syntax = "proto3";

package com.mlp.gate;

import "google/protobuf/duration.proto";

option java_multiple_files = true;

service Gate {
  rpc healthCheck (HeartBeatProto) returns (HeartBeatProto);

  rpc processAsync(stream ServiceToGateProto) returns (stream GateToServiceProto) {}

  rpc process(ClientRequestProto) returns (ClientResponseProto) {}

  rpc processResponseStream(ClientRequestProto) returns (stream ClientResponseProto) {}

  rpc processStream(stream ClientRequestProto) returns (stream ClientResponseProto) {}

  rpc processSynthesis(ClientTtsRequestProto) returns (stream ClientTtsResponseProto) {}

  rpc processRecognition(stream ClientAsrRequestProto) returns (stream ClientAsrResponseProto) {}
}

message ServiceDescriptorProto {
  string name = 1;

  bool fittable = 10;
  map<string, MethodDescriptorProto> methods = 20;

  // Контент протобаф-спеки для специфических типов данных, используемых данных экшеном.
  // пока что это поле предназначено для экспериментов
  map<string, string> schemaFiles = 30;
}

message MethodDescriptorProto {
  map<string, ParamDescriptorProto> input = 1;
  ParamDescriptorProto output = 2;
  optional bool fitted = 3;
}

message ParamDescriptorProto {
  string type = 1;
}

message ServiceInfoProto {
  int64 accountId = 1;
  int64 modelId = 2;
  string modelName = 3;
  string authToken = 4;
  optional string bucketName = 5;
}

message ServiceToGateProto {
  int64 requestId = 1;

  oneof body {
    HeartBeatProto heartBeat = 2;
    StartServingProto startServing = 3;
    StopServingProto stopServing = 4;

    PredictResponseProto predict = 101;
    FitResponseProto fit = 102;
    ExtendedResponseProto ext = 103;

    StatusResponseProto status = 105;

    BatchResponseProto batch = 106;

    PartialPredictResponseProto partialPredict = 107;

    FitStatusProto fitStatus = 108;

    DeferredBillingChargeRequestProto deferredBillingCharge = 109;

    DeferredBillingChargesProto deferredBillingCharges = 110;

    ApiErrorProto error = 201;
  }

  map<string, string> headers = 1000;
}

message GateToServiceProto {
  int64 requestId = 1;

  oneof body {
    HeartBeatProto heartBeat = 2;
    ClusterUpdateProto cluster = 3;
    ServiceInfoProto serviceInfo = 4;
    StopServingProto stopServing = 5;

    PredictRequestProto predict = 101;
    FitRequestProto fit = 102;
    ExtendedRequestProto ext = 103;

    StatusRequestProto status = 105;

    BatchRequestProto batch = 106;

    PartialPredictRequestProto partialPredict = 107;
    CancelRequestProto cancel = 108;

    ApiErrorProto error = 201;
  }

  map<string, string> headers = 1000;
}

message HeartBeatProto {
  string status = 1;
  int32  interval = 2;
}

message StartServingProto {
  // Чтобы зарегистрироваться, сервис должен указать секретный ключ. Этот ключ генерирует Кайла
  // и передаёт в Cloud через env-переменные
  string connectionToken = 1;

  ServiceDescriptorProto serviceDescriptor = 3;

  optional string hostname = 4;
  optional int64 version = 5;
  optional string image = 6;

  // Unique identifier of the current SDK process lifetime.
  // Stable within a single process run; regenerated on instance restart.
  // Used by gateway to distinguish transient network disconnect from instance restart.
  // Field 7 is reserved (used as `imageVersionJson` in legacy mlp-python-sdk proto copy).
  optional string instanceBootUuid = 8;
}

message StopServingProto {
  optional string reason = 1;
}

message PredictRequestProto {
  PayloadProto data = 101;
  optional PayloadProto config = 102;
}

message PartialPredictRequestProto {
  PayloadProto data = 101;
  optional PayloadProto config = 102;
  optional bool start = 103;
  optional bool finish = 104;
}

message BatchRequestProto {
  repeated BatchPayloadProto data = 101;
  optional PayloadProto config = 102;
}

message PredictResponseProto {
  PayloadProto data = 101;

  map<string, string> headers = 102;
  optional int32 statusCode = 103;
}

message PartialPredictResponseProto {
  PayloadProto data = 101;
  optional bool start = 103;
  optional bool finish = 104;

  map<string, string> headers = 105;
  optional int32 statusCode = 106;
}

message FitStatusProto {
  int32 percentage = 101;
}

message DeferredBillingChargeRequestProto {
  string billingRequestId = 1;
  int64 amountInUnits = 2;
}

message DeferredBillingChargesProto {
  repeated DeferredBillingChargeProto charges = 1;
  DeferredBillingParamsProto params = 2;
}

message DeferredBillingParamsProto {
  optional int64 callerAccountId = 1;
  optional string apiKeyName = 2;
  optional string billingKeyName = 3;
  optional int64 billingAccountId = 4;
  optional string billingUserId = 5;
}

message DeferredBillingChargeProto {
  string idempotencyKey = 1;
  int64 amount = 2;
  optional string currency = 3;
  optional string llmModelName = 4;
  optional string billingDetails = 5;
}

message BatchResponseProto {
  repeated BatchPayloadResponseProto data = 101;
}

message DatasetInfoProto {
  int64 accountId = 1;
  int64 datasetId = 2;
  string name = 3;
  string type = 4;
}

message FitRequestProto {
  PayloadProto trainData = 101;
  optional PayloadProto targetsData = 102;
  optional PayloadProto config = 103;
  string modelDir = 104;
  optional string previousModelDir = 105;
  ServiceInfoProto targetServiceInfo = 106;
  DatasetInfoProto datasetInfo = 107;
}

message FitResponseProto {
}

message ExtendedRequestProto {
  string methodName = 1;
  map<string, PayloadProto> params = 2;
}

message ExtendedResponseProto {
  PayloadProto data = 1;

  map<string, string> headers = 102;
  optional int32 statusCode = 103;
}

message BatchPayloadProto {
  int64 requestId = 1;
  PayloadProto data = 101;

  map<string, string> headers = 102;
  optional int32 statusCode = 103;
}

message BatchPayloadResponseProto {
  int64 requestId = 1;

  oneof body {
    PredictResponseProto predict = 101;
    ApiErrorProto error = 201;
  }

  map<string, string> headers = 1000;
}

message StatusRequestProto {
}

message StatusResponseProto {
  repeated string connectedGates = 1;
  bool connectedToAllGates = 2;
}

message CancelRequestProto {
  int64 requestIdToCancel = 101;
}

message PayloadProto {
  // имя класса, закодированного в text или bytes
  optional string dataType = 1;

  oneof body {
    string json = 101;
    bytes protobuf = 102;
  }
}

enum SimpleStatusProto {
  // Success
  OK = 0;

  // Client errors (4xx)
  BAD_REQUEST = 400;
  UNAUTHORIZED = 401;
  PAYMENT_REQUIRED = 402;
  FORBIDDEN = 403;
  NOT_FOUND = 404;
  METHOD_NOT_ALLOWED = 405;
  NOT_ACCEPTABLE = 406;
  PROXY_AUTHENTICATION_REQUIRED = 407;
  REQUEST_TIMEOUT = 408;
  CONFLICT = 409;
  GONE = 410;
  LENGTH_REQUIRED = 411;
  PRECONDITION_FAILED = 412;
  PAYLOAD_TOO_LARGE = 413;
  URI_TOO_LONG = 414;
  UNSUPPORTED_MEDIA_TYPE = 415;
  RANGE_NOT_SATISFIABLE = 416;
  EXPECTATION_FAILED = 417;
  IM_A_TEAPOT = 418;
  MISDIRECTED_REQUEST = 421;
  UNPROCESSABLE_ENTITY = 422;
  LOCKED = 423;
  FAILED_DEPENDENCY = 424;
  TOO_EARLY = 425;
  UPGRADE_REQUIRED = 426;
  PRECONDITION_REQUIRED = 428;
  TOO_MANY_REQUESTS = 429;
  REQUEST_HEADER_FIELDS_TOO_LARGE = 431;
  UNAVAILABLE_FOR_LEGAL_REASONS = 451;

  // Nginx specific (4xx)
  NO_RESPONSE = 444;                        // Nginx: Connection closed without response
  REQUEST_HEADER_TOO_LARGE = 494;           // Nginx: Request header too large
  SSL_CERTIFICATE_ERROR = 495;              // Nginx: SSL certificate error
  SSL_CERTIFICATE_REQUIRED = 496;           // Nginx: SSL certificate required
  HTTP_REQUEST_SENT_TO_HTTPS_PORT = 497;    // Nginx: HTTP request sent to HTTPS port
  CLIENT_CLOSED_REQUEST = 499;              // Nginx: Client closed request

  // Server errors (5xx)
  INTERNAL_SERVER_ERROR = 500;
  NOT_IMPLEMENTED = 501;
  BAD_GATEWAY = 502;
  SERVICE_UNAVAILABLE = 503;
  GATEWAY_TIMEOUT = 504;
  HTTP_VERSION_NOT_SUPPORTED = 505;
  VARIANT_ALSO_NEGOTIATES = 506;
  INSUFFICIENT_STORAGE = 507;
  LOOP_DETECTED = 508;
  NOT_EXTENDED = 510;
  NETWORK_AUTHENTICATION_REQUIRED = 511;

  // Cloudflare specific (5xx)
  WEB_SERVER_RETURNED_UNKNOWN_ERROR = 520;  // Cloudflare: Web server returned unknown error
  WEB_SERVER_IS_DOWN = 521;                 // Cloudflare: Web server is down
  CONNECTION_TIMED_OUT = 522;               // Cloudflare: Connection timed out
  ORIGIN_IS_UNREACHABLE = 523;              // Cloudflare: Origin is unreachable
  A_TIMEOUT_OCCURRED = 524;                 // Cloudflare: A timeout occurred
  SSL_HANDSHAKE_FAILED = 525;               // Cloudflare: SSL handshake failed
  INVALID_SSL_CERTIFICATE = 526;            // Cloudflare: Invalid SSL certificate
  RAILGUN_ERROR = 527;                      // Cloudflare: Railgun error
  SITE_IS_OVERLOADED = 529;                 // Cloudflare: Site is overloaded
  SITE_IS_FROZEN = 530;                     // Cloudflare: Site is frozen

  // AWS specific
  ORIGIN_CONNECTION_TIME_OUT = 561;         // AWS ELB: Origin connection time-out
}

message ApiErrorProto {
  string code = 1;
  string message = 2;
  optional SimpleStatusProto status = 3;
  map<string, string> args = 4;
  optional int32 statusCode = 5;  // произвольный HTTP-статус код (приоритетнее status)
}

message ClientRequestProto {
  string account = 1; // тут можно передавать либо id либо shortName аккаунта и модели соответственно
  string model = 2;   // обработка будет происходить так: если только цифры - значит id, иначе name

  // Чтобы обратиться к экшену, клиент должен указать секретный токен.
  // Это может быть либо id-сессии в AA, либо отдельный токен доступа, который можно сгенерировать и получить
  // через MLP API. Внутри системы, токен будет ассоциирован с конкретным аккаунтом в АА.
  string authToken = 3;

  // Таймаут в секундах. Если не указан, то используется значение по умолчанию
  int32 timeoutSec = 4;

  oneof body {
    PredictRequestProto predict = 101;
    ExtendedRequestProto ext = 103;
    PartialPredictRequestProto partialPredict = 104;
  }

  map<string, string> headers = 1000;
}

message ClientResponseProto {
  oneof body {
    PredictResponseProto predict = 101;
    ExtendedResponseProto ext = 103;
    PartialPredictResponseProto partialPredict = 104;

    ApiErrorProto error = 201;
  }

  map<string, string> headers = 1000;
}

message ClientTokenRequestProto {
}

message ClientTokenResponseProto {
  string token = 1;
}

message ClusterUpdateProto {
  repeated string servers = 1;
  string currentServer = 2;
}

message SimpleTextProto {
  string text = 1;
}

// GRPC TTS API
message ClientTtsRequestProto {
  string account = 1;
  string model = 2;
  string authToken = 3;

  TtsRequestProto tts = 101;
}

message TtsRequestProto {
  string text = 1;
  optional string voice = 2;
  optional AudioFormatOptions output_audio_spec = 3;
}

message AudioFormatOptions {
  enum AudioEncoding {
    LINEAR16_PCM = 0;
  }
  optional AudioEncoding audio_encoding = 1;
  optional int64 sample_rate_hertz = 2;
  optional int64 chunk_size_kb = 3;
}

message ClientTtsResponseProto {
  oneof body {
    TtsResponseProto tts = 101;
    ApiErrorProto error = 201;
  }
}

message TtsResponseProto {
  message AudioChunk {
    bytes data = 1;
  }

  AudioChunk audio_chunk = 1;

  optional bool first = 10;
  optional bool last = 12;
}

message ClientAsrRequestProto {
  string account = 1;
  string model = 2;
  string authToken = 3;

  AsrRequestProto asr = 101;
}

message AsrRequestProto {
  oneof streaming_request {
    RecognitionConfig config = 1;
    bytes audio_content = 2;
  }
}

message RecognitionConfig {
  enum AudioEncoding {
    AUDIO_ENCODING_UNSPECIFIED = 0;
    LINEAR16_PCM = 1;
    OGG_OPUS = 2;
    MP3 = 3;
    MULAW = 4;
    ALAW = 5;
    FLAC = 6;
  }

  AudioEncoding audio_encoding = 1;
  int64 sample_rate_hertz = 2;
  // code in BCP-47
  string language_code = 3;
  bool enable_profanity_filter = 4;
  string model = 5;
  // If set true, tentative hypotheses may be returned as they become available (final=false flag)
  // If false or omitted, only final=true result(s) are returned.
  // Makes sense only for StreamingRecognize requests.
  bool enable_partial_results = 7;
  bool enable_single_utterance = 8;
  // Used only for long running recognize.
  int64 audio_channel_count = 9;
  // This mark allows disable normalization text
  bool enable_raw_results = 10;
  // Rewrite text in literature style (default: false)
  bool enable_literature_text = 11;
  bool enable_automatic_punctuation = 12;
  string provider_specific = 13;
}

message ClientAsrResponseProto {
  oneof body {
    AsrResponseProto asr = 101;
    ApiErrorProto error = 201;
  }
}

message AsrResponseProto {
  repeated SpeechRecognitionChunk chunks = 1;
  optional bool final = 2;
  string provider_specific = 3;
}

message SpeechRecognitionChunk {
  repeated SpeechRecognitionAlternative alternatives = 1;
  // This flag shows that the received chunk contains a part of the recognized text that won't be changed.
  bool final = 2;
  // This flag shows that the received chunk is the end of an utterance.
  bool end_of_utterance = 3;
}

message SpeechRecognitionAlternative {
  string text = 1;
  float confidence = 2;
  repeated WordInfo words = 3;
}

message WordInfo {
  google.protobuf.Duration start_time = 1;
  google.protobuf.Duration end_time = 2;
  string word = 3;
  float confidence = 4;
}