fix: retry embedding 5xx errors

CHANGELOG.md

@@ -7,6 +7,9 @@ Docs: https://docs.clawd.bot
 ### Changes
 - Memory: add OpenAI Batch API indexing for embeddings when configured.
 
+### Fixes
+- Memory: retry transient 5xx errors (Cloudflare) during embedding indexing.
+
 ## 2026.1.17-2
 
 ### Changes

src/memory/manager.batch.test.ts

@@ -138,11 +138,18 @@ describe("memory indexing with OpenAI batches", () => {
     expect(result.manager).not.toBeNull();
     if (!result.manager) throw new Error("manager missing");
     manager = result.manager;
-    await manager.sync({ force: true });
+    const labels: string[] = [];
+    await manager.sync({
+      force: true,
+      progress: (update) => {
+        if (update.label) labels.push(update.label);
+      },
+    });
 
     const status = manager.status();
     expect(status.chunks).toBeGreaterThan(0);
     expect(embedBatch).not.toHaveBeenCalled();
     expect(fetchMock).toHaveBeenCalled();
+    expect(labels.some((label) => label.toLowerCase().includes("batch"))).toBe(true);
   });
 });

src/memory/manager.embedding-batches.test.ts

@@ -192,6 +192,47 @@ describe("memory embedding batches", () => {
     expect(calls).toBe(3);
   }, 10000);
 
+  it("retries embeddings on transient 5xx errors", async () => {
+    const line = "e".repeat(120);
+    const content = Array.from({ length: 12 }, () => line).join("\n");
+    await fs.writeFile(path.join(workspaceDir, "memory", "2026-01-08.md"), content);
+
+    let calls = 0;
+    embedBatch.mockImplementation(async (texts: string[]) => {
+      calls += 1;
+      if (calls < 3) {
+        throw new Error("openai embeddings failed: 502 Bad Gateway (cloudflare)");
+      }
+      return texts.map(() => [0, 1, 0]);
+    });
+
+    const cfg = {
+      agents: {
+        defaults: {
+          workspace: workspaceDir,
+          memorySearch: {
+            provider: "openai",
+            model: "mock-embed",
+            store: { path: indexPath },
+            chunking: { tokens: 200, overlap: 0 },
+            sync: { watch: false, onSessionStart: false, onSearch: false },
+            query: { minScore: 0 },
+          },
+        },
+        list: [{ id: "main", default: true }],
+      },
+    };
+
+    const result = await getMemorySearchManager({ cfg, agentId: "main" });
+    expect(result.manager).not.toBeNull();
+    if (!result.manager) throw new Error("manager missing");
+    manager = result.manager;
+
+    await manager.sync({ force: true });
+
+    expect(calls).toBe(3);
+  }, 10000);
+
   it("skips empty chunks so embeddings input stays valid", async () => {
     await fs.writeFile(path.join(workspaceDir, "memory", "2026-01-07.md"), "\n\n\n");
 

src/memory/manager.ts

@@ -733,7 +733,7 @@ export class MemoryIndexManager {
       params.progress.report({
         completed: params.progress.completed,
         total: params.progress.total,
-        label: "Indexing memory files…",
+        label: this.batch.enabled ? "Indexing memory files (batch)..." : "Indexing memory files…",
       });
     }
 
@@ -784,7 +784,7 @@ export class MemoryIndexManager {
       params.progress.report({
         completed: params.progress.completed,
         total: params.progress.total,
-        label: "Indexing session files…",
+        label: this.batch.enabled ? "Indexing session files (batch)..." : "Indexing session files…",
       });
     }
 
@@ -1357,7 +1357,7 @@ export class MemoryIndexManager {
         return await this.provider.embedBatch(texts);
       } catch (err) {
         const message = err instanceof Error ? err.message : String(err);
-        if (!this.isRateLimitError(message) || attempt >= EMBEDDING_RETRY_MAX_ATTEMPTS) {
+        if (!this.isRetryableEmbeddingError(message) || attempt >= EMBEDDING_RETRY_MAX_ATTEMPTS) {
           throw err;
         }
         const waitMs = Math.min(
@@ -1372,8 +1372,10 @@ export class MemoryIndexManager {
     }
   }
 
-  private isRateLimitError(message: string): boolean {
+  private isRetryableEmbeddingError(message: string): boolean {
-    return /(rate[_ ]limit|too many requests|429|resource has been exhausted)/i.test(message);
+    return /(rate[_ ]limit|too many requests|429|resource has been exhausted|5\d\d|cloudflare)/i.test(
+      message,
+    );
   }
 
   private async runWithConcurrency<T>(tasks: Array<() => Promise<T>>, limit: number): Promise<T[]> {