package config import ( "encoding/json" "errors" "log/slog" "net/url" "os" "strconv" "strings" ) const ( DefaultLocalGeneratedStorageDir = "data/static/generated" DefaultLocalUploadedStorageDir = "data/static/uploaded" ) type Config struct { AppEnv string HTTPAddr string DatabaseURL string IdentityMode string JWTSecret string ServerMainBaseURL string ServerMainInternalToken string ServerMainInternalKey string ServerMainInternalSecret string IdentitySecretStore string IdentitySecretDir string IdentityKubernetesNamespace string IdentityKubernetesSecretName string IdentityKubernetesAPIServer string IdentityKubernetesTokenFile string IdentityKubernetesCAFile string IdentitySecurityEventHeartbeatIntervalSeconds int IdentitySecurityEventStaleAfterSeconds int IdentitySecurityEventClockSkewSeconds int PublicBaseURL string WebBaseURL string LocalGeneratedStorageDir string LocalUploadedStorageDir string LocalTempAssetTTLHours int LocalResultTTLHours int LocalResultMinFreeBytes int64 LocalResultMaxBytes int64 LocalResultMaxTaskBytes int64 TaskProgressCallbackEnabled bool TaskProgressCallbackURL string TaskProgressCallbackTimeoutMS int TaskProgressCallbackMaxAttempts int TaskCleanupEnabled bool TaskRetentionDays int TaskAnalysisRetentionDays int TaskCleanupIntervalSeconds int TaskCleanupBatchSize int CORSAllowedOrigin string GlobalHTTPProxy string GlobalHTTPProxySource string PlatformProxyBypassIDs string LogLevel slog.Level BillingEngineMode string ProcessRole string DatabaseMaxConns int DatabaseCriticalMaxConns int DatabaseRiverMaxConns int DatabaseMinIdleConns int DatabaseMaxConnIdleSeconds int DatabaseIdleInTransactionTimeoutSeconds int DatabaseLockTimeoutSeconds int MediaRequestConcurrency int MediaMaterializationConcurrency int MediaImageNormalizationConcurrency int MediaOSSDirectEnabled bool MediaOSSEndpoint string MediaOSSBucket string MediaOSSAccessKeyID string MediaOSSAccessKeySecret string MediaOSSPublicBaseURL string MediaOSSObjectPrefix string AsyncQueueWorkerEnabled bool AsyncWorkerHardLimit int AsyncWorkerInstanceHardLimit int AsyncWorkerRefreshIntervalSeconds int AsyncWorkerLoadMode string AsyncAdmissionMicrobatchSize int AsyncAdmissionDispatcherEnabled bool AsyncAdmissionDispatcherConfigured bool RoutingMode string ExecutionPoolID string ExecutionPoolLabels string WorkerID string WorkerAdvertiseEndpoint string WorkerOrchestratorInstanceRef string WorkerEndpointAllowedSuffixes string WorkerEndpointAllowPrivate bool WorkerExecutionSecret string WorkerExecutionCAFile string WorkerExecutionCertFile string WorkerExecutionKeyFile string RouteProbeEnabled bool RouteProbeHotIntervalSeconds int RouteProbeColdIntervalSeconds int RouteProbeTimeoutMS int WorkerAutoscalingEnabled bool CapacityOrchestratorAdapter string CapacityPoolsJSON string WorkerTargetOutstandingPerReplica int WorkerScaleUpWindowSeconds int WorkerScaleDownStabilizationSeconds int WorkerDrainTimeoutSeconds int NodeMemoryTargetPercent int NodeMemoryHardPercent int NodeCPUTargetPercent int PostgresConnectionBudget int PostgresNonWorkerConnectionBudget int WorkerDatabaseMaxConns int } func Load() Config { globalProxy := LoadGlobalHTTPProxyStatus() appEnv := env("APP_ENV", "development") return Config{ AppEnv: appEnv, HTTPAddr: env("HTTP_ADDR", ":8088"), DatabaseURL: gatewayDatabaseURL(), IdentityMode: env("IDENTITY_MODE", "hybrid"), JWTSecret: env("CONFIG_JWT_SECRET", "this is a very secret secret"), ServerMainBaseURL: strings.TrimRight( env("SERVER_MAIN_BASE_URL", "http://localhost:3000"), "/", ), ServerMainInternalToken: env("SERVER_MAIN_INTERNAL_TOKEN", ""), ServerMainInternalKey: env("SERVER_MAIN_INTERNAL_KEY", "gateway"), ServerMainInternalSecret: env("SERVER_MAIN_INTERNAL_SECRET", env("SERVER_MAIN_INTERNAL_TOKEN", "")), IdentitySecretStore: env("IDENTITY_SECRET_STORE", "file"), IdentitySecretDir: env("IDENTITY_SECRET_DIR", ".local-secrets/identity"), IdentityKubernetesNamespace: env("IDENTITY_KUBERNETES_NAMESPACE", env("POD_NAMESPACE", "")), IdentityKubernetesSecretName: env("IDENTITY_KUBERNETES_SECRET_NAME", "easyai-gateway-identity"), IdentityKubernetesAPIServer: env("IDENTITY_KUBERNETES_API_SERVER", "https://kubernetes.default.svc"), IdentityKubernetesTokenFile: env("IDENTITY_KUBERNETES_TOKEN_FILE", "/var/run/secrets/kubernetes.io/serviceaccount/token"), IdentityKubernetesCAFile: env("IDENTITY_KUBERNETES_CA_FILE", "/var/run/secrets/kubernetes.io/serviceaccount/ca.crt"), IdentitySecurityEventHeartbeatIntervalSeconds: envInt("IDENTITY_SECURITY_EVENTS_HEARTBEAT_INTERVAL_SECONDS", 60), IdentitySecurityEventStaleAfterSeconds: envInt("IDENTITY_SECURITY_EVENTS_STALE_AFTER_SECONDS", 180), IdentitySecurityEventClockSkewSeconds: envInt("IDENTITY_SECURITY_EVENTS_CLOCK_SKEW_SECONDS", 60), PublicBaseURL: strings.TrimRight(env("AI_GATEWAY_PUBLIC_BASE_URL", env("PUBLIC_BASE_URL", "")), "/"), WebBaseURL: strings.TrimRight(env("AI_GATEWAY_WEB_BASE_URL", env("GATEWAY_WEB_BASE_URL", env("PUBLIC_WEB_BASE_URL", ""))), "/"), LocalGeneratedStorageDir: env("AI_GATEWAY_GENERATED_STORAGE_DIR", env("LOCAL_GENERATED_STORAGE_DIR", env("AI_GATEWAY_STATIC_STORAGE_DIR", DefaultLocalGeneratedStorageDir))), LocalUploadedStorageDir: env("AI_GATEWAY_UPLOADED_STORAGE_DIR", env("LOCAL_UPLOADED_STORAGE_DIR", DefaultLocalUploadedStorageDir)), LocalTempAssetTTLHours: envInt("AI_GATEWAY_LOCAL_TEMP_ASSET_TTL_HOURS", 24), LocalResultTTLHours: envIntValidated("AI_GATEWAY_LOCAL_RESULT_TTL_HOURS", 24), LocalResultMinFreeBytes: envInt64Validated("AI_GATEWAY_LOCAL_RESULT_MIN_FREE_BYTES", 10*1024*1024*1024), LocalResultMaxBytes: envInt64Validated("AI_GATEWAY_LOCAL_RESULT_MAX_BYTES", 256*1024*1024), LocalResultMaxTaskBytes: envInt64Validated("AI_GATEWAY_LOCAL_RESULT_MAX_TASK_BYTES", 512*1024*1024), TaskProgressCallbackEnabled: env("TASK_PROGRESS_CALLBACK_ENABLED", "true") == "true", TaskProgressCallbackURL: env("TASK_PROGRESS_CALLBACK_URL", strings.TrimRight(env("SERVER_MAIN_BASE_URL", "http://localhost:3000"), "/")+"/internal/platform/task-progress-callbacks", ), TaskProgressCallbackTimeoutMS: envIntValidated("TASK_PROGRESS_CALLBACK_TIMEOUT_MS", 5000), TaskProgressCallbackMaxAttempts: envIntValidated("TASK_PROGRESS_CALLBACK_MAX_ATTEMPTS", 10), TaskCleanupEnabled: env("AI_GATEWAY_TASK_CLEANUP_ENABLED", "false") == "true", TaskRetentionDays: envIntValidated("AI_GATEWAY_TASK_RETENTION_DAYS", 30), TaskAnalysisRetentionDays: envIntValidated("AI_GATEWAY_TASK_ANALYSIS_RETENTION_DAYS", 7), TaskCleanupIntervalSeconds: envIntValidated("AI_GATEWAY_TASK_CLEANUP_INTERVAL_SECONDS", 300), TaskCleanupBatchSize: envIntValidated("AI_GATEWAY_TASK_CLEANUP_BATCH_SIZE", 1000), CORSAllowedOrigin: env("CORS_ALLOWED_ORIGIN", "http://localhost:5178,http://127.0.0.1:5178"), GlobalHTTPProxy: globalProxy.HTTPProxy, GlobalHTTPProxySource: globalProxy.Source, PlatformProxyBypassIDs: env("AI_GATEWAY_PLATFORM_PROXY_BYPASS_IDS", ""), LogLevel: logLevel(env("LOG_LEVEL", "info")), BillingEngineMode: strings.ToLower(env("BILLING_ENGINE_MODE", "observe")), ProcessRole: strings.ToLower(strings.TrimSpace(env("AI_GATEWAY_PROCESS_ROLE", "all"))), DatabaseMaxConns: envInt("AI_GATEWAY_DATABASE_MAX_CONNS", 0), DatabaseCriticalMaxConns: envOptionalIntValidated("AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS", 0), DatabaseRiverMaxConns: envOptionalIntValidated("AI_GATEWAY_DATABASE_RIVER_MAX_CONNS", 0), DatabaseMinIdleConns: envOptionalIntValidated("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS", 0), DatabaseMaxConnIdleSeconds: envOptionalIntValidated("AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS", 0), DatabaseIdleInTransactionTimeoutSeconds: envOptionalIntValidated( "AI_GATEWAY_DATABASE_IDLE_IN_TRANSACTION_TIMEOUT_SECONDS", 60, ), DatabaseLockTimeoutSeconds: envOptionalIntValidated("AI_GATEWAY_DATABASE_LOCK_TIMEOUT_SECONDS", 30), MediaRequestConcurrency: envIntValidated("AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY", 16), MediaMaterializationConcurrency: envIntValidated("AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY", 8), MediaImageNormalizationConcurrency: envIntValidated( "AI_GATEWAY_MEDIA_IMAGE_NORMALIZATION_CONCURRENCY", 2, ), MediaOSSDirectEnabled: env("AI_GATEWAY_MEDIA_OSS_DIRECT_ENABLED", "false") == "true", MediaOSSEndpoint: strings.TrimRight(env("AI_GATEWAY_MEDIA_OSS_ENDPOINT", ""), "/"), MediaOSSBucket: strings.TrimSpace(env("AI_GATEWAY_MEDIA_OSS_BUCKET", "")), MediaOSSAccessKeyID: strings.TrimSpace(env("AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_ID", "")), MediaOSSAccessKeySecret: strings.TrimSpace(env("AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_SECRET", "")), MediaOSSPublicBaseURL: strings.TrimRight(env("AI_GATEWAY_MEDIA_OSS_PUBLIC_BASE_URL", ""), "/"), MediaOSSObjectPrefix: strings.Trim(env("AI_GATEWAY_MEDIA_OSS_OBJECT_PREFIX", "easyai-ai-gateway/media"), "/"), AsyncQueueWorkerEnabled: env("AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED", "true") == "true", AsyncWorkerHardLimit: envIntValidated( "AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT", envOptionalIntValidated("AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT", 2048), ), AsyncWorkerInstanceHardLimit: envIntValidated("AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT", 32), AsyncWorkerRefreshIntervalSeconds: envIntValidated("AI_GATEWAY_ASYNC_WORKER_REFRESH_INTERVAL_SECONDS", 5), AsyncWorkerLoadMode: strings.ToLower(strings.TrimSpace(env("AI_GATEWAY_WORKER_LOAD_MODE", "adaptive"))), AsyncAdmissionMicrobatchSize: envIntValidated("AI_GATEWAY_ASYNC_ADMISSION_MICROBATCH_SIZE", 8), AsyncAdmissionDispatcherEnabled: envValue("AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED") == "true", AsyncAdmissionDispatcherConfigured: envValue( "AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED", ) != "", RoutingMode: strings.ToLower(strings.TrimSpace(env("AI_GATEWAY_ROUTING_MODE", "legacy"))), ExecutionPoolID: strings.TrimSpace(env("AI_GATEWAY_EXECUTION_POOL_ID", "legacy-default")), ExecutionPoolLabels: strings.TrimSpace(env("AI_GATEWAY_EXECUTION_POOL_LABELS", "{}")), WorkerID: strings.TrimSpace(env("AI_GATEWAY_WORKER_ID", "")), WorkerAdvertiseEndpoint: workerAdvertiseEndpoint(), WorkerOrchestratorInstanceRef: strings.TrimSpace(env("AI_GATEWAY_ORCHESTRATOR_INSTANCE_REF", "")), WorkerEndpointAllowedSuffixes: strings.TrimSpace(env("AI_GATEWAY_WORKER_ENDPOINT_ALLOWED_SUFFIXES", "svc,cluster.local")), WorkerEndpointAllowPrivate: env("AI_GATEWAY_WORKER_ENDPOINT_ALLOW_PRIVATE", "true") == "true", WorkerExecutionSecret: env("AI_GATEWAY_WORKER_EXECUTION_SECRET", env("SERVER_MAIN_INTERNAL_SECRET", env("SERVER_MAIN_INTERNAL_TOKEN", ""))), WorkerExecutionCAFile: strings.TrimSpace(env("AI_GATEWAY_WORKER_EXECUTION_CA_FILE", "")), WorkerExecutionCertFile: strings.TrimSpace(env("AI_GATEWAY_WORKER_EXECUTION_CERT_FILE", "")), WorkerExecutionKeyFile: strings.TrimSpace(env("AI_GATEWAY_WORKER_EXECUTION_KEY_FILE", "")), RouteProbeEnabled: env("AI_GATEWAY_ROUTE_PROBE_ENABLED", "true") == "true", RouteProbeHotIntervalSeconds: envIntValidated("AI_GATEWAY_ROUTE_PROBE_HOT_INTERVAL_SECONDS", 15), RouteProbeColdIntervalSeconds: envIntValidated("AI_GATEWAY_ROUTE_PROBE_COLD_INTERVAL_SECONDS", 60), RouteProbeTimeoutMS: envIntValidated("AI_GATEWAY_ROUTE_PROBE_TIMEOUT_MS", 3000), WorkerAutoscalingEnabled: env("AI_GATEWAY_WORKER_AUTOSCALING_ENABLED", "false") == "true", CapacityOrchestratorAdapter: strings.ToLower(strings.TrimSpace(env("AI_GATEWAY_CAPACITY_ORCHESTRATOR_ADAPTER", "kubernetes"))), CapacityPoolsJSON: strings.TrimSpace(env("AI_GATEWAY_CAPACITY_POOLS", "")), WorkerTargetOutstandingPerReplica: envOptionalIntValidated("AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA", 0), WorkerScaleUpWindowSeconds: envIntValidated("AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS", 20), WorkerScaleDownStabilizationSeconds: envIntValidated( "AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS", 600, ), WorkerDrainTimeoutSeconds: envIntValidated("AI_GATEWAY_WORKER_DRAIN_TIMEOUT_SECONDS", 600), NodeMemoryTargetPercent: envIntValidated("AI_GATEWAY_NODE_MEMORY_TARGET_PERCENT", 75), NodeMemoryHardPercent: envIntValidated("AI_GATEWAY_NODE_MEMORY_HARD_PERCENT", 85), NodeCPUTargetPercent: envIntValidated("AI_GATEWAY_NODE_CPU_TARGET_PERCENT", 70), PostgresConnectionBudget: envIntValidated("AI_GATEWAY_POSTGRES_CONNECTION_BUDGET", 150), PostgresNonWorkerConnectionBudget: envIntValidated( "AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET", 72, ), WorkerDatabaseMaxConns: envIntValidated("AI_GATEWAY_WORKER_DATABASE_MAX_CONNS", 32), } } func (c Config) Validate() error { switch strings.ToLower(strings.TrimSpace(c.ProcessRole)) { case "", "all", "api", "worker", "capacity-controller": default: return errors.New("AI_GATEWAY_PROCESS_ROLE must be all, api, worker, or capacity-controller") } if c.DatabaseMaxConns < 0 || c.DatabaseMaxConns > 1000 { return errors.New("AI_GATEWAY_DATABASE_MAX_CONNS must be between 1 and 1000 when configured") } if c.DatabaseMinIdleConns < 0 || c.DatabaseMinIdleConns > 1000 || (c.DatabaseMaxConns > 0 && c.DatabaseMinIdleConns > c.DatabaseMaxConns) { return errors.New("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS must be between 0 and AI_GATEWAY_DATABASE_MAX_CONNS") } if c.DatabaseCriticalMaxConns < 0 || c.DatabaseCriticalMaxConns > 64 || (c.DatabaseMaxConns > 0 && c.DatabaseCriticalMaxConns >= c.DatabaseMaxConns) { return errors.New("AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS must be between 0 and less than AI_GATEWAY_DATABASE_MAX_CONNS") } if c.DatabaseRiverMaxConns < 0 || c.DatabaseRiverMaxConns > 256 || (c.DatabaseMaxConns > 0 && c.DatabaseCriticalMaxConns+c.DatabaseRiverMaxConns >= c.DatabaseMaxConns) { return errors.New("critical and River database pools must leave at least one execution connection") } if c.DatabaseMaxConns > 0 && c.DatabaseMinIdleConns > c.DatabaseMaxConns-c.DatabaseCriticalMaxConns-c.DatabaseRiverMaxConns { return errors.New("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS must not exceed the execution pool after reserving critical and River connections") } if c.DatabaseMaxConnIdleSeconds < 0 || c.DatabaseMaxConnIdleSeconds > 3600 { return errors.New("AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS must be between 0 and 3600") } if c.DatabaseIdleInTransactionTimeoutSeconds < 0 || c.DatabaseIdleInTransactionTimeoutSeconds > 3600 { return errors.New("AI_GATEWAY_DATABASE_IDLE_IN_TRANSACTION_TIMEOUT_SECONDS must be between 0 and 3600") } if c.DatabaseLockTimeoutSeconds < 0 || c.DatabaseLockTimeoutSeconds > 3600 { return errors.New("AI_GATEWAY_DATABASE_LOCK_TIMEOUT_SECONDS must be between 0 and 3600") } if c.MediaMaterializationConcurrency != 0 && (c.MediaMaterializationConcurrency < 1 || c.MediaMaterializationConcurrency > 256) { return errors.New("AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY must be between 1 and 256") } if c.MediaRequestConcurrency != 0 && (c.MediaRequestConcurrency < 1 || c.MediaRequestConcurrency > 1024) { return errors.New("AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY must be between 1 and 1024") } if c.MediaImageNormalizationConcurrency != 0 && (c.MediaImageNormalizationConcurrency < 1 || c.MediaImageNormalizationConcurrency > 64) { return errors.New("AI_GATEWAY_MEDIA_IMAGE_NORMALIZATION_CONCURRENCY must be between 1 and 64") } if c.MediaOSSDirectEnabled && !c.RunsCapacityController() { if strings.TrimSpace(c.MediaOSSAccessKeyID) == "" || strings.TrimSpace(c.MediaOSSAccessKeySecret) == "" { return errors.New("AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_ID and AI_GATEWAY_MEDIA_OSS_ACCESS_KEY_SECRET are required when direct media OSS is enabled") } if !validMediaOSSBucket(c.MediaOSSBucket) { return errors.New("AI_GATEWAY_MEDIA_OSS_BUCKET is invalid") } if !validMediaOSSURL(c.MediaOSSEndpoint, c.AppEnv) { return errors.New("AI_GATEWAY_MEDIA_OSS_ENDPOINT must be an HTTPS URL") } if !validMediaOSSURL(c.MediaOSSPublicBaseURL, c.AppEnv) { return errors.New("AI_GATEWAY_MEDIA_OSS_PUBLIC_BASE_URL must be an HTTPS URL") } if !validMediaOSSObjectPrefix(c.MediaOSSObjectPrefix) { return errors.New("AI_GATEWAY_MEDIA_OSS_OBJECT_PREFIX is invalid") } } switch strings.ToLower(strings.TrimSpace(c.BillingEngineMode)) { case "", "observe", "enforce", "hold": default: return errors.New("BILLING_ENGINE_MODE must be observe, enforce, or hold") } if c.AsyncWorkerHardLimit < 1 || c.AsyncWorkerHardLimit > 10000 { return errors.New("AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT must be between 1 and 10000") } if c.AsyncWorkerInstanceHardLimit < 1 || c.AsyncWorkerInstanceHardLimit > 10000 { return errors.New("AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT must be between 1 and 10000") } if c.AsyncWorkerRefreshIntervalSeconds < 1 { return errors.New("AI_GATEWAY_ASYNC_WORKER_REFRESH_INTERVAL_SECONDS must be positive") } switch strings.ToLower(strings.TrimSpace(c.AsyncWorkerLoadMode)) { case "", "adaptive", "legacy": default: return errors.New("AI_GATEWAY_WORKER_LOAD_MODE must be adaptive or legacy") } switch strings.ToLower(strings.TrimSpace(c.RoutingMode)) { case "", "legacy", "shadow", "enforced": default: return errors.New("AI_GATEWAY_ROUTING_MODE must be legacy, shadow, or enforced") } if strings.TrimSpace(c.ExecutionPoolID) != "" && !validExecutionPoolID(c.ExecutionPoolID) { return errors.New("AI_GATEWAY_EXECUTION_POOL_ID is invalid") } if strings.TrimSpace(c.ExecutionPoolLabels) != "" && !jsonObject(c.ExecutionPoolLabels) { return errors.New("AI_GATEWAY_EXECUTION_POOL_LABELS must be a JSON object") } if c.RouteProbeHotIntervalSeconds != 0 && (c.RouteProbeHotIntervalSeconds < 5 || c.RouteProbeHotIntervalSeconds > 300) { return errors.New("AI_GATEWAY_ROUTE_PROBE_HOT_INTERVAL_SECONDS must be between 5 and 300") } if c.RouteProbeColdIntervalSeconds != 0 && (c.RouteProbeColdIntervalSeconds < c.RouteProbeHotIntervalSeconds || c.RouteProbeColdIntervalSeconds > 3600) { return errors.New("AI_GATEWAY_ROUTE_PROBE_COLD_INTERVAL_SECONDS must be between the hot interval and 3600") } if c.RouteProbeTimeoutMS != 0 && (c.RouteProbeTimeoutMS < 250 || c.RouteProbeTimeoutMS > 10000) { return errors.New("AI_GATEWAY_ROUTE_PROBE_TIMEOUT_MS must be between 250 and 10000") } if strings.EqualFold(c.RoutingMode, "enforced") && len(c.WorkerExecutionSecret) < 32 { return errors.New("AI_GATEWAY_WORKER_EXECUTION_SECRET must be at least 32 bytes in enforced routing mode") } if (c.WorkerExecutionCertFile == "") != (c.WorkerExecutionKeyFile == "") { return errors.New("AI_GATEWAY_WORKER_EXECUTION_CERT_FILE and AI_GATEWAY_WORKER_EXECUTION_KEY_FILE must be configured together") } if c.AsyncAdmissionMicrobatchSize < 1 || c.AsyncAdmissionMicrobatchSize > 32 { return errors.New("AI_GATEWAY_ASYNC_ADMISSION_MICROBATCH_SIZE must be between 1 and 32") } pools, err := c.CapacityPools() if err != nil { return err } if c.RunsCapacityController() && len(pools) == 0 { return errors.New("AI_GATEWAY_CAPACITY_POOLS must contain at least one pool for the capacity-controller role") } switch strings.ToLower(strings.TrimSpace(c.CapacityOrchestratorAdapter)) { case "", "kubernetes", "static": default: return errors.New("AI_GATEWAY_CAPACITY_ORCHESTRATOR_ADAPTER must be kubernetes or static") } if c.WorkerTargetOutstandingPerReplica < 0 || c.WorkerTargetOutstandingPerReplica > 100000 { return errors.New("AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA must be between 0 and 100000") } if c.WorkerScaleUpWindowSeconds != 0 && (c.WorkerScaleUpWindowSeconds < 5 || c.WorkerScaleUpWindowSeconds > 600) { return errors.New("AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS must be between 5 and 600") } if c.WorkerScaleDownStabilizationSeconds != 0 && (c.WorkerScaleDownStabilizationSeconds < 60 || c.WorkerScaleDownStabilizationSeconds > 86400) { return errors.New("AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS must be between 60 and 86400") } if c.WorkerDrainTimeoutSeconds != 0 && (c.WorkerDrainTimeoutSeconds < 60 || c.WorkerDrainTimeoutSeconds > 86400) { return errors.New("AI_GATEWAY_WORKER_DRAIN_TIMEOUT_SECONDS must be between 60 and 86400") } if (c.NodeMemoryTargetPercent != 0 || c.NodeMemoryHardPercent != 0) && (c.NodeMemoryTargetPercent < 50 || c.NodeMemoryTargetPercent > 90 || c.NodeMemoryHardPercent <= c.NodeMemoryTargetPercent || c.NodeMemoryHardPercent > 95) { return errors.New("Worker node memory target/hard percentages must be ordered within 50..95") } if c.NodeCPUTargetPercent != 0 && (c.NodeCPUTargetPercent < 40 || c.NodeCPUTargetPercent > 90) { return errors.New("AI_GATEWAY_NODE_CPU_TARGET_PERCENT must be between 40 and 90") } if c.PostgresConnectionBudget < 0 || c.PostgresConnectionBudget > 1000 { return errors.New("AI_GATEWAY_POSTGRES_CONNECTION_BUDGET must be between 1 and 1000") } if c.PostgresNonWorkerConnectionBudget < 0 || (c.PostgresConnectionBudget > 0 && c.PostgresNonWorkerConnectionBudget >= c.PostgresConnectionBudget) { return errors.New("AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET must be below the PostgreSQL connection budget") } if c.WorkerDatabaseMaxConns < 0 || c.WorkerDatabaseMaxConns > 256 { return errors.New("AI_GATEWAY_WORKER_DATABASE_MAX_CONNS must be between 1 and 256") } if c.TaskProgressCallbackTimeoutMS != 0 && (c.TaskProgressCallbackTimeoutMS < 100 || c.TaskProgressCallbackTimeoutMS > 60000) { return errors.New("TASK_PROGRESS_CALLBACK_TIMEOUT_MS must be between 100 and 60000") } if c.TaskProgressCallbackMaxAttempts != 0 && (c.TaskProgressCallbackMaxAttempts < 1 || c.TaskProgressCallbackMaxAttempts > 100) { return errors.New("TASK_PROGRESS_CALLBACK_MAX_ATTEMPTS must be between 1 and 100") } if c.TaskRetentionDays != 0 && (c.TaskRetentionDays < 1 || c.TaskRetentionDays > 3650) { return errors.New("AI_GATEWAY_TASK_RETENTION_DAYS must be between 1 and 3650") } if c.TaskAnalysisRetentionDays != 0 && (c.TaskAnalysisRetentionDays < 1 || (c.TaskRetentionDays != 0 && c.TaskAnalysisRetentionDays > c.TaskRetentionDays)) { return errors.New("AI_GATEWAY_TASK_ANALYSIS_RETENTION_DAYS must be between 1 and AI_GATEWAY_TASK_RETENTION_DAYS") } if c.TaskCleanupIntervalSeconds != 0 && c.TaskCleanupIntervalSeconds < 60 { return errors.New("AI_GATEWAY_TASK_CLEANUP_INTERVAL_SECONDS must be at least 60") } if c.TaskCleanupBatchSize != 0 && (c.TaskCleanupBatchSize < 100 || c.TaskCleanupBatchSize > 5000) { return errors.New("AI_GATEWAY_TASK_CLEANUP_BATCH_SIZE must be between 100 and 5000") } if c.LocalResultTTLHours != 0 && (c.LocalResultTTLHours < 1 || c.LocalResultTTLHours > 24*30) { return errors.New("AI_GATEWAY_LOCAL_RESULT_TTL_HOURS must be between 1 and 720") } if c.LocalResultMinFreeBytes < 0 { return errors.New("AI_GATEWAY_LOCAL_RESULT_MIN_FREE_BYTES must not be negative") } if c.LocalResultMaxBytes != 0 && c.LocalResultMaxBytes < 1024 { return errors.New("AI_GATEWAY_LOCAL_RESULT_MAX_BYTES must be at least 1024") } if c.LocalResultMaxTaskBytes != 0 && c.LocalResultMaxTaskBytes < c.LocalResultMaxBytes { return errors.New("AI_GATEWAY_LOCAL_RESULT_MAX_TASK_BYTES must be at least AI_GATEWAY_LOCAL_RESULT_MAX_BYTES") } switch strings.ToLower(strings.TrimSpace(c.IdentitySecretStore)) { case "": case "file": if strings.TrimSpace(c.IdentitySecretDir) == "" { return errors.New("IDENTITY_SECRET_DIR is required for the file SecretStore") } case "kubernetes": if strings.TrimSpace(c.IdentityKubernetesNamespace) == "" || strings.TrimSpace(c.IdentityKubernetesSecretName) == "" { return errors.New("Kubernetes identity SecretStore requires namespace and Secret name") } default: return errors.New("IDENTITY_SECRET_STORE must be file or kubernetes") } if c.IdentitySecurityEventHeartbeatIntervalSeconds != 0 || c.IdentitySecurityEventStaleAfterSeconds != 0 || c.IdentitySecurityEventClockSkewSeconds != 0 { if c.IdentitySecurityEventHeartbeatIntervalSeconds <= 0 || c.IdentitySecurityEventStaleAfterSeconds < 2*c.IdentitySecurityEventHeartbeatIntervalSeconds || c.IdentitySecurityEventClockSkewSeconds < 0 || c.IdentitySecurityEventClockSkewSeconds > 300 { return errors.New("identity security event heartbeat, stale threshold, or clock skew is invalid") } } return nil } func validMediaOSSBucket(value string) bool { value = strings.TrimSpace(value) if len(value) < 3 || len(value) > 63 { return false } for index, item := range value { if (item >= 'a' && item <= 'z') || (item >= '0' && item <= '9') || (item == '-' && index > 0 && index < len(value)-1) { continue } return false } return true } func validMediaOSSURL(value string, appEnv string) bool { parsed, err := url.Parse(strings.TrimSpace(value)) if err != nil || parsed.Host == "" || parsed.RawQuery != "" || parsed.Fragment != "" || parsed.User != nil { return false } if parsed.Scheme == "https" { return true } host := strings.ToLower(parsed.Hostname()) return !strings.EqualFold(strings.TrimSpace(appEnv), "production") && parsed.Scheme == "http" && (host == "127.0.0.1" || host == "localhost" || host == "::1") } func validMediaOSSObjectPrefix(value string) bool { value = strings.Trim(value, "/") if value == "" || len(value) > 256 || strings.Contains(value, "..") { return false } for _, item := range value { if (item >= 'a' && item <= 'z') || (item >= 'A' && item <= 'Z') || (item >= '0' && item <= '9') || item == '-' || item == '_' || item == '/' { continue } return false } return true } func (c Config) EffectiveProcessRole() string { role := strings.ToLower(strings.TrimSpace(c.ProcessRole)) if role == "" { return "all" } return role } func (c Config) RunsPublicHTTP() bool { switch c.EffectiveProcessRole() { case "worker", "capacity-controller": return false default: return true } } func (c Config) RunsAsyncExecutionWorker() bool { switch c.EffectiveProcessRole() { case "api", "capacity-controller": return false case "worker": return true default: return c.AsyncQueueWorkerEnabled } } // RunsAsyncAdmissionDispatcher keeps the legacy all-in-one/Worker behaviour // unless a deployment explicitly separates admission coordination from task // execution. Production enables this on both API sites so the instance nearest // the current PostgreSQL primary wins the existing database scope locks, while // remote Worker replicas only execute River jobs. func (c Config) RunsAsyncAdmissionDispatcher() bool { if c.AsyncAdmissionDispatcherConfigured { return c.AsyncAdmissionDispatcherEnabled } return c.RunsAsyncExecutionWorker() } func (c Config) RunsBackgroundWorkers() bool { switch c.EffectiveProcessRole() { case "api", "capacity-controller": return false default: return true } } func (c Config) RunsCapacityController() bool { return c.EffectiveProcessRole() == "capacity-controller" } func (c Config) RunsRouteProber() bool { return c.RunsAsyncExecutionWorker() && c.RouteProbeEnabled && !strings.EqualFold(c.RoutingMode, "legacy") } func workerAdvertiseEndpoint() string { return strings.TrimRight(strings.TrimSpace(os.Getenv("AI_GATEWAY_WORKER_ADVERTISE_ENDPOINT")), "/") } func validExecutionPoolID(value string) bool { value = strings.TrimSpace(value) if value == "" || len(value) > 128 { return false } for index, character := range value { valid := (character >= 'a' && character <= 'z') || (character >= 'A' && character <= 'Z') || (character >= '0' && character <= '9') || character == '.' || character == '_' || character == ':' || character == '/' || character == '-' if !valid || (index == 0 && !((character >= 'a' && character <= 'z') || (character >= 'A' && character <= 'Z') || (character >= '0' && character <= '9'))) { return false } } return true } func jsonObject(value string) bool { var object map[string]any return json.Unmarshal([]byte(value), &object) == nil && object != nil } type GlobalHTTPProxyStatus struct { HTTPProxy string Source string } type ExecutionPoolCapacityConfig struct { ID string `json:"id"` AdapterRef string `json:"adapterRef"` BootstrapReplicas int `json:"bootstrapReplicas"` MinReplicas int `json:"minReplicas"` MaxReplicas int `json:"maxReplicas"` } func (c Config) CapacityPools() ([]ExecutionPoolCapacityConfig, error) { raw := strings.TrimSpace(c.CapacityPoolsJSON) if raw == "" { return nil, nil } var pools []ExecutionPoolCapacityConfig if err := json.Unmarshal([]byte(raw), &pools); err != nil { return nil, errors.New("AI_GATEWAY_CAPACITY_POOLS must be a JSON array") } if len(pools) == 0 { return nil, errors.New("AI_GATEWAY_CAPACITY_POOLS must contain at least one pool") } seen := make(map[string]struct{}, len(pools)) for index := range pools { pools[index].ID = strings.TrimSpace(pools[index].ID) pools[index].AdapterRef = strings.TrimSpace(pools[index].AdapterRef) if pools[index].AdapterRef == "" { pools[index].AdapterRef = pools[index].ID } if !validExecutionPoolID(pools[index].ID) || pools[index].MinReplicas < 0 || pools[index].MaxReplicas < pools[index].MinReplicas || pools[index].MaxReplicas > 64 || pools[index].BootstrapReplicas < pools[index].MinReplicas || pools[index].BootstrapReplicas > pools[index].MaxReplicas { return nil, errors.New("AI_GATEWAY_CAPACITY_POOLS contains an invalid pool definition") } if _, duplicate := seen[pools[index].ID]; duplicate { return nil, errors.New("AI_GATEWAY_CAPACITY_POOLS contains duplicate pool IDs") } seen[pools[index].ID] = struct{}{} } return pools, nil } func LoadGlobalHTTPProxyStatus() GlobalHTTPProxyStatus { for _, key := range []string{ "AI_GATEWAY_GLOBAL_HTTP_PROXY", "GLOBAL_HTTP_PROXY", "HTTPS_PROXY", "https_proxy", "HTTP_PROXY", "http_proxy", "ALL_PROXY", "all_proxy", } { if value := envValue(key); value != "" { return GlobalHTTPProxyStatus{HTTPProxy: value, Source: key} } } return GlobalHTTPProxyStatus{} } func gatewayDatabaseURL() string { if value := envValue("AI_GATEWAY_DATABASE_URL"); value != "" { return normalizePostgresURL(value) } if value := envValue("DATABASE_URL"); value != "" { return normalizePostgresURL(value) } if memoryURL := envValue("MEMORY_DATABASE_URL"); memoryURL != "" { return normalizePostgresURL(withDatabase(memoryURL, env("AI_GATEWAY_DATABASE_NAME", "easyai_ai_gateway"))) } return normalizePostgresURL("postgresql://easyai:easyai2025@localhost:5432/easyai_ai_gateway?sslmode=disable") } func normalizePostgresURL(raw string) string { parsed, err := url.Parse(raw) if err != nil { return raw } values := parsed.Query() schema := values.Get("schema") if schema == "" { return raw } values.Del("schema") if values.Get("search_path") == "" { values.Set("search_path", schema) } parsed.RawQuery = values.Encode() return parsed.String() } func withDatabase(raw string, databaseName string) string { parsed, err := url.Parse(raw) if err != nil || databaseName == "" { return raw } parsed.Path = "/" + databaseName return parsed.String() } func envValue(key string) string { return strings.TrimSpace(os.Getenv(key)) } func env(key string, fallback string) string { if value := envValue(key); value != "" { return value } return fallback } func envInt(key string, fallback int) int { value := envValue(key) if value == "" { return fallback } parsed, err := strconv.Atoi(value) if err != nil { return fallback } return parsed } func envIntValidated(key string, fallback int) int { value := envValue(key) if value == "" { return fallback } parsed, err := strconv.Atoi(value) if err != nil { return 0 } return parsed } func envOptionalIntValidated(key string, fallback int) int { value := envValue(key) if value == "" { return fallback } parsed, err := strconv.Atoi(value) if err != nil { return -1 } return parsed } func envInt64Validated(key string, fallback int64) int64 { value := envValue(key) if value == "" { return fallback } parsed, err := strconv.ParseInt(value, 10, 64) if err != nil { return 0 } return parsed } func logLevel(value string) slog.Level { switch strings.ToLower(value) { case "debug": return slog.LevelDebug case "warn", "warning": return slog.LevelWarn case "error": return slog.LevelError default: return slog.LevelInfo } }