fix(gateway): gracefully handle AbortError and transient network errors

Addresses issues #1851, #1997, and #2034.

During config reload (SIGUSR1), in-flight requests are aborted, causing
AbortError exceptions. Similarly, transient network errors (fetch failed,
ECONNRESET, ETIMEDOUT, etc.) can crash the gateway unnecessarily.

This change:
- Adds isAbortError() to detect intentional cancellations
- Adds isTransientNetworkError() to detect temporary connectivity issues
- Logs these errors appropriately instead of crashing
- Handles nested cause chains and AggregateError

AbortError is logged as a warning (expected during shutdown).
Network errors are logged as non-fatal errors (will resolve on their own).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

src/infra/unhandled-rejections.test.ts

@@ -0,0 +1,129 @@
+import { describe, expect, it } from "vitest";
+
+import { isAbortError, isTransientNetworkError } from "./unhandled-rejections.js";
+
+describe("isAbortError", () => {
+  it("returns true for error with name AbortError", () => {
+    const error = new Error("aborted");
+    error.name = "AbortError";
+    expect(isAbortError(error)).toBe(true);
+  });
+
+  it('returns true for error with "This operation was aborted" message', () => {
+    const error = new Error("This operation was aborted");
+    expect(isAbortError(error)).toBe(true);
+  });
+
+  it("returns true for undici-style AbortError", () => {
+    // Node's undici throws errors with this exact message
+    const error = Object.assign(new Error("This operation was aborted"), { name: "AbortError" });
+    expect(isAbortError(error)).toBe(true);
+  });
+
+  it("returns true for object with AbortError name", () => {
+    expect(isAbortError({ name: "AbortError", message: "test" })).toBe(true);
+  });
+
+  it("returns false for regular errors", () => {
+    expect(isAbortError(new Error("Something went wrong"))).toBe(false);
+    expect(isAbortError(new TypeError("Cannot read property"))).toBe(false);
+    expect(isAbortError(new RangeError("Invalid array length"))).toBe(false);
+  });
+
+  it("returns false for errors with similar but different messages", () => {
+    expect(isAbortError(new Error("Operation aborted"))).toBe(false);
+    expect(isAbortError(new Error("aborted"))).toBe(false);
+    expect(isAbortError(new Error("Request was aborted"))).toBe(false);
+  });
+
+  it("returns false for null and undefined", () => {
+    expect(isAbortError(null)).toBe(false);
+    expect(isAbortError(undefined)).toBe(false);
+  });
+
+  it("returns false for non-error values", () => {
+    expect(isAbortError("string error")).toBe(false);
+    expect(isAbortError(42)).toBe(false);
+  });
+
+  it("returns false for plain objects without AbortError name", () => {
+    expect(isAbortError({ message: "plain object" })).toBe(false);
+  });
+});
+
+describe("isTransientNetworkError", () => {
+  it("returns true for errors with transient network codes", () => {
+    const codes = [
+      "ECONNRESET",
+      "ECONNREFUSED",
+      "ENOTFOUND",
+      "ETIMEDOUT",
+      "ESOCKETTIMEDOUT",
+      "ECONNABORTED",
+      "EPIPE",
+      "EHOSTUNREACH",
+      "ENETUNREACH",
+      "EAI_AGAIN",
+      "UND_ERR_CONNECT_TIMEOUT",
+      "UND_ERR_SOCKET",
+      "UND_ERR_HEADERS_TIMEOUT",
+      "UND_ERR_BODY_TIMEOUT",
+    ];
+
+    for (const code of codes) {
+      const error = Object.assign(new Error("test"), { code });
+      expect(isTransientNetworkError(error), `code: ${code}`).toBe(true);
+    }
+  });
+
+  it('returns true for TypeError with "fetch failed" message', () => {
+    const error = new TypeError("fetch failed");
+    expect(isTransientNetworkError(error)).toBe(true);
+  });
+
+  it("returns true for fetch failed with network cause", () => {
+    const cause = Object.assign(new Error("getaddrinfo ENOTFOUND"), { code: "ENOTFOUND" });
+    const error = Object.assign(new TypeError("fetch failed"), { cause });
+    expect(isTransientNetworkError(error)).toBe(true);
+  });
+
+  it("returns true for nested cause chain with network error", () => {
+    const innerCause = Object.assign(new Error("connection reset"), { code: "ECONNRESET" });
+    const outerCause = Object.assign(new Error("wrapper"), { cause: innerCause });
+    const error = Object.assign(new TypeError("fetch failed"), { cause: outerCause });
+    expect(isTransientNetworkError(error)).toBe(true);
+  });
+
+  it("returns true for AggregateError containing network errors", () => {
+    const networkError = Object.assign(new Error("timeout"), { code: "ETIMEDOUT" });
+    const error = new AggregateError([networkError], "Multiple errors");
+    expect(isTransientNetworkError(error)).toBe(true);
+  });
+
+  it("returns false for regular errors without network codes", () => {
+    expect(isTransientNetworkError(new Error("Something went wrong"))).toBe(false);
+    expect(isTransientNetworkError(new TypeError("Cannot read property"))).toBe(false);
+    expect(isTransientNetworkError(new RangeError("Invalid array length"))).toBe(false);
+  });
+
+  it("returns false for errors with non-network codes", () => {
+    const error = Object.assign(new Error("test"), { code: "INVALID_CONFIG" });
+    expect(isTransientNetworkError(error)).toBe(false);
+  });
+
+  it("returns false for null and undefined", () => {
+    expect(isTransientNetworkError(null)).toBe(false);
+    expect(isTransientNetworkError(undefined)).toBe(false);
+  });
+
+  it("returns false for non-error values", () => {
+    expect(isTransientNetworkError("string error")).toBe(false);
+    expect(isTransientNetworkError(42)).toBe(false);
+    expect(isTransientNetworkError({ message: "plain object" })).toBe(false);
+  });
+
+  it("returns false for AggregateError with only non-network errors", () => {
+    const error = new AggregateError([new Error("regular error")], "Multiple errors");
+    expect(isTransientNetworkError(error)).toBe(false);
+  });
+});

src/infra/unhandled-rejections.ts

@@ -1,11 +1,88 @@
 import process from "node:process";
 
-import { formatErrorMessage, formatUncaughtError } from "./errors.js";
+import { formatUncaughtError } from "./errors.js";
 
 type UnhandledRejectionHandler = (reason: unknown) => boolean;
 
 const handlers = new Set<UnhandledRejectionHandler>();
 
+/**
+ * Checks if an error is an AbortError.
+ * These are typically intentional cancellations (e.g., during shutdown) and shouldn't crash.
+ */
+export function isAbortError(err: unknown): boolean {
+  if (!err || typeof err !== "object") return false;
+  const name = "name" in err ? String(err.name) : "";
+  if (name === "AbortError") return true;
+  // Check for "This operation was aborted" message from Node's undici
+  const message = "message" in err && typeof err.message === "string" ? err.message : "";
+  if (message === "This operation was aborted") return true;
+  return false;
+}
+
+// Network error codes that indicate transient failures (shouldn't crash the gateway)
+const TRANSIENT_NETWORK_CODES = new Set([
+  "ECONNRESET",
+  "ECONNREFUSED",
+  "ENOTFOUND",
+  "ETIMEDOUT",
+  "ESOCKETTIMEDOUT",
+  "ECONNABORTED",
+  "EPIPE",
+  "EHOSTUNREACH",
+  "ENETUNREACH",
+  "EAI_AGAIN",
+  "UND_ERR_CONNECT_TIMEOUT",
+  "UND_ERR_SOCKET",
+  "UND_ERR_HEADERS_TIMEOUT",
+  "UND_ERR_BODY_TIMEOUT",
+]);
+
+function getErrorCode(err: unknown): string | undefined {
+  if (!err || typeof err !== "object") return undefined;
+  const code = (err as { code?: unknown }).code;
+  return typeof code === "string" ? code : undefined;
+}
+
+function getErrorCause(err: unknown): unknown {
+  if (!err || typeof err !== "object") return undefined;
+  return (err as { cause?: unknown }).cause;
+}
+
+/**
+ * Checks if an error is a transient network error that shouldn't crash the gateway.
+ * These are typically temporary connectivity issues that will resolve on their own.
+ */
+export function isTransientNetworkError(err: unknown): boolean {
+  if (!err) return false;
+
+  // Check the error itself
+  const code = getErrorCode(err);
+  if (code && TRANSIENT_NETWORK_CODES.has(code)) return true;
+
+  // "fetch failed" TypeError from undici (Node's native fetch)
+  if (err instanceof TypeError && err.message === "fetch failed") {
+    const cause = getErrorCause(err);
+    // The cause often contains the actual network error
+    if (cause) return isTransientNetworkError(cause);
+    // Even without a cause, "fetch failed" is typically a network issue
+    return true;
+  }
+
+  // Check the cause chain recursively
+  const cause = getErrorCause(err);
+  if (cause && cause !== err) {
+    return isTransientNetworkError(cause);
+  }
+
+  // AggregateError may wrap multiple causes
+  if (err instanceof AggregateError && err.errors?.length) {
+    return err.errors.some((e) => isTransientNetworkError(e));
+  }
+
+  return false;
+}
+
 export function registerUnhandledRejectionHandler(handler: UnhandledRejectionHandler): () => void {
   handlers.add(handler);
   return () => {
@@ -13,36 +90,6 @@ export function registerUnhandledRejectionHandler(handler: UnhandledRejectionHan
   };
 }
 
-/**
- * Check if an error is a recoverable/transient error that shouldn't crash the process.
- * These include network errors and abort signals during shutdown.
- */
-function isRecoverableError(reason: unknown): boolean {
-  if (!reason) return false;
-
-  // Check error name for AbortError
-  if (reason instanceof Error && reason.name === "AbortError") {
-    return true;
-  }
-
-  const message = reason instanceof Error ? reason.message : formatErrorMessage(reason);
-  const lowerMessage = message.toLowerCase();
-  return (
-    lowerMessage.includes("fetch failed") ||
-    lowerMessage.includes("network request") ||
-    lowerMessage.includes("econnrefused") ||
-    lowerMessage.includes("econnreset") ||
-    lowerMessage.includes("etimedout") ||
-    lowerMessage.includes("socket hang up") ||
-    lowerMessage.includes("enotfound") ||
-    lowerMessage.includes("network error") ||
-    lowerMessage.includes("getaddrinfo") ||
-    lowerMessage.includes("client network socket disconnected") ||
-    lowerMessage.includes("this operation was aborted") ||
-    lowerMessage.includes("aborted")
-  );
-}
-
 export function isUnhandledRejectionHandled(reason: unknown): boolean {
   for (const handler of handlers) {
     try {
@@ -61,9 +108,17 @@ export function installUnhandledRejectionHandler(): void {
   process.on("unhandledRejection", (reason, _promise) => {
     if (isUnhandledRejectionHandled(reason)) return;
 
-    // Don't crash on recoverable/transient errors - log them and continue
+    // AbortError is typically an intentional cancellation (e.g., during shutdown)
-    if (isRecoverableError(reason)) {
+    // Log it but don't crash - these are expected during graceful shutdown
-      console.error("[clawdbot] Recoverable error (not crashing):", formatUncaughtError(reason));
+    if (isAbortError(reason)) {
+      console.warn("[clawdbot] Suppressed AbortError:", formatUncaughtError(reason));
+      return;
+    }
+
+    // Transient network errors (fetch failed, connection reset, etc.) shouldn't crash
+    // These are temporary connectivity issues that will resolve on their own
+    if (isTransientNetworkError(reason)) {
+      console.error("[clawdbot] Network error (non-fatal):", formatUncaughtError(reason));
       return;
     }