fix: add embed retry + graceful fallback + proactive cache warming (#77)

- AiService.embedWithRetry(): up to 3 attempts with 1s/2s/4s exponential
  backoff for transient network errors (Premature close, ECONNRESET, etc.)
- FeedService: catch embed failures and return empty feed instead of 500
- UsersService.saveInterests(): fire-and-forget warmEmbedding() after tags
  are saved so the cache is always warm before the next /feed request
- UsersModule: import AiModule so AiService can be injected into UsersService

Fixes: feed returning 500 when user edits interests and /feed is called
before the OpenAI embedding succeeds on a cold Render container

Author: zavxai

apps/api/src/ai/ai.service.ts

@@ -61,14 +61,43 @@ export class AiService {
   }
 
   async embed(text: string): Promise<number[]> {
-    const response = await this.client.embeddings.create({
-      model: 'text-embedding-3-small',
-      input: text,
-    });
+    return this.embedWithRetry(text, 3);
+  }
+
+  private async embedWithRetry(
+    text: string,
+    maxAttempts: number,
+    attempt = 1,
+  ): Promise<number[]> {
+    try {
+      const response = await this.client.embeddings.create({
+        model: 'text-embedding-3-small',
+        input: text,
+      });
+      const embedding = response.data[0]?.embedding;
+      if (!embedding) throw new Error('OpenAI embeddings returned no data');
+      return embedding;
+    } catch (err: unknown) {
+      const msg = err instanceof Error ? err.message : String(err);
+      const isTransient =
+        /premature close|connection error|econnreset|econnrefused|socket hang up|network/i.test(
+          msg,
+        );
 
-    const embedding = response.data[0]?.embedding;
-    if (!embedding) throw new Error('OpenAI embeddings returned no data');
-    return embedding;
+      if (isTransient && attempt < maxAttempts) {
+        const delayMs = Math.pow(2, attempt - 1) * 1000; // 1 s, 2 s, 4 s
+        this.logger.warn(
+          `embed() attempt ${attempt} failed (${msg}). Retrying in ${delayMs}ms…`,
+        );
+        await new Promise((r) => setTimeout(r, delayMs));
+        return this.embedWithRetry(text, maxAttempts, attempt + 1);
+      }
+
+      this.logger.error(
+        `embed() failed after ${attempt} attempt(s): ${msg}`,
+      );
+      throw err;
+    }
   }
 
   async processUnsummarized(

apps/api/src/feed/feed.service.ts

@@ -85,12 +85,19 @@ export class FeedService {
       this.logger.log(
         `Generating embedding for user ${userId} | query: "${queryText}"`,
       );
-      embedding = await this.aiService.embed(queryText);
-      if (interestRow.length > 0) {
-        await this.db
-          .update(userInterests)
-          .set({ queryEmbedding: embedding })
-          .where(eq(userInterests.userId, userId));
+      try {
+        embedding = await this.aiService.embed(queryText);
+        if (interestRow.length > 0) {
+          await this.db
+            .update(userInterests)
+            .set({ queryEmbedding: embedding })
+            .where(eq(userInterests.userId, userId));
+        }
+      } catch (err) {
+        this.logger.error(
+          `Failed to generate embedding for user ${userId} — returning empty feed: ${err}`,
+        );
+        return { hasMatches: false, articles: [], fallback: [] };
       }
     }
 
@@ -227,12 +234,19 @@ export class FeedService {
       this.logger.log(
         `Generating embedding for user ${userId} (debug) | query: "${queryText}"`,
       );
-      embedding = await this.aiService.embed(queryText);
-      if (interestRow.length > 0) {
-        await this.db
-          .update(userInterests)
-          .set({ queryEmbedding: embedding })
-          .where(eq(userInterests.userId, userId));
+      try {
+        embedding = await this.aiService.embed(queryText);
+        if (interestRow.length > 0) {
+          await this.db
+            .update(userInterests)
+            .set({ queryEmbedding: embedding })
+            .where(eq(userInterests.userId, userId));
+        }
+      } catch (err) {
+        this.logger.error(
+          `Failed to generate embedding for user ${userId} (debug): ${err}`,
+        );
+        return { queryText, articles: [] };
       }
     }
 

apps/api/src/users/users.module.ts

@@ -2,9 +2,10 @@ import { Module, forwardRef } from '@nestjs/common';
 import { UsersService } from './users.service';
 import { UsersController } from './users.controller';
 import { AuthModule } from '../auth/auth.module';
+import { AiModule } from '../ai/ai.module';
 
 @Module({
-  imports: [forwardRef(() => AuthModule)],
+  imports: [forwardRef(() => AuthModule), AiModule],
   providers: [UsersService],
   controllers: [UsersController],
   exports: [UsersService],

apps/api/src/users/users.service.ts

@@ -1,12 +1,18 @@
-import { Inject, Injectable } from '@nestjs/common';
+import { Inject, Injectable, Logger } from '@nestjs/common';
 import { eq } from 'drizzle-orm';
 import { DRIZZLE } from '../db/drizzle.provider';
 import type { DrizzleDB } from '../db/drizzle.provider';
 import { users, userInterests, User, NewUser } from '../db/schema';
+import { AiService } from '../ai/ai.service';
 
 @Injectable()
 export class UsersService {
-  constructor(@Inject(DRIZZLE) private db: DrizzleDB) {}
+  private readonly logger = new Logger(UsersService.name);
+
+  constructor(
+    @Inject(DRIZZLE) private db: DrizzleDB,
+    private readonly aiService: AiService,
+  ) {}
 
   async upsert(googleProfile: {
     id: string;
@@ -78,6 +84,24 @@ export class UsersService {
         target: userInterests.userId,
         set: { tags, queryEmbedding: null },
       });
+
+    // Fire-and-forget: warm the embedding cache so the next /feed request
+    // doesn't have to call OpenAI cold (which risks a transient 500).
+    this.warmEmbedding(userId, tags).catch((err) =>
+      this.logger.warn(`Background embed warming failed for user ${userId}: ${err}`),
+    );
+  }
+
+  private async warmEmbedding(userId: string, tags: string[]): Promise<void> {
+    if (tags.length === 0) return;
+    const queryText = `software engineering articles about ${tags.join(', ')} for developers`;
+    this.logger.log(`Warming embedding for user ${userId} after interests update`);
+    const embedding = await this.aiService.embed(queryText);
+    await this.db
+      .update(userInterests)
+      .set({ queryEmbedding: embedding })
+      .where(eq(userInterests.userId, userId));
+    this.logger.log(`Embedding warmed for user ${userId}`);
   }
 
   async getInterests(userId: string): Promise<{ tags: string[] }> {