fix(POSTV1-05): 修复生成请求与 Worker 重入

apps/worker/src/generation-polling-loop.ts

@@ -0,0 +1,35 @@
+export interface GenerationPollingProcessor {
+  processNext(): Promise<unknown>;
+}
+
+export class GenerationPollingLoop {
+  private closed = false;
+  private inFlight = false;
+  private readonly timer: ReturnType<typeof setInterval>;
+
+  constructor(
+    private readonly processor: GenerationPollingProcessor,
+    intervalMilliseconds = 250,
+  ) {
+    if (!Number.isSafeInteger(intervalMilliseconds) || intervalMilliseconds <= 0) {
+      throw new Error("generation_polling_interval_invalid");
+    }
+    this.timer = setInterval(() => this.run(), intervalMilliseconds);
+  }
+
+  close() {
+    if (this.closed) return;
+    this.closed = true;
+    clearInterval(this.timer);
+  }
+
+  private run() {
+    if (this.closed || this.inFlight) return;
+    this.inFlight = true;
+    void this.processor.processNext()
+      .catch(() => undefined)
+      .finally(() => {
+        this.inFlight = false;
+      });
+  }
+}

apps/worker/src/oneapi-generation-adapter.ts

@@ -16,6 +16,7 @@ const gptImageEndpoint = "https://oneapi.intelligrow.cn/v1/images/generations";
 const gptImageReferenceEndpoint = "https://oneapi.intelligrow.cn/v1/images/edits";
 const maximumResponseBytes = 32 * 1024 * 1024;
 const requestTimeoutMilliseconds = 180_000;
+const geminiImageSystemInstruction = "Generate exactly one image from the user's description. Return the generated image and do not answer with text only.";
 
 type FetchLike = typeof fetch;
 
@@ -141,7 +142,10 @@ function buildRequest(request: GenerationAdapterRequest) {
     return {
       body: JSON.stringify({
         extra_body: { google: { image_config: { aspect_ratio: request.ratio, image_size: "1K" } } },
-        messages: [{ content, role: "user" }],
+        messages: [
+          { content: geminiImageSystemInstruction, role: "system" },
+          { content, role: "user" },
+        ],
         model: geminiProviderModelId,
         stream: false,
       }),

apps/worker/src/worker.ts

@@ -3,6 +3,7 @@ import { join } from "node:path";
 
 import { WorkerAiCallGate } from "./ai-call-gate.js";
 import { runAiRuntimeProbe } from "./ai-runtime-probe.js";
+import { GenerationPollingLoop } from "./generation-polling-loop.js";
 import { GenerationProcessor } from "./generation-processor.js";
 import { OneApiGenerationAdapter } from "./oneapi-generation-adapter.js";
 import { readConfiguredLocalDataRoot } from "./runtime-config.js";
@@ -52,13 +53,13 @@ if (!workerPort && process.argv.includes("--dada-ai-probe")) {
   let projectCleanup: ProjectPurgeCleanup | undefined;
   let retentionTimer: ReturnType<typeof setInterval> | undefined;
   let processor: GenerationProcessor | undefined;
-  let generationTimer: ReturnType<typeof setInterval> | undefined;
+  let generationLoop: GenerationPollingLoop | undefined;
   const control = attachWorkerSupervisorControl(controlPipe, () => {
     clearInterval(keepAlive);
     if (retentionTimer) clearInterval(retentionTimer);
     retention?.close();
     projectCleanup?.close();
-    if (generationTimer) clearInterval(generationTimer);
+    generationLoop?.close();
     processor?.close();
     storage?.close();
   });
@@ -102,7 +103,7 @@ if (!workerPort && process.argv.includes("--dada-ai-probe")) {
     });
     logger.write({ error_category: "none", status_category: "ready" });
     new WorkerAiCallGate({ getStorageStatus: () => storageStatus === "unavailable" ? storageStatus : (storage?.getStatus() ?? "unavailable"), logger });
-    generationTimer = setInterval(() => { void processor?.processNext().catch(() => undefined); }, 250);
+    generationLoop = new GenerationPollingLoop(processor);
   } catch {
     storageStatus = "unavailable";
     control.reportStatus("storage_unavailable");

tests/worker/postv1-generation-polling-loop.test.ts

@@ -0,0 +1,28 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+import { GenerationPollingLoop } from "../../apps/worker/src/generation-polling-loop.js";
+
+describe("POSTV1-05 generation polling loop", () => {
+  afterEach(() => {
+    vi.useRealTimers();
+  });
+
+  it("does not start another processor call while the current call is unresolved", async () => {
+    vi.useFakeTimers();
+    let finishCurrentCall: (() => void) | undefined;
+    const processNext = vi.fn(() => new Promise<void>((resolve) => {
+      finishCurrentCall = resolve;
+    }));
+    const loop = new GenerationPollingLoop({ processNext }, 250);
+
+    await vi.advanceTimersByTimeAsync(1_000);
+    expect(processNext).toHaveBeenCalledTimes(1);
+
+    finishCurrentCall?.();
+    await Promise.resolve();
+    await vi.advanceTimersByTimeAsync(250);
+    expect(processNext).toHaveBeenCalledTimes(2);
+
+    loop.close();
+  });
+});

tests/worker/postv1-oneapi-runtime.test.ts

@@ -26,6 +26,14 @@ describe("POSTV1-02 OneAPI runtime adapter", () => {
       const headers = new Headers(init?.headers);
       expect(headers.get("authorization")).toBe("Bearer synthetic-runtime-token");
       expect(init?.redirect).toBe("error");
+      const payload = JSON.parse(String(init?.body)) as { messages: Array<{ content: unknown; role: string }> };
+      expect(payload.messages).toEqual([
+        {
+          content: "Generate exactly one image from the user's description. Return the generated image and do not answer with text only.",
+          role: "system",
+        },
+        { content: "一张用于本机验收的抽象色彩图", role: "user" },
+      ]);
       return new Response(JSON.stringify({
         choices: [{ message: { content: `![result](data:image/png;base64,${source.toString("base64")})` } }],
       }), { headers: { "content-type": "application/json" }, status: 200 });