From bdbef04ac312e52fb46f3c40928120b53402e647 Mon Sep 17 00:00:00 2001
From: Trevin Chow <trevin@bigcartel.com>
Date: Thu, 29 Jan 2026 12:23:02 -0800
Subject: [PATCH] fix(gateway): expand transient network error detection
 patterns

Add more DNS error codes (EAI_NODATA, EAI_NONAME), TLS certificate
errors, HTTP status codes (429, 502, 503, 504), and message-based
fallbacks to prevent gateway crash loops on transient network issues.
---
 src/infra/unhandled-rejections.test.ts | 60 ++++++++++++++++++++++++++
 src/infra/unhandled-rejections.ts      | 32 ++++++++++++++
 2 files changed, 92 insertions(+)

diff --git a/src/infra/unhandled-rejections.test.ts b/src/infra/unhandled-rejections.test.ts
index 1ec144ba1..a4f637c5a 100644
--- a/src/infra/unhandled-rejections.test.ts
+++ b/src/infra/unhandled-rejections.test.ts
@@ -127,3 +127,63 @@ describe("isTransientNetworkError", () => {
     expect(isTransientNetworkError(error)).toBe(false);
   });
 });
+
+describe("isTransientNetworkError - expanded patterns", () => {
+  it("recognizes EAI_NODATA as transient", () => {
+    const err = { code: "EAI_NODATA", message: "DNS lookup failed" };
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+
+  it("recognizes EAI_NONAME as transient", () => {
+    const err = { code: "EAI_NONAME", message: "DNS name not found" };
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+
+  it("recognizes HTTP 502 as transient", () => {
+    const err = { status: 502, message: "Bad Gateway" };
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+
+  it("recognizes HTTP 503 as transient", () => {
+    const err = { statusCode: 503, message: "Service Unavailable" };
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+
+  it("recognizes HTTP 429 rate limit as transient", () => {
+    const err = { status: 429, message: "Too Many Requests" };
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+
+  it("recognizes socket closed message as transient", () => {
+    const err = new Error("socket closed unexpectedly");
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+
+  it("recognizes client network socket disconnected as transient", () => {
+    const err = new Error("Client network socket disconnected before secure TLS connection");
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+
+  it("recognizes TLS certificate errors as transient", () => {
+    const err1 = { code: "CERT_HAS_EXPIRED", message: "certificate has expired" };
+    expect(isTransientNetworkError(err1)).toBe(true);
+
+    const err2 = { code: "ERR_TLS_CERT_ALTNAME_INVALID", message: "Hostname mismatch" };
+    expect(isTransientNetworkError(err2)).toBe(true);
+  });
+
+  it("does not recognize HTTP 400 as transient", () => {
+    const err = { status: 400, message: "Bad Request" };
+    expect(isTransientNetworkError(err)).toBe(false);
+  });
+
+  it("does not recognize HTTP 401 as transient", () => {
+    const err = { status: 401, message: "Unauthorized" };
+    expect(isTransientNetworkError(err)).toBe(false);
+  });
+
+  it("handles string status codes", () => {
+    const err = { status: "503", message: "Service Unavailable" };
+    expect(isTransientNetworkError(err)).toBe(true);
+  });
+});
diff --git a/src/infra/unhandled-rejections.ts b/src/infra/unhandled-rejections.ts
index d186c6a78..3e3041157 100644
--- a/src/infra/unhandled-rejections.ts
+++ b/src/infra/unhandled-rejections.ts
@@ -28,6 +28,10 @@ const TRANSIENT_NETWORK_CODES = new Set([
   "EHOSTUNREACH",
   "ENETUNREACH",
   "EAI_AGAIN",
+  "EAI_NODATA",
+  "EAI_NONAME",
+  "CERT_HAS_EXPIRED",
+  "ERR_TLS_CERT_ALTNAME_INVALID",
   "UND_ERR_CONNECT_TIMEOUT",
   "UND_ERR_DNS_RESOLVE_FAILED",
   "UND_ERR_CONNECT",
@@ -99,6 +103,34 @@ export function isTransientNetworkError(err: unknown): boolean {
     return err.errors.some((e) => isTransientNetworkError(e));
   }
 
+  // Message-based fallback detection
+  const message = err instanceof Error ? err.message?.toLowerCase() : "";
+  if (
+    message.includes("fetch failed") ||
+    message.includes("network error") ||
+    message.includes("socket hang up") ||
+    message.includes("socket closed") ||
+    message.includes("client network socket disconnected")
+  ) {
+    return true;
+  }
+
+  // Check for transient HTTP error responses (handle both number and string status codes)
+  const statusRaw =
+    (err as { status?: unknown }).status ?? (err as { statusCode?: unknown }).statusCode;
+  const status =
+    typeof statusRaw === "number"
+      ? statusRaw
+      : typeof statusRaw === "string"
+        ? parseInt(statusRaw, 10)
+        : NaN;
+  if (
+    Number.isFinite(status) &&
+    (status === 429 || status === 502 || status === 503 || status === 504)
+  ) {
+    return true;
+  }
+
   return false;
 }