fix(sandbox): retain submitted build IDs on polling failure

Author: chenanran555

packages/commands/src/commands/sandbox/template.ts

@@ -355,12 +355,41 @@ async function emitTemplateMutationResult(options: {
     ),
     options.endpoint,
   ).toString();
-  const build = await waitForTemplateBuild(
-    options.client,
-    options.settings,
-    buildEndpoint,
-    options.pollInterval ?? 5,
-  );
+  let build: TemplateBuildStatus;
+  try {
+    build = await waitForTemplateBuild(
+      options.client,
+      options.settings,
+      buildEndpoint,
+      options.pollInterval ?? 5,
+    );
+  } catch (error) {
+    const hint =
+      `Submitted build / 已提交的构建: templateID=${options.response.templateID}, buildID=${options.response.buildID}.\n` +
+      "Check this build with sandbox template build-status before submitting another build. / 请先通过 sandbox template build-status 查询本次构建，再决定是否重新提交。";
+    if (error instanceof BailianError) {
+      throw new BailianError(
+        error.message,
+        error.exitCode,
+        [error.hint, hint].filter(Boolean).join("\n"),
+        {
+          api: error.api,
+          rawResponse: error.rawResponse,
+          cause: error.cause,
+        },
+      );
+    }
+    // Leave transport errors intact so runtime retains timeout/network classification.
+    const recovery = {
+      templateID: options.response.templateID,
+      buildID: options.response.buildID,
+      hint,
+    };
+    process.stderr.write(
+      format === "json" ? `${JSON.stringify(recovery, null, 2)}\n\n` : `${hint}\n`,
+    );
+    throw error;
+  }
   if (options.settings.quiet) emitBare(displayValue(options.response.templateID));
   else emitResult({ template: options.response, build }, format);
 }

packages/commands/tests/e2e/sandbox-base-url.e2e.test.ts

@@ -231,3 +231,88 @@ describe("e2e: Sandbox custom gateway transport", () => {
     },
   );
 });
+
+describe("e2e: Sandbox submitted build recovery", () => {
+  test.each([
+    { action: "create", output: "json", failure: "timeout", exitCode: 5 },
+    { action: "update", output: "text", failure: "service", exitCode: 1 },
+    { action: "create", output: "json", failure: "network", exitCode: 6 },
+  ])(
+    "$action retains IDs after a $failure in $output output",
+    async ({ action, output, failure, exitCode }) => {
+      let submissionCount = 0;
+      const server = createServer((request, response) => {
+        request.resume();
+        response.setHeader("content-type", "application/json");
+        if (request.method !== "GET") {
+          submissionCount += 1;
+          response.end(
+            JSON.stringify({
+              templateID: "template-recovery",
+              buildID: "build-recovery",
+              buildStatus: "building",
+            }),
+          );
+        } else if (failure === "network") {
+          request.socket.destroy();
+        } else if (failure === "service") {
+          response.writeHead(503);
+          response.end(
+            JSON.stringify({
+              code: 100005,
+              message: "original service failure",
+              requestID: "request-recovery",
+            }),
+          );
+        } else {
+          response.end(JSON.stringify({ status: "building" }));
+        }
+      });
+      await new Promise<void>((resolve, reject) => {
+        server.once("error", reject);
+        server.listen(0, "127.0.0.1", resolve);
+      });
+      servers.push(server);
+      const address = server.address();
+      if (!address || typeof address === "string") throw new Error("Expected a local TCP server.");
+      const args =
+        action === "create"
+          ? ["--name", "recovery", "--cpu-count", "1", "--memory-mb", "2048"]
+          : ["--template-id", "template-recovery", "--description", "updated"];
+      const result = await runCommandE2e(
+        ROUTES,
+        ["sandbox", "template", action, ...args, "--timeout", "1", "--quiet", "--output", output],
+        makeConfigEnv({
+          api_key: "sk-recovery-test",
+          base_url: `http://127.0.0.1:${address.port}`,
+        }),
+      );
+      expect(result.exitCode, result.stderr).toBe(exitCode);
+      expect(result.stdout).toBe("");
+      expect(submissionCount).toBe(1);
+      expect(result.stderr).toContain("templateID=template-recovery, buildID=build-recovery");
+      expect(result.stderr).toContain("sandbox template build-status");
+      if (output === "json") {
+        const diagnostics = result.stderr
+          .trim()
+          .split(/\n\s*\n/)
+          .map((diagnostic) => JSON.parse(diagnostic));
+        expect(diagnostics.at(-1)).toMatchObject({ error: { code: exitCode } });
+        if (failure === "timeout") {
+          expect(diagnostics).toHaveLength(1);
+          expect(diagnostics[0].error.message).toBe("Template build polling timed out.");
+        } else {
+          expect(diagnostics[0]).toMatchObject({
+            templateID: "template-recovery",
+            buildID: "build-recovery",
+          });
+          expect(diagnostics.at(-1).error.message).toContain("Network request failed");
+        }
+      } else {
+        expect(result.stderr).toContain("original service failure");
+        expect(result.stderr).toContain("HTTP 503 (100005)");
+        expect(result.stderr).toContain("request-recovery");
+      }
+    },
+  );
+});

packages/commands/tests/sandbox.test.ts

@@ -445,6 +445,113 @@ describe("Sandbox template build polling", () => {
     expect(requestJson).not.toHaveBeenCalled();
   });
 
+  test.each([
+    { name: "create", command: sandboxTemplateCreate },
+    { name: "update", command: sandboxTemplateUpdate },
+  ])("$name preserves submitted IDs and the original polling failure", async ({ command }) => {
+    const submission = { templateID: "template-test", buildID: "build-test" };
+    const serviceCause = new Error("original cause");
+    const serviceError = new BailianError("service message", ExitCode.GENERAL, "original hint", {
+      api: { httpStatus: 503, apiCode: "Unavailable", requestId: "request-test" },
+      rawResponse: "original response",
+      cause: serviceCause,
+    });
+    const scenarios = [
+      {
+        timeout: 0,
+        response: { status: "building" },
+        exitCode: ExitCode.TIMEOUT,
+        message: "Template build polling timed out.",
+      },
+      {
+        timeout: 30,
+        response: { status: "error", reason: { message: "image download failed" } },
+        exitCode: ExitCode.GENERAL,
+        message: "image download failed",
+      },
+      {
+        timeout: 30,
+        error: serviceError,
+        exitCode: ExitCode.GENERAL,
+        message: serviceError.message,
+      },
+    ];
+    const stdout = vi.spyOn(process.stdout, "write").mockImplementation(() => true);
+    for (const scenario of scenarios) {
+      const requestJson = vi.fn().mockResolvedValueOnce(submission);
+      if (scenario.error) requestJson.mockRejectedValue(scenario.error);
+      else requestJson.mockResolvedValue(scenario.response);
+      const operation = command.run({
+        identity: { binName: "bl" },
+        settings: { ...SETTINGS, timeout: scenario.timeout },
+        flags: {
+          workspaceId: "ws-test",
+          templateId: "template-test",
+          name: "python",
+          cpuCount: 1,
+          memoryMb: 2048,
+          async: false,
+        },
+        client: { requestJson, url: createUrlResolver() },
+      } as never);
+      await expect(operation).rejects.toMatchObject({
+        message: scenario.message,
+        exitCode: scenario.exitCode,
+        hint: expect.stringContaining("templateID=template-test, buildID=build-test"),
+      });
+      if (scenario.error) {
+        await expect(operation).rejects.toMatchObject({
+          api: serviceError.api,
+          rawResponse: serviceError.rawResponse,
+          cause: serviceCause,
+          hint: expect.stringContaining("original hint"),
+        });
+      }
+      expect(requestJson.mock.calls.filter(([request]) => request.method !== "GET")).toHaveLength(
+        1,
+      );
+    }
+    expect(stdout).not.toHaveBeenCalled();
+  });
+
+  test.each(["json", "text"] as const)(
+    "transport failures retain their identity and emit build recovery in %s diagnostics",
+    async (output) => {
+      const failure = new TypeError("fetch failed", { cause: { code: "ECONNRESET" } });
+      const requestJson = vi
+        .fn()
+        .mockResolvedValueOnce({ templateID: "template-test", buildID: "build-test" })
+        .mockRejectedValue(failure);
+      let stderr = "";
+      vi.spyOn(process.stderr, "write").mockImplementation((chunk) => {
+        stderr += String(chunk);
+        return true;
+      });
+      await expect(
+        sandboxTemplateCreate.run({
+          identity: { binName: "bl" },
+          settings: { ...SETTINGS, output },
+          flags: {
+            workspaceId: "ws-test",
+            name: "python",
+            cpuCount: 1,
+            memoryMb: 2048,
+            async: false,
+          },
+          client: { requestJson, url: createUrlResolver() },
+        } as never),
+      ).rejects.toBe(failure);
+      if (output === "json") {
+        expect(JSON.parse(stderr)).toMatchObject({
+          templateID: "template-test",
+          buildID: "build-test",
+        });
+      } else {
+        expect(stderr).toContain("templateID=template-test, buildID=build-test");
+      }
+    },
+  );
+
   test("async template creation returns after the submit request", async () => {
     let stdout = "";
     vi.spyOn(process.stdout, "write").mockImplementation((chunk) => {