@loonylabs/tti-middleware 1.5.1 → 1.7.0

package/README.md

@@ -45,6 +45,7 @@
   - **IONOS**: German cloud provider with OpenAI-compatible API (experimental)
 - **Character Consistency**: Generate consistent characters across multiple images (perfect for children's book illustrations)
 - **GDPR/DSGVO Compliance**: Built-in EU region support with automatic fallback
+- **Region Rotation**: Opt-in region rotation on quota errors (429) for Google Cloud — rotate through regions instead of retrying the same exhausted region
 - **Retry Logic**: Exponential backoff with jitter for transient errors (429, 408, 5xx, timeouts)
 - **TypeScript First**: Full type safety with comprehensive interfaces
 - **Logging Control**: Configurable log levels via environment or API
@@ -400,6 +401,31 @@ interface TTIResponse {
 
 ## Advanced Features
 
+<details>
+<summary><strong>Region Rotation (Google Cloud)</strong></summary>
+
+When Vertex AI returns 429 (Resource Exhausted) due to Dynamic Shared Quota, the middleware can rotate through a list of regions instead of retrying the same exhausted region:
+
+```typescript
+const provider = new GoogleCloudTTIProvider({
+  projectId: 'my-project',
+  region: 'europe-west4',
+  regionRotation: {
+    regions: ['europe-west4', 'europe-west1', 'europe-north1', 'europe-central2'],
+    fallback: 'global',
+    alwaysTryFallback: true, // Default: one bonus attempt on fallback after budget exhausted
+  },
+});
+```
+
+**Key behavior:**
+- `maxRetries` is the **total budget** across all regions (no multiplier)
+- Only **quota errors** (429, Resource Exhausted) trigger rotation — server errors (500, 503) retry the same region
+- `alwaysTryFallback: true` (default): one bonus attempt on fallback even if retry budget is exhausted
+- Without `regionRotation`: existing behavior unchanged
+
+</details>
+
 <details>
 <summary><strong>Retry Configuration</strong></summary>
 

package/dist/middleware/services/tti/providers/base-tti-provider.d.ts

@@ -112,6 +112,12 @@ export declare abstract class BaseTTIProvider implements ITTIProvider {
      * Check if an error is a timeout error (from our withTimeout wrapper).
      */
     private isTimeoutError;
+    /**
+     * Check if an error is a quota/rate-limit error (429 / Resource Exhausted).
+     * Used by providers to distinguish quota errors from other retryable errors
+     * (e.g., for region rotation on quota errors only).
+     */
+    protected isQuotaError(error: Error): boolean;
     /**
      * Execute a generation function with retry logic for transient errors.
      * Retries on: 429, 408, 5xx, network timeouts, TCP disconnects.
@@ -121,8 +127,15 @@ export declare abstract class BaseTTIProvider implements ITTIProvider {
      * retry.timeoutMs, default 45s). Timeout errors have their own retry
      * counter (timeoutRetries, default 2) independent from the general
      * maxRetries used for quota/server errors.
-     */
-    protected executeWithRetry<T>(request: TTIRequest, operation: () => Promise<T>, operationName: string): Promise<T>;
+     *
+     * @param options.onRetry - Optional callback invoked before each retry.
+     *   Receives the error that triggered the retry and the current general
+     *   retry count. Providers can use this to adjust state between retries
+     *   (e.g., rotate regions on quota errors).
+     */
+    protected executeWithRetry<T>(request: TTIRequest, operation: () => Promise<T>, operationName: string, options?: {
+        onRetry?: (error: Error, generalRetryCount: number) => void;
+    }): Promise<T>;
     /**
      * Check if an error is retryable (transient).
      * Retryable: 429, 408, 500, 502, 503, 504, network errors, timeouts.

package/dist/middleware/services/tti/providers/base-tti-provider.js

@@ -301,6 +301,19 @@ class BaseTTIProvider {
     isTimeoutError(error) {
         return error.message.toLowerCase().startsWith('timeout:');
     }
+    /**
+     * Check if an error is a quota/rate-limit error (429 / Resource Exhausted).
+     * Used by providers to distinguish quota errors from other retryable errors
+     * (e.g., for region rotation on quota errors only).
+     */
+    isQuotaError(error) {
+        const message = error.message.toLowerCase();
+        return (message.includes('429') ||
+            message.includes('resource exhausted') ||
+            message.includes('quota exceeded') ||
+            message.includes('rate limit') ||
+            message.includes('too many requests'));
+    }
     /**
      * Execute a generation function with retry logic for transient errors.
      * Retries on: 429, 408, 5xx, network timeouts, TCP disconnects.
@@ -310,8 +323,13 @@ class BaseTTIProvider {
      * retry.timeoutMs, default 45s). Timeout errors have their own retry
      * counter (timeoutRetries, default 2) independent from the general
      * maxRetries used for quota/server errors.
+     *
+     * @param options.onRetry - Optional callback invoked before each retry.
+     *   Receives the error that triggered the retry and the current general
+     *   retry count. Providers can use this to adjust state between retries
+     *   (e.g., rotate regions on quota errors).
      */
-    async executeWithRetry(request, operation, operationName) {
+    async executeWithRetry(request, operation, operationName, options) {
         const retryConfig = this.resolveRetryConfig(request);
         // No retry configured
         if (!retryConfig) {
@@ -371,6 +389,10 @@ class BaseTTIProvider {
                         this.log('error', `${operationName} general retry budget exhausted (${maxGeneralRetries} retries): ${error.message}`, { attempt, generalRetryCount, durationMs: duration });
                         throw error;
                     }
+                    // Notify provider before retry (e.g., for region rotation)
+                    if (options?.onRetry) {
+                        options.onRetry(error, generalRetryCount);
+                    }
                     const delay = this.calculateRetryDelay(generalRetryCount, retryConfig);
                     this.log('warn', `Transient error during ${operationName} after ${duration}ms. Retry ${generalRetryCount}/${maxGeneralRetries} in ${delay}ms: ${error.message}`, { attempt, generalRetryCount, maxGeneralRetries, delayMs: delay, durationMs: duration });
                     await this.sleep(delay);

package/dist/middleware/services/tti/providers/google-cloud-provider.d.ts

@@ -8,6 +8,7 @@
  * - Imagen 4 Ultra (imagen-4.0-ultra-generate-001) - Highest quality variant
  * - Gemini 2.5 Flash Image - Text-to-image with character consistency
  * - Gemini 3 Pro Image (gemini-3-pro-image-preview) - 4K, text rendering
+ * - Gemini 3.1 Flash Image (gemini-3.1-flash-image-preview) - 4K, improved text rendering (global endpoint)
  *
  * All requests go through Google Cloud (Vertex AI) with proper DPA.
  * EU-compliant when using EU regions.
@@ -15,12 +16,12 @@
  * @see https://cloud.google.com/vertex-ai/generative-ai/pricing
  * @see https://cloud.google.com/terms/data-processing-addendum
  */
-import { TTIRequest, TTIResponse, ModelInfo, GoogleCloudRegion } from '../../../types';
+import { TTIRequest, TTIResponse, ModelInfo, GoogleCloudRegion, RegionRotationConfig } from '../../../types';
 import { BaseTTIProvider } from './base-tti-provider';
 interface GoogleCloudConfig {
     /** Google Cloud Project ID */
     projectId: string;
-    /** Default region for requests */
+    /** Default region for requests (used when regionRotation is not configured) */
     region: GoogleCloudRegion;
     /** Path to service account JSON file */
     keyFilename?: string;
@@ -30,11 +31,17 @@ interface GoogleCloudConfig {
         private_key: string;
         project_id?: string;
     };
+    /**
+     * Opt-in region rotation for quota errors (429 / Resource Exhausted).
+     * When configured, the middleware rotates through the listed regions
+     * on quota errors instead of retrying the same region.
+     */
+    regionRotation?: RegionRotationConfig;
 }
 export declare class GoogleCloudTTIProvider extends BaseTTIProvider {
     private config;
     private lastUsedRegion;
-    private aiplatformClient;
+    private aiplatformClients;
     private genaiClients;
     constructor(config?: Partial<GoogleCloudConfig>);
     getDisplayName(): string;

package/dist/middleware/services/tti/providers/google-cloud-provider.js

@@ -9,6 +9,7 @@
  * - Imagen 4 Ultra (imagen-4.0-ultra-generate-001) - Highest quality variant
  * - Gemini 2.5 Flash Image - Text-to-image with character consistency
  * - Gemini 3 Pro Image (gemini-3-pro-image-preview) - 4K, text rendering
+ * - Gemini 3.1 Flash Image (gemini-3.1-flash-image-preview) - 4K, improved text rendering (global endpoint)
  *
  * All requests go through Google Cloud (Vertex AI) with proper DPA.
  * EU-compliant when using EU regions.
@@ -163,6 +164,20 @@ const GOOGLE_CLOUD_MODELS = [
         availableRegions: ['global'],
         pricingUrl: 'https://cloud.google.com/vertex-ai/generative-ai/pricing',
     },
+    {
+        id: 'gemini-flash-image-2',
+        displayName: 'Gemini 3.1 Flash Image',
+        capabilities: {
+            textToImage: true,
+            characterConsistency: true, // Up to 5 characters + 14 objects
+            imageEditing: false,
+            maxImagesPerRequest: 1,
+        },
+        // Preview model — requires global endpoint (same as gemini-pro-image).
+        // Will likely get regional endpoints once GA.
+        availableRegions: ['global'],
+        pricingUrl: 'https://cloud.google.com/vertex-ai/generative-ai/pricing',
+    },
 ];
 // Internal model IDs used in Vertex AI API calls
 const MODEL_ID_MAP = {
@@ -172,9 +187,10 @@ const MODEL_ID_MAP = {
     'imagen-4-ultra': 'imagen-4.0-ultra-generate-001',
     'gemini-flash-image': 'gemini-2.5-flash-image',
     'gemini-pro-image': 'gemini-3-pro-image-preview',
+    'gemini-flash-image-2': 'gemini-3.1-flash-image-preview',
 };
 // Models that use the Gemini generateContent API (vs Imagen predict API)
-const GEMINI_API_MODELS = new Set(['gemini-flash-image', 'gemini-pro-image']);
+const GEMINI_API_MODELS = new Set(['gemini-flash-image', 'gemini-pro-image', 'gemini-flash-image-2']);
 // ============================================================
 // PROVIDER IMPLEMENTATION
 // ============================================================
@@ -182,8 +198,8 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
     constructor(config) {
         super(types_1.TTIProvider.GOOGLE_CLOUD);
         this.lastUsedRegion = null;
-        // Lazy-loaded SDK clients
-        this.aiplatformClient = null;
+        // Lazy-loaded SDK clients (one per region, since region is baked into the client)
+        this.aiplatformClients = new Map();
         // eslint-disable-next-line @typescript-eslint/no-explicit-any
         this.genaiClients = new Map();
         const projectId = config?.projectId ||
@@ -202,10 +218,27 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
             region,
             keyFilename: config?.keyFilename || process.env.GOOGLE_APPLICATION_CREDENTIALS,
             credentials: config?.credentials,
+            regionRotation: config?.regionRotation,
         };
+        // Validate regionRotation config
+        if (this.config.regionRotation) {
+            if (!this.config.regionRotation.regions || this.config.regionRotation.regions.length === 0) {
+                throw new base_tti_provider_1.InvalidConfigError(types_1.TTIProvider.GOOGLE_CLOUD, 'regionRotation.regions must contain at least one region');
+            }
+            if (!this.config.regionRotation.fallback) {
+                throw new base_tti_provider_1.InvalidConfigError(types_1.TTIProvider.GOOGLE_CLOUD, 'regionRotation.fallback is required');
+            }
+        }
         this.log('info', 'Google Cloud TTI Provider initialized', {
             projectId: this.config.projectId,
             region: this.config.region,
+            regionRotation: this.config.regionRotation
+                ? {
+                    regions: this.config.regionRotation.regions,
+                    fallback: this.config.regionRotation.fallback,
+                    alwaysTryFallback: this.config.regionRotation.alwaysTryFallback ?? true,
+                }
+                : undefined,
             isEURegion: (0, base_tti_provider_1.isEURegion)(this.config.region),
             models: this.listModels().map((m) => m.id),
         });
@@ -232,29 +265,66 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
                 .map((m) => m.id)
                 .join(', ')}`);
         }
-        // Validate region availability
-        const effectiveRegion = this.getEffectiveRegion(modelId);
+        // Determine base region (handles global-only models, region availability)
+        const baseRegion = this.getEffectiveRegion(modelId);
+        // Region rotation: only for non-global models with rotation configured
+        const rotation = this.config.regionRotation;
+        const useRotation = !!(rotation && baseRegion !== 'global');
+        // Mutable region — the onRetry callback advances this on quota errors
+        let currentRegion = useRotation ? rotation.regions[0] : baseRegion;
+        // Build region sequence: [...regions, fallback]
+        let regionIndex = 0;
+        let regionSequence = [];
+        if (useRotation) {
+            regionSequence = [...rotation.regions, rotation.fallback];
+            this.log('info', 'Region rotation enabled', {
+                sequence: regionSequence,
+                fallback: rotation.fallback,
+                alwaysTryFallback: rotation.alwaysTryFallback ?? true,
+            });
+        }
         // Create debug info for logging
         let debugInfo = null;
         if (debug_tti_utils_1.TTIDebugger.isEnabled) {
-            debugInfo = debug_tti_utils_1.TTIDebugger.createDebugInfo(request, this.providerName, modelId, { region: effectiveRegion });
+            debugInfo = debug_tti_utils_1.TTIDebugger.createDebugInfo(request, this.providerName, modelId, { region: currentRegion });
             await debug_tti_utils_1.TTIDebugger.logRequest(debugInfo);
         }
         this.log('debug', 'Generating image', {
             model: modelId,
-            region: effectiveRegion,
+            region: currentRegion,
+            regionRotation: useRotation,
             hasReferenceImages: (0, base_tti_provider_1.hasReferenceImages)(request),
         });
-        try {
-            // Route to appropriate API based on model type
-            let response;
-            if (GEMINI_API_MODELS.has(modelId)) {
-                response = await this.executeWithRetry(request, () => this.generateWithGemini(request, modelId, effectiveRegion), 'Gemini API call');
+        const isGeminiModel = GEMINI_API_MODELS.has(modelId);
+        const operationName = isGeminiModel ? 'Gemini API call' : 'Imagen API call';
+        // Operation lambda reads currentRegion from closure
+        const operation = () => {
+            if (isGeminiModel) {
+                return this.generateWithGemini(request, modelId, currentRegion);
             }
             else {
-                response = await this.executeWithRetry(request, () => this.generateWithImagen(request, modelId, effectiveRegion), 'Imagen API call');
+                return this.generateWithImagen(request, modelId, currentRegion);
             }
-            // Log successful response
+        };
+        // onRetry: advance region on quota errors, stay on same region otherwise
+        const onRetry = useRotation
+            ? (error) => {
+                if (this.isQuotaError(error) && regionIndex < regionSequence.length - 1) {
+                    regionIndex++;
+                    currentRegion = regionSequence[regionIndex];
+                    this.log('info', `Quota error — rotating to region ${currentRegion}`, {
+                        regionIndex,
+                        totalRegions: regionSequence.length,
+                        region: currentRegion,
+                    });
+                }
+                // Non-quota retryable errors: stay on same region
+            }
+            : undefined;
+        try {
+            const response = await this.executeWithRetry(request, operation, operationName, {
+                onRetry,
+            });
             if (debugInfo) {
                 debugInfo = debug_tti_utils_1.TTIDebugger.updateWithResponse(debugInfo, response);
                 await debug_tti_utils_1.TTIDebugger.logResponse(debugInfo);
@@ -262,7 +332,32 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
             return response;
         }
         catch (error) {
-            // Log error
+            // alwaysTryFallback: one bonus attempt on fallback after budget exhausted
+            if (useRotation &&
+                this.isQuotaError(error) &&
+                (rotation.alwaysTryFallback !== false) &&
+                currentRegion !== rotation.fallback) {
+                this.log('info', `Retry budget exhausted — bonus attempt on fallback region ${rotation.fallback}`, {
+                    exhaustedRegion: currentRegion,
+                    fallback: rotation.fallback,
+                });
+                currentRegion = rotation.fallback;
+                try {
+                    const response = await operation();
+                    if (debugInfo) {
+                        debugInfo = debug_tti_utils_1.TTIDebugger.updateWithResponse(debugInfo, response);
+                        await debug_tti_utils_1.TTIDebugger.logResponse(debugInfo);
+                    }
+                    return response;
+                }
+                catch (fallbackError) {
+                    if (debugInfo) {
+                        debugInfo = debug_tti_utils_1.TTIDebugger.updateWithError(debugInfo, fallbackError);
+                        await debug_tti_utils_1.TTIDebugger.logError(debugInfo);
+                    }
+                    throw fallbackError;
+                }
+            }
             if (debugInfo) {
                 debugInfo = debug_tti_utils_1.TTIDebugger.updateWithError(debugInfo, error);
                 await debug_tti_utils_1.TTIDebugger.logError(debugInfo);
@@ -325,7 +420,7 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
         const internalModelId = MODEL_ID_MAP[modelId];
         this.lastUsedRegion = region;
         try {
-            const { client, helpers } = await this.getAiplatformClient();
+            const { client, helpers } = await this.getAiplatformClient(region);
             const endpoint = `projects/${this.config.projectId}/locations/${region}/publishers/google/models/${internalModelId}`;
             // Build instance
             const instanceValue = { prompt: request.prompt };
@@ -370,12 +465,12 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
             throw this.handleError(error, 'during Imagen API call');
         }
     }
-    async getAiplatformClient() {
-        if (!this.aiplatformClient) {
+    async getAiplatformClient(region) {
+        if (!this.aiplatformClients.has(region)) {
             try {
                 const { v1, helpers } = await Promise.resolve().then(() => __importStar(require('@google-cloud/aiplatform')));
                 const clientOptions = {
-                    apiEndpoint: `${this.config.region}-aiplatform.googleapis.com`,
+                    apiEndpoint: `${region}-aiplatform.googleapis.com`,
                 };
                 if (this.config.keyFilename) {
                     clientOptions.keyFilename = this.config.keyFilename;
@@ -384,9 +479,13 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
                     clientOptions.credentials = this.config.credentials;
                 }
                 // eslint-disable-next-line @typescript-eslint/no-explicit-any
-                this.aiplatformClient = new v1.PredictionServiceClient(clientOptions);
+                this.aiplatformClients.set(region, new v1.PredictionServiceClient(clientOptions));
+                this.log('debug', 'Initialized @google-cloud/aiplatform client', {
+                    region,
+                    apiEndpoint: clientOptions.apiEndpoint,
+                });
                 return {
-                    client: this.aiplatformClient,
+                    client: this.aiplatformClients.get(region),
                     helpers: helpers,
                 };
             }
@@ -396,7 +495,7 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
         }
         const { helpers } = await Promise.resolve().then(() => __importStar(require('@google-cloud/aiplatform')));
         return {
-            client: this.aiplatformClient,
+            client: this.aiplatformClients.get(region),
             helpers: helpers,
         };
     }
@@ -470,11 +569,17 @@ class GoogleCloudTTIProvider extends base_tti_provider_1.BaseTTIProvider {
             const config = {
                 responseModalities: ['TEXT', 'IMAGE'],
             };
-            // Add imageConfig with aspectRatio if provided
-            if (request.aspectRatio) {
-                config.imageConfig = {
-                    aspectRatio: request.aspectRatio,
-                };
+            // Add imageConfig with aspectRatio and/or imageSize if provided
+            if (request.aspectRatio || request.providerOptions?.imageSize) {
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                const imageConfig = {};
+                if (request.aspectRatio) {
+                    imageConfig.aspectRatio = request.aspectRatio;
+                }
+                if (request.providerOptions?.imageSize) {
+                    imageConfig.imageSize = request.providerOptions.imageSize;
+                }
+                config.imageConfig = imageConfig;
             }
             // Add temperature if provided
             if (request.providerOptions?.temperature !== undefined) {

package/dist/middleware/types/index.d.ts

@@ -50,7 +50,29 @@ export interface ModelInfo {
  * Google Cloud regions
  * EU regions are GDPR-compliant
  */
-export type GoogleCloudRegion = 'global' | 'europe-west1' | 'europe-west2' | 'europe-west3' | 'europe-west4' | 'europe-west9' | 'us-central1' | 'us-east4';
+export type GoogleCloudRegion = 'global' | 'europe-west1' | 'europe-west2' | 'europe-west3' | 'europe-west4' | 'europe-west6' | 'europe-west8' | 'europe-west9' | 'europe-north1' | 'europe-central2' | 'europe-southwest1' | 'us-central1' | 'us-east1' | 'us-east4' | 'us-east5' | 'us-south1' | 'us-west1' | 'us-west4' | 'asia-east1' | 'asia-east2' | 'asia-northeast1' | 'asia-northeast3' | 'asia-south1' | 'asia-southeast1' | 'australia-southeast1' | 'me-central1' | 'me-central2' | 'me-west1';
+/**
+ * Configuration for region rotation on quota errors (429 / Resource Exhausted).
+ *
+ * When Vertex AI returns a quota error, the middleware rotates through the
+ * configured regions instead of retrying the same region. This is useful
+ * when Dynamic Shared Quota is temporarily exhausted in a single region.
+ *
+ * The total retry budget (from RetryOptions.maxRetries) is shared across
+ * all regions — region rotation does NOT multiply the retry count.
+ */
+export interface RegionRotationConfig {
+    /** Ordered list of regions to try. First entry = primary region. */
+    regions: GoogleCloudRegion[];
+    /** Last-resort region after all regions exhausted (typically 'global'). */
+    fallback: GoogleCloudRegion;
+    /**
+     * If true: when maxRetries is exhausted before reaching the fallback,
+     * one final bonus attempt on the fallback region is made.
+     * @default true
+     */
+    alwaysTryFallback?: boolean;
+}
 /**
  * Reference image for character consistency
  */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@loonylabs/tti-middleware",
-  "version": "1.5.1",
+  "version": "1.7.0",
   "description": "Provider-agnostic Text-to-Image middleware with GDPR compliance. Supports Google Cloud (Imagen, Gemini), Eden AI, and IONOS.",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",