f89e074dd1
Governed substrate for autonomous agents: scoped identity (passports), audited actions, MCP workspace. Infra IPs and secrets redacted for public release.
73 lines
3.1 KiB
JavaScript
73 lines
3.1 KiB
JavaScript
/**
|
|
* Live sync-agent activity registry (in-process).
|
|
*
|
|
* Sync agents (POST /chat, POST /run via agentChatController / agentExecutionController)
|
|
* set conversations.is_processing = true DIRECTLY and never create an `agent_jobs`
|
|
* row. The watchdog's orphan-reaper (services/agent-job/watchdog.js) therefore treats
|
|
* every live sync agent as a candidate orphan, and the only thing protecting it is the
|
|
* flat `updated_at < NOW() - 2 minutes` guard. A single long blocking call with no
|
|
* intermediate saveStepMessage (Copilot CLI, a slow Anthropic/OpenAI request, a long
|
|
* tool) lets `updated_at` go stale, so the watchdog prematurely clears the lock of a
|
|
* still-working agent and posts «… перестал отвечать».
|
|
*
|
|
* Fix: while a sync agent is executing in THIS process we register its conversation
|
|
* here. The watchdog skips reaping any conversation present in the registry — unless it
|
|
* has been active longer than SYNC_ACTIVITY_MAX_MS (a genuinely-hung agent), in which
|
|
* case it is reaped anyway so a chat can never wedge permanently.
|
|
*
|
|
* PM2 runs godcrm as `instances: 1, exec_mode: 'fork'` (ecosystem.config.cjs), so the
|
|
* watchdog timer and the sync controllers live in the SAME process and share this Map.
|
|
* On a process crash the Map is lost together with the agent, so the leaked DB lock is
|
|
* correctly reaped on the next watchdog tick after restart.
|
|
*/
|
|
|
|
// Upper bound on how long a sync conversation is protected from the watchdog.
|
|
// Matches the 30-minute request socket timeout set in the sync controllers
|
|
// (BUG-504: req.socket.setTimeout(1800 * 1000)). Beyond this the agent is
|
|
// considered hung and the watchdog is allowed to reap it.
|
|
export const SYNC_ACTIVITY_MAX_MS = 30 * 60 * 1000;
|
|
|
|
/** Map<conversationId:number, startedAtMs:number> */
|
|
const _activeSync = new Map();
|
|
|
|
/**
|
|
* Mark a conversation as actively processed by a sync agent in this process.
|
|
* Idempotent: re-marking refreshes the start timestamp.
|
|
* @param {number|string} conversationId
|
|
*/
|
|
export function markSyncActive(conversationId) {
|
|
if (conversationId == null) return;
|
|
_activeSync.set(Number(conversationId), Date.now());
|
|
}
|
|
|
|
/**
|
|
* Clear the live mark for a conversation (call in finally on every exit path).
|
|
* @param {number|string} conversationId
|
|
*/
|
|
export function markSyncInactive(conversationId) {
|
|
if (conversationId == null) return;
|
|
_activeSync.delete(Number(conversationId));
|
|
}
|
|
|
|
/**
|
|
* Whether a conversation is currently being processed by a LIVE sync agent in
|
|
* this process — i.e. registered and still within the max-age cap. A stale entry
|
|
* (older than the cap) is treated as not-live AND evicted, so the watchdog reaps it.
|
|
* @param {number|string} conversationId
|
|
* @returns {boolean}
|
|
*/
|
|
export function isSyncActive(conversationId) {
|
|
if (conversationId == null) return false;
|
|
const startedAt = _activeSync.get(Number(conversationId));
|
|
if (startedAt === undefined) return false;
|
|
if (Date.now() - startedAt >= SYNC_ACTIVITY_MAX_MS) {
|
|
_activeSync.delete(Number(conversationId));
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
/** Test/diagnostics helper: number of live sync conversations. */
|
|
export function activeSyncCount() {
|
|
return _activeSync.size;
|
|
}
|