metrics

type ChatCompletionMetrics interface {
	// StartRequest initializes timing for a new request.
	StartRequest(headers map[string]string)
	// SetOriginalModel sets the original model from the incoming request body before any virtualization applies.
	// This is usually called after parsing the request body. Example: gpt-5
	SetOriginalModel(originalModel internalapi.OriginalModel)
	// SetRequestModel sets the model from the request. This is usually called after parsing the request body.
	// Example: gpt-5-nano
	SetRequestModel(requestModel internalapi.RequestModel)
	// SetResponseModel sets the model that ultimately generated the response.
	// Example: gpt-5-nano-2025-08-07
	SetResponseModel(responseModel internalapi.ResponseModel)
	// SetBackend sets the selected backend when the routing decision has been made. This is usually called
	// after parsing the request body to determine the model and invoke the routing logic.
	SetBackend(backend *filterapi.Backend)

	// RecordTokenUsage records token usage metrics.
	RecordTokenUsage(ctx context.Context, inputTokens, cachedInputTokens, outputTokens uint32, requestHeaderLabelMapping map[string]string)
	// RecordRequestCompletion records latency metrics for the entire request.
	RecordRequestCompletion(ctx context.Context, success bool, requestHeaderLabelMapping map[string]string)
	// RecordTokenLatency records latency metrics for token generation.
	RecordTokenLatency(ctx context.Context, tokens uint32, endOfStream bool, requestHeaderLabelMapping map[string]string)
	// GetTimeToFirstTokenMs returns the time to first token in stream mode in milliseconds.
	GetTimeToFirstTokenMs() float64
	// GetInterTokenLatencyMs returns the inter token latency in stream mode in milliseconds.
	GetInterTokenLatencyMs() float64
}

type ChatCompletionMetricsFactory func() ChatCompletionMetrics

type CompletionMetrics interface {
	// StartRequest initializes timing for a new request.
	StartRequest(headers map[string]string)
	// SetOriginalModel sets the original model from the incoming request body before any virtualization applies.
	// This is usually called after parsing the request body. Example: gpt-3.5-turbo-instruct
	SetOriginalModel(originalModel internalapi.OriginalModel)
	// SetRequestModel sets the model from the request. This is usually called after parsing the request body.
	// Example: gpt-3.5-turbo-instruct-override
	SetRequestModel(requestModel internalapi.RequestModel)
	// SetResponseModel sets the model that ultimately generated the response.
	// Example: gpt-3.5-turbo-instruct-0914
	SetResponseModel(responseModel internalapi.ResponseModel)
	// SetBackend sets the selected backend when the routing decision has been made. This is usually called
	// after parsing the request body to determine the model and invoke the routing logic.
	SetBackend(backend *filterapi.Backend)

	// RecordTokenUsage records token usage metrics.
	RecordTokenUsage(ctx context.Context, inputTokens, outputTokens uint32, requestHeaderLabelMapping map[string]string)
	// RecordRequestCompletion records latency metrics for the entire request.
	RecordRequestCompletion(ctx context.Context, success bool, requestHeaderLabelMapping map[string]string)
	// RecordTokenLatency records latency metrics for token generation.
	RecordTokenLatency(ctx context.Context, tokens uint32, endOfStream bool, requestHeaderLabelMapping map[string]string)
	// GetTimeToFirstTokenMs returns the time to first token in stream mode in milliseconds.
	GetTimeToFirstTokenMs() float64
	// GetInterTokenLatencyMs returns the inter token latency in stream mode in milliseconds.
	GetInterTokenLatencyMs() float64
}

type CompletionMetricsFactory func() CompletionMetrics

type EmbeddingsMetrics interface {
	// StartRequest initializes timing for a new request.
	StartRequest(headers map[string]string)
	// SetOriginalModel sets the original model from the incoming request body before any virtualization applies.
	// This is usually called after parsing the request body. Example: text-embedding-3-small
	SetOriginalModel(originalModel internalapi.OriginalModel)
	// SetRequestModel sets the model from the request. This is usually called after parsing the request body.
	// Example: text-embedding-3-small
	SetRequestModel(requestModel internalapi.RequestModel)
	// SetResponseModel sets the model that ultimately generated the response.
	// Example: text-embedding-3-small-2025-02-18
	SetResponseModel(responseModel internalapi.ResponseModel)
	// SetBackend sets the selected backend when the routing decision has been made. This is usually called
	// after parsing the request body to determine the model and invoke the routing logic.
	SetBackend(backend *filterapi.Backend)

	// RecordTokenUsage records token usage metrics for embeddings (only input tokens are relevant).
	RecordTokenUsage(ctx context.Context, inputTokens uint32, requestHeaderLabelMapping map[string]string)
	// RecordRequestCompletion records latency metrics for the entire request.
	RecordRequestCompletion(ctx context.Context, success bool, requestHeaderLabelMapping map[string]string)
}

type EmbeddingsMetricsFactory func() EmbeddingsMetrics

type ImageGenerationMetrics interface {
	// StartRequest initializes timing for a new request.
	StartRequest(headers map[string]string)
	// SetOriginalModel sets the original model from the incoming request body before any virtualization applies.
	// This is usually called after parsing the request body. Example: dall-e-3
	SetOriginalModel(originalModel internalapi.OriginalModel)
	// SetRequestModel sets the request model name.
	SetRequestModel(requestModel internalapi.RequestModel)
	// SetResponseModel sets the response model name.
	SetResponseModel(responseModel internalapi.ResponseModel)
	// SetBackend sets the selected backend when the routing decision has been made. This is usually called
	// after parsing the request body to determine the model and invoke the routing logic.
	SetBackend(backend *filterapi.Backend)

	// RecordTokenUsage records token usage metrics (image gen typically 0, but supported).
	RecordTokenUsage(ctx context.Context, inputTokens, outputTokens uint32, requestHeaderLabelMapping map[string]string)
	// RecordRequestCompletion records latency metrics for the entire request.
	RecordRequestCompletion(ctx context.Context, success bool, requestHeaderLabelMapping map[string]string)
	// RecordImageGeneration records metrics specific to image generation (request duration only).
	RecordImageGeneration(ctx context.Context, requestHeaderLabelMapping map[string]string)
}

type ImageGenerationMetricsFactory func() ImageGenerationMetrics

type MCPErrorType string

type MCPMetrics interface {
	// WithRequestAttributes returns a new MCPMetrics instance with default attributes extracted from the HTTP request.
	WithRequestAttributes(req *http.Request) MCPMetrics
	// RecordRequestDuration records the duration of a success MCP request.
	RecordRequestDuration(ctx context.Context, startAt *time.Time, meta mcpsdk.Params)
	// RecordRequestErrorDuration records the duration of an MCP request that resulted in an error.
	RecordRequestErrorDuration(ctx context.Context, startAt *time.Time, errType MCPErrorType, meta mcpsdk.Params)
	// RecordMethodCount records the count of method invocations.
	RecordMethodCount(ctx context.Context, methodName string, meta mcpsdk.Params)
	// RecordMethodErrorCount records the count of method invocations with error status.
	RecordMethodErrorCount(ctx context.Context, meta mcpsdk.Params)
	// RecordInitializationDuration records the duration of MCP initialization.
	RecordInitializationDuration(ctx context.Context, startAt *time.Time, meta mcpsdk.Params)
	// RecordClientCapabilities records the negotiated client capabilities.
	RecordClientCapabilities(ctx context.Context, capabilities *mcpsdk.ClientCapabilities, meta mcpsdk.Params)
	// RecordServerCapabilities records the negotiated server capabilities.
	RecordServerCapabilities(ctx context.Context, capabilities *mcpsdk.ServerCapabilities, meta mcpsdk.Params)
	// RecordProgress records a progress notification sent/received.
	RecordProgress(ctx context.Context, meta mcpsdk.Params)
}

type MessagesMetrics interface {
	ChatCompletionMetrics
}

type MessagesMetricsFactory func() MessagesMetrics

type RerankMetrics interface {
	// StartRequest initializes timing for a new request.
	StartRequest(headers map[string]string)
	// SetOriginalModel sets the original model from the incoming request body before any virtualization applies.
	// This is usually called after parsing the request body. Example: rerank-english-v3
	SetOriginalModel(originalModel internalapi.OriginalModel)
	// SetRequestModel sets the model from the request. This is usually called after parsing the request body.
	// Example: rerank-english-v3
	SetRequestModel(requestModel internalapi.RequestModel)
	// SetResponseModel sets the model that ultimately generated the response.
	// Example: rerank-english-v3-2025-02-18
	SetResponseModel(responseModel internalapi.ResponseModel)
	// SetBackend sets the selected backend when the routing decision has been made. This is usually called
	// after parsing the request body to determine the model and invoke the routing logic.
	SetBackend(backend *filterapi.Backend)

	// RecordTokenUsage records token usage metrics for rerank (only input tokens are relevant).
	RecordTokenUsage(ctx context.Context, inputTokens uint32, requestHeaderLabelMapping map[string]string)
	// RecordRequestCompletion records latency metrics for the entire request.
	RecordRequestCompletion(ctx context.Context, success bool, requestHeaderLabelMapping map[string]string)
}

type RerankMetricsFactory func() RerankMetrics