Files
openclaw/src/auto-reply/reply/reply-dispatcher.ts
Bridgerz ab4a08a82a fix: defer gateway restart until all replies are sent (#12970)
* fix: defer gateway restart until all replies are sent

Fixes a race condition where gateway config changes (e.g., enabling
plugins via iMessage) trigger an immediate SIGUSR1 restart, killing the
iMessage RPC connection before replies are delivered.

Both restart paths (config watcher and RPC-triggered) now defer until
all queued operations, pending replies, and embedded agent runs complete
(polling every 500ms, 30s timeout). A shared emitGatewayRestart() guard
prevents double SIGUSR1 when both paths fire simultaneously.

Key changes:
- Dispatcher registry tracks active reply dispatchers globally
- markComplete() called in finally block for guaranteed cleanup
- Pre-restart deferral hook registered at gateway startup
- Centralized extractDeliveryInfo() for session key parsing
- Post-restart sentinel messages delivered directly (not via agent)
- config-patch distinguished from config-apply in sentinel kind

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: single-source gateway restart authorization

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 00:29:29 +01:00

237 lines
8.1 KiB
TypeScript

import type { HumanDelayConfig } from "../../config/types.js";
import type { GetReplyOptions, ReplyPayload } from "../types.js";
import type { ResponsePrefixContext } from "./response-prefix-template.js";
import type { TypingController } from "./typing.js";
import { sleep } from "../../utils.js";
import { registerDispatcher } from "./dispatcher-registry.js";
import { normalizeReplyPayload, type NormalizeReplySkipReason } from "./normalize-reply.js";
export type ReplyDispatchKind = "tool" | "block" | "final";
type ReplyDispatchErrorHandler = (err: unknown, info: { kind: ReplyDispatchKind }) => void;
type ReplyDispatchSkipHandler = (
payload: ReplyPayload,
info: { kind: ReplyDispatchKind; reason: NormalizeReplySkipReason },
) => void;
type ReplyDispatchDeliverer = (
payload: ReplyPayload,
info: { kind: ReplyDispatchKind },
) => Promise<void>;
const DEFAULT_HUMAN_DELAY_MIN_MS = 800;
const DEFAULT_HUMAN_DELAY_MAX_MS = 2500;
/** Generate a random delay within the configured range. */
function getHumanDelay(config: HumanDelayConfig | undefined): number {
const mode = config?.mode ?? "off";
if (mode === "off") {
return 0;
}
const min =
mode === "custom" ? (config?.minMs ?? DEFAULT_HUMAN_DELAY_MIN_MS) : DEFAULT_HUMAN_DELAY_MIN_MS;
const max =
mode === "custom" ? (config?.maxMs ?? DEFAULT_HUMAN_DELAY_MAX_MS) : DEFAULT_HUMAN_DELAY_MAX_MS;
if (max <= min) {
return min;
}
return Math.floor(Math.random() * (max - min + 1)) + min;
}
export type ReplyDispatcherOptions = {
deliver: ReplyDispatchDeliverer;
responsePrefix?: string;
/** Static context for response prefix template interpolation. */
responsePrefixContext?: ResponsePrefixContext;
/** Dynamic context provider for response prefix template interpolation.
* Called at normalization time, after model selection is complete. */
responsePrefixContextProvider?: () => ResponsePrefixContext;
onHeartbeatStrip?: () => void;
onIdle?: () => void;
onError?: ReplyDispatchErrorHandler;
// AIDEV-NOTE: onSkip lets channels detect silent/empty drops (e.g. Telegram empty-response fallback).
onSkip?: ReplyDispatchSkipHandler;
/** Human-like delay between block replies for natural rhythm. */
humanDelay?: HumanDelayConfig;
};
export type ReplyDispatcherWithTypingOptions = Omit<ReplyDispatcherOptions, "onIdle"> & {
onReplyStart?: () => Promise<void> | void;
onIdle?: () => void;
/** Called when the typing controller is cleaned up (e.g., on NO_REPLY). */
onCleanup?: () => void;
};
type ReplyDispatcherWithTypingResult = {
dispatcher: ReplyDispatcher;
replyOptions: Pick<GetReplyOptions, "onReplyStart" | "onTypingController" | "onTypingCleanup">;
markDispatchIdle: () => void;
};
export type ReplyDispatcher = {
sendToolResult: (payload: ReplyPayload) => boolean;
sendBlockReply: (payload: ReplyPayload) => boolean;
sendFinalReply: (payload: ReplyPayload) => boolean;
waitForIdle: () => Promise<void>;
getQueuedCounts: () => Record<ReplyDispatchKind, number>;
markComplete: () => void;
};
type NormalizeReplyPayloadInternalOptions = Pick<
ReplyDispatcherOptions,
"responsePrefix" | "responsePrefixContext" | "responsePrefixContextProvider" | "onHeartbeatStrip"
> & {
onSkip?: (reason: NormalizeReplySkipReason) => void;
};
function normalizeReplyPayloadInternal(
payload: ReplyPayload,
opts: NormalizeReplyPayloadInternalOptions,
): ReplyPayload | null {
// Prefer dynamic context provider over static context
const prefixContext = opts.responsePrefixContextProvider?.() ?? opts.responsePrefixContext;
return normalizeReplyPayload(payload, {
responsePrefix: opts.responsePrefix,
responsePrefixContext: prefixContext,
onHeartbeatStrip: opts.onHeartbeatStrip,
onSkip: opts.onSkip,
});
}
export function createReplyDispatcher(options: ReplyDispatcherOptions): ReplyDispatcher {
let sendChain: Promise<void> = Promise.resolve();
// Track in-flight deliveries so we can emit a reliable "idle" signal.
// Start with pending=1 as a "reservation" to prevent premature gateway restart.
// This is decremented when markComplete() is called to signal no more replies will come.
let pending = 1;
let completeCalled = false;
// Track whether we've sent a block reply (for human delay - skip delay on first block).
let sentFirstBlock = false;
// Serialize outbound replies to preserve tool/block/final order.
const queuedCounts: Record<ReplyDispatchKind, number> = {
tool: 0,
block: 0,
final: 0,
};
// Register this dispatcher globally for gateway restart coordination.
const { unregister } = registerDispatcher({
pending: () => pending,
waitForIdle: () => sendChain,
});
const enqueue = (kind: ReplyDispatchKind, payload: ReplyPayload) => {
const normalized = normalizeReplyPayloadInternal(payload, {
responsePrefix: options.responsePrefix,
responsePrefixContext: options.responsePrefixContext,
responsePrefixContextProvider: options.responsePrefixContextProvider,
onHeartbeatStrip: options.onHeartbeatStrip,
onSkip: (reason) => options.onSkip?.(payload, { kind, reason }),
});
if (!normalized) {
return false;
}
queuedCounts[kind] += 1;
pending += 1;
// Determine if we should add human-like delay (only for block replies after the first).
const shouldDelay = kind === "block" && sentFirstBlock;
if (kind === "block") {
sentFirstBlock = true;
}
sendChain = sendChain
.then(async () => {
// Add human-like delay between block replies for natural rhythm.
if (shouldDelay) {
const delayMs = getHumanDelay(options.humanDelay);
if (delayMs > 0) {
await sleep(delayMs);
}
}
// Safe: deliver is called inside an async .then() callback, so even a synchronous
// throw becomes a rejection that flows through .catch()/.finally(), ensuring cleanup.
await options.deliver(normalized, { kind });
})
.catch((err) => {
options.onError?.(err, { kind });
})
.finally(() => {
pending -= 1;
// Clear reservation if:
// 1. pending is now 1 (just the reservation left)
// 2. markComplete has been called
// 3. No more replies will be enqueued
if (pending === 1 && completeCalled) {
pending -= 1; // Clear the reservation
}
if (pending === 0) {
// Unregister from global tracking when idle.
unregister();
options.onIdle?.();
}
});
return true;
};
const markComplete = () => {
if (completeCalled) {
return;
}
completeCalled = true;
// If no replies were enqueued (pending is still 1 = just the reservation),
// schedule clearing the reservation after current microtasks complete.
// This gives any in-flight enqueue() calls a chance to increment pending.
void Promise.resolve().then(() => {
if (pending === 1 && completeCalled) {
// Still just the reservation, no replies were enqueued
pending -= 1;
if (pending === 0) {
unregister();
options.onIdle?.();
}
}
});
};
return {
sendToolResult: (payload) => enqueue("tool", payload),
sendBlockReply: (payload) => enqueue("block", payload),
sendFinalReply: (payload) => enqueue("final", payload),
waitForIdle: () => sendChain,
getQueuedCounts: () => ({ ...queuedCounts }),
markComplete,
};
}
export function createReplyDispatcherWithTyping(
options: ReplyDispatcherWithTypingOptions,
): ReplyDispatcherWithTypingResult {
const { onReplyStart, onIdle, onCleanup, ...dispatcherOptions } = options;
let typingController: TypingController | undefined;
const dispatcher = createReplyDispatcher({
...dispatcherOptions,
onIdle: () => {
typingController?.markDispatchIdle();
onIdle?.();
},
});
return {
dispatcher,
replyOptions: {
onReplyStart,
onTypingCleanup: onCleanup,
onTypingController: (typing) => {
typingController = typing;
},
},
markDispatchIdle: () => {
typingController?.markDispatchIdle();
onIdle?.();
},
};
}