fix(agents): classify Zhipu GLM overload as overloaded for failover (#93241)

0xghost42 · steipete · web-flow · commit 6c1041339da6 · 2026-06-20T23:49:39.000-04:00
Merged via squash. Prepared head SHA: db79e94 Co-authored-by: 0xghost42 <151941421+0xghost42@users.noreply.github.com> Co-authored-by: steipete <58493+steipete@users.noreply.github.com> Reviewed-by: @steipete
diff --git a/src/agents/embedded-agent-helpers/failover-matches.test.ts b/src/agents/embedded-agent-helpers/failover-matches.test.ts
@@ -103,6 +103,19 @@ describe("Z.ai vendor error codes (#48988)", () => {
   });
 });
 
+describe("Chinese provider overload messages", () => {
+  const ZHIPU_OVERLOAD = "[1305][该模型当前访问量过大，请您稍后再试]";
+
+  it("classifies the Zhipu GLM overload body as overloaded", () => {
+    expect(isOverloadedErrorMessage(ZHIPU_OVERLOAD)).toBe(true);
+  });
+
+  it("does not misclassify the GLM overload body as rate limit or auth", () => {
+    expect(isRateLimitErrorMessage(ZHIPU_OVERLOAD)).toBe(false);
+    expect(isAuthErrorMessage(ZHIPU_OVERLOAD)).toBe(false);
+  });
+});
+
 describe("Volcengine Coding Plan subscription errors", () => {
   it("classifies InvalidSubscription JSON body as billing", () => {
     const raw =
diff --git a/src/agents/embedded-agent-helpers/failover-matches.ts b/src/agents/embedded-agent-helpers/failover-matches.ts
@@ -107,6 +107,7 @@ const ERROR_PATTERNS = {
     // Chinese provider overloaded messages
     "服务过载",
     "当前负载过高",
+    "访问量过大",
   ],
   serverError: [
     "an error occurred while processing",
diff --git a/src/agents/model-fallback.test.ts b/src/agents/model-fallback.test.ts
@@ -746,6 +746,24 @@ describe("runWithModelFallback", () => {
     expect(result.attempts[0].reason).toBe("unknown");
   });
 
+  it("falls back on a Zhipu GLM 1305 overload body and classifies it as overloaded", async () => {
+    const cfg = makeCfg();
+    const glmOverload = new Error("[1305][该模型当前访问量过大，请您稍后再试]");
+    const run = vi.fn().mockRejectedValueOnce(glmOverload).mockResolvedValueOnce("ok");
+
+    const result = await runWithModelFallback({
+      cfg,
+      provider: "glm",
+      model: "GLM-5.2",
+      run,
+    });
+    expect(result.result).toBe("ok");
+    expect(run).toHaveBeenCalledTimes(2);
+    expect(requireMockCall(run, 1, "fallback run")).toEqual(["anthropic", "claude-haiku-3-5"]);
+    expect(result.attempts).toHaveLength(1);
+    expect(result.attempts[0].reason).toBe("overloaded");
+  });
+
   it("does not prepare agent harness plugins for forced OpenClaw candidates", async () => {
     const cfg = makeCfg({
       models: {
