fix(agents): fallback on upstream provider errors

markoub · markoub · commit 86c1cf9cfc68 · 2026-06-23T22:42:39.000+02:00
diff --git a/src/agents/embedded-agent-helpers/failover-matches.ts b/src/agents/embedded-agent-helpers/failover-matches.ts
@@ -119,6 +119,7 @@ const ERROR_PATTERNS = {
     "bad gateway",
     "gateway timeout",
     "upstream error",
+    "upstream_error",
     "upstream connect error",
     "connection reset",
     // Chinese provider server error messages
diff --git a/src/agents/embedded-agent-runner/result-fallback-classifier.test.ts b/src/agents/embedded-agent-runner/result-fallback-classifier.test.ts
@@ -282,15 +282,49 @@ describe("classifyEmbeddedAgentRunResultForModelFallback", () => {
     expect(result).toBeNull();
   });
 
-  it("does not retry non-business transport error payloads", () => {
+  it("classifies transient provider error payloads as fallback-worthy", () => {
+    const errorText = JSON.stringify({
+      error: {
+        message: "Upstream request failed",
+        type: "upstream_error",
+        param: "",
+        code: null,
+      },
+    });
+
+    const result = classifyEmbeddedAgentRunResultForModelFallback({
+      provider: "custom",
+      model: "llama-3.1",
+      result: {
+        payloads: [
+          {
+            isError: true,
+            text: errorText,
+          },
+        ],
+        meta: {
+          durationMs: 42,
+        },
+      },
+    });
+
+    expect(result).toEqual({
+      message: `custom/llama-3.1 ended with a provider error: ${errorText}`,
+      reason: "timeout",
+      code: "embedded_error_payload",
+      rawError: errorText,
+    });
+  });
+
+  it("does not retry schema-rejection error payloads", () => {
     const result = classifyEmbeddedAgentRunResultForModelFallback({
       provider: "custom",
       model: "llama-3.1",
       result: {
         payloads: [
           {
             isError: true,
-            text: "HTTP 500: internal server error",
+            text: "LLM request failed: provider rejected the request schema or tool payload.",
           },
         ],
         meta: {
diff --git a/src/agents/embedded-agent-runner/result-fallback-classifier.ts b/src/agents/embedded-agent-runner/result-fallback-classifier.ts
@@ -146,11 +146,16 @@ function classifyHarnessResult(params: {
   }
 }
 
-/** Maps provider error payloads to fallback-safe business reasons. */
-function classifyBusinessDenialErrorPayloadReason(
+type EmbeddedErrorPayloadFailoverReason = Extract<
+  FailoverReason,
+  "auth" | "auth_permanent" | "billing" | "overloaded" | "rate_limit" | "server_error" | "timeout"
+>;
+
+/** Maps provider error payloads to fallback-safe reasons. */
+function classifyProviderErrorPayloadReason(
   errorText: string,
   provider: string,
-): Extract<FailoverReason, "auth" | "auth_permanent" | "billing" | "rate_limit"> | null {
+): EmbeddedErrorPayloadFailoverReason | null {
   if (!errorText.trim()) {
     return null;
   }
@@ -159,7 +164,10 @@ function classifyBusinessDenialErrorPayloadReason(
     case "auth":
     case "auth_permanent":
     case "billing":
+    case "overloaded":
     case "rate_limit":
+    case "server_error":
+    case "timeout":
       return failoverReason;
     default:
       return null;
@@ -252,7 +260,7 @@ export function classifyEmbeddedAgentRunResultForModelFallback(params: {
     .filter((payload) => payload?.isError === true)
     .map((payload) => (typeof payload.text === "string" ? payload.text : ""))
     .join("\n");
-  const failoverReason = classifyBusinessDenialErrorPayloadReason(errorText, params.provider);
+  const failoverReason = classifyProviderErrorPayloadReason(errorText, params.provider);
   if (failoverReason) {
     return {
       message: `${params.provider}/${params.model} ended with a provider error: ${errorText}`,
