Add gateway.unhandledRejections config (warn|exit)

PR_UNHANDLED_REJECTIONS.md

@@ -0,0 +1,28 @@
+# Proposal: Configurable gateway unhandled promise rejection policy
+
+## Motivation
+Running Clawdbot Gateway in production-like environments can encounter transient network errors (undici `fetch failed`, DNS hiccups, Telegram API blips). Today, any *unhandled* promise rejection can terminate the gateway process (`process.exit(1)`), causing user-visible downtime and missed replies.
+
+Diego’s deployment uses systemd with `Restart=always`, but the restart still interrupts message handling.
+
+## Desired behavior
+Add a JSON config knob to control what happens on unhandled promise rejections:
+
+```json5
+{
+  gateway: {
+    unhandledRejections: "warn" // or "exit"
+  }
+}
+```
+
+- `exit` (default): keep current behavior for safety; exit 1 so supervisors restart.
+- `warn`: never exit the gateway for unhandled rejections; log as error/warn and continue.
+
+## Notes
+- Existing suppression for AbortError / transient network errors should remain regardless of mode.
+- In `warn` mode, non-network/unexpected unhandled rejections should still be logged loudly.
+
+## Acceptance criteria
+- With `gateway.unhandledRejections="warn"`, a forced Telegram DNS failure should result in a logged error but the gateway process should remain running.
+- With `exit` (default), behavior remains unchanged.

src/cli/run-main.ts

@@ -10,6 +10,7 @@ import { ensureMoltbotCliOnPath } from "../infra/path-env.js";
 import { assertSupportedRuntime } from "../infra/runtime-guard.js";
 import { formatUncaughtError } from "../infra/errors.js";
 import { installUnhandledRejectionHandler } from "../infra/unhandled-rejections.js";
+import { loadConfig } from "../config/config.js";
 import { enableConsoleCapture } from "../logging.js";
 import { getPrimaryCommand, hasHelpOrVersion } from "./argv.js";
 import { tryRouteCli } from "./route.js";
@@ -41,8 +42,10 @@ export async function runCli(argv: string[] = process.argv) {
   const program = buildProgram();
 
   // Global error handlers to prevent silent crashes from unhandled rejections/exceptions.
-  // These log the error and exit gracefully instead of crashing without trace.
-  installUnhandledRejectionHandler();
+  // Default behavior is "exit" (preserves historical behavior). Users can opt into
+  // "warn" via config: gateway.unhandledRejections.
+  const cfg = loadConfig();
+  installUnhandledRejectionHandler({ mode: cfg.gateway?.unhandledRejections ?? "exit" });
 
   process.on("uncaughtException", (error) => {
     console.error("[moltbot] Uncaught exception:", formatUncaughtError(error));

src/config/types.gateway.ts

@@ -205,9 +205,17 @@ export type GatewayNodesConfig = {
   denyCommands?: string[];
 };
 
+export type GatewayUnhandledRejectionsMode = "exit" | "warn";
+
 export type GatewayConfig = {
   /** Single multiplexed port for Gateway WS + HTTP (default: 18789). */
   port?: number;
+  /**
+   * What to do on unhandled promise rejections in the gateway process.
+   * - exit (default): log + exit(1)
+   * - warn: log but keep running
+   */
+  unhandledRejections?: GatewayUnhandledRejectionsMode;
   /**
    * Explicit gateway mode. When set to "remote", local gateway start is disabled.
    * When set to "local", the CLI may start the gateway locally.

src/config/zod-schema.ts

@@ -304,6 +304,7 @@ export const MoltbotSchema = z
       .object({
         port: z.number().int().positive().optional(),
         mode: z.union([z.literal("local"), z.literal("remote")]).optional(),
+        unhandledRejections: z.union([z.literal("exit"), z.literal("warn")]).optional(),
         bind: z
           .union([
             z.literal("auto"),

src/index.ts

@@ -30,6 +30,7 @@ import {
 import { assertSupportedRuntime } from "./infra/runtime-guard.js";
 import { formatUncaughtError } from "./infra/errors.js";
 import { installUnhandledRejectionHandler } from "./infra/unhandled-rejections.js";
+import { loadConfig } from "./config/config.js";
 import { enableConsoleCapture } from "./logging.js";
 import { runCommandWithTimeout, runExec } from "./process/exec.js";
 import { assertWebChannel, normalizeE164, toWhatsappJid } from "./utils.js";
@@ -79,8 +80,10 @@ const isMain = isMainModule({
 
 if (isMain) {
   // Global error handlers to prevent silent crashes from unhandled rejections/exceptions.
-  // These log the error and exit gracefully instead of crashing without trace.
-  installUnhandledRejectionHandler();
+  // Default behavior is "exit" (preserves historical behavior). Users can opt into
+  // "warn" via config: gateway.unhandledRejections.
+  const cfg = loadConfig();
+  installUnhandledRejectionHandler({ mode: cfg.gateway?.unhandledRejections ?? "exit" });
 
   process.on("uncaughtException", (error) => {
     console.error("[moltbot] Uncaught exception:", formatUncaughtError(error));

src/infra/unhandled-rejections.ts

@@ -104,7 +104,11 @@ export function isUnhandledRejectionHandled(reason: unknown): boolean {
   return false;
 }
 
-export function installUnhandledRejectionHandler(): void {
+export type UnhandledRejectionsMode = "exit" | "warn";
+
+export function installUnhandledRejectionHandler(opts: { mode?: UnhandledRejectionsMode } = {}): void {
+  const mode: UnhandledRejectionsMode = opts.mode ?? "exit";
+
   process.on("unhandledRejection", (reason, _promise) => {
     if (isUnhandledRejectionHandled(reason)) return;
 
@@ -123,6 +127,7 @@ export function installUnhandledRejectionHandler(): void {
     }
 
     console.error("[moltbot] Unhandled promise rejection:", formatUncaughtError(reason));
+    if (mode === "warn") return;
     process.exit(1);
   });
 }