
|
|
1318
|
+
return None
|
|
1180
1319
|
|
|
1181
1320
|
try:
|
|
1182
1321
|
# 204 No Content == the rule has no compiled reading yet, which is a
|
{revenium_python_sdk-0.12.0 → revenium_python_sdk-0.13.0}/revenium_middleware/_core/log_sanitize.py
RENAMED
|
@@ -264,3 +264,9 @@ def redact_error_text(text: str, max_len: int) -> str:
|
|
|
264
264
|
redacted = _EMAIL_WORD_PATTERN.sub(_REDACTED, redacted)
|
|
265
265
|
redacted = _SECRET_SHAPED_WORD_PATTERN.sub(_REDACTED, redacted)
|
|
266
266
|
return redacted[:max_len]
|
|
267
|
+
|
|
268
|
+
|
|
269
|
+
def address_without_credentials(url) -> str:
|
|
270
|
+
"""Scheme, host, port and path of an ``httpx.URL``: never its userinfo, query or fragment."""
|
|
271
|
+
port = f":{url.port}" if url.port is not None else ""
|
|
272
|
+
return f"{url.scheme}://{url.host}{port}{url.path}"
|