@triflux/remote 10.42.0 → 10.44.0

package/cto/hygiene.mjs

@@ -11,7 +11,7 @@ import {
 } from "node:fs";
 import { homedir } from "node:os";
 import { basename, dirname, join } from "node:path";
-import { getCtoHygieneApplyMode } from "@triflux/core/hub/lib/cto-env.mjs";
+import { getCtoHygieneApplyMode } from "../hub/lib/cto-env.mjs";
 import { appendCtoEvent } from "./events.mjs";
 import { applyHygieneArchiveActions } from "./hygiene-actions.mjs";
 import { resolveLakeRootDir } from "./lake-root.mjs";

package/hub/lib/cto-env.mjs

@@ -0,0 +1 @@
+export * from "@triflux/core/hub/lib/cto-env.mjs";

package/hub/pipe.mjs

@@ -269,6 +269,15 @@ export function createPipeServer({
           payload.heartbeat_ttl_ms || heartbeatTtlMs,
         );
         if (client) touchClient(client);
+        if (result?.effective?.updated === false) {
+          return {
+            ok: false,
+            error: {
+              code: "AGENT_NOT_FOUND",
+              message: `heartbeat target not registered: ${agentId}`,
+            },
+          };
+        }
         return { ok: true, data: result };
       }
 

package/hub/schema.sql

@@ -11,7 +11,11 @@ CREATE TABLE IF NOT EXISTS agents (
   last_seen_ms INTEGER NOT NULL,
   lease_expires_ms INTEGER NOT NULL,
   status TEXT NOT NULL CHECK (status IN ('online','stale','offline')),
-  metadata_json TEXT NOT NULL DEFAULT '{}'
+  metadata_json TEXT NOT NULL DEFAULT '{}',
+  project_id TEXT,
+  session_id TEXT,
+  host_id TEXT,
+  transport_locators_json TEXT NOT NULL DEFAULT '[]'
 );
 
 -- 메시지 테이블
@@ -28,7 +32,61 @@ CREATE TABLE IF NOT EXISTS messages (
   correlation_id TEXT NOT NULL,
   trace_id TEXT NOT NULL,
   payload_json TEXT NOT NULL DEFAULT '{}',
-  status TEXT NOT NULL CHECK (status IN ('queued','delivered','acked','expired','dead_letter'))
+  status TEXT NOT NULL CHECK (status IN ('queued','delivered','acked','expired','dead_letter')),
+  role_key_wire TEXT
+);
+
+-- 동적 역할 레지스트리 (schema v6)
+CREATE TABLE IF NOT EXISTS role_registry (
+  role_key_wire TEXT PRIMARY KEY,
+  project_id TEXT NOT NULL,
+  role_kind TEXT NOT NULL CHECK (role_kind IN ('cto','lead')),
+  scope_id TEXT NOT NULL DEFAULT '',
+  state TEXT NOT NULL CHECK (
+    state IN (
+      'electable',
+      'elected-probing',
+      'active',
+      'unreachable',
+      'quarantined'
+    )
+  ),
+  holder_agent_id TEXT,
+  previous_holder_agent_id TEXT,
+  epoch INTEGER NOT NULL DEFAULT 0 CHECK (epoch >= 0),
+  activation_seq INTEGER NOT NULL DEFAULT 0 CHECK (activation_seq >= 0),
+  activation_id TEXT,
+  activation_deadline_ms INTEGER,
+  holder_lease_expires_ms INTEGER,
+  charter_version TEXT,
+  charter_acked_epoch INTEGER,
+  retry_count INTEGER NOT NULL DEFAULT 0,
+  next_probe_ms INTEGER,
+  blocked_reason TEXT,
+  last_transition_ms INTEGER NOT NULL,
+  last_reason TEXT NOT NULL,
+  version INTEGER NOT NULL DEFAULT 0,
+  CHECK (
+    (role_kind = 'cto' AND scope_id = '') OR
+    (role_kind = 'lead' AND scope_id <> '')
+  ),
+  UNIQUE(project_id, role_kind, scope_id)
+);
+
+CREATE TABLE IF NOT EXISTS role_candidates (
+  role_key_wire TEXT NOT NULL,
+  agent_id TEXT NOT NULL,
+  source TEXT NOT NULL CHECK (source IN ('grant','standup','operator')),
+  priority INTEGER NOT NULL DEFAULT 0,
+  transport_locators_json TEXT NOT NULL DEFAULT '[]',
+  consecutive_probe_failures INTEGER NOT NULL DEFAULT 0,
+  excluded_until_ms INTEGER,
+  last_probe_ms INTEGER,
+  last_probe_code TEXT,
+  updated_at_ms INTEGER NOT NULL,
+  PRIMARY KEY(role_key_wire, agent_id),
+  FOREIGN KEY(role_key_wire)
+    REFERENCES role_registry(role_key_wire) ON DELETE CASCADE
 );
 
 -- 메시지 수신함 (배달 추적)
@@ -73,6 +131,7 @@ CREATE INDEX IF NOT EXISTS idx_messages_correlation ON messages(correlation_id);
 CREATE INDEX IF NOT EXISTS idx_messages_trace ON messages(trace_id);
 CREATE INDEX IF NOT EXISTS idx_messages_expires ON messages(expires_at_ms);
 CREATE INDEX IF NOT EXISTS idx_messages_priority ON messages(priority DESC, created_at_ms ASC);
+CREATE INDEX IF NOT EXISTS idx_messages_role_key ON messages(role_key_wire, status, expires_at_ms);
 CREATE INDEX IF NOT EXISTS idx_inbox_agent ON message_inbox(agent_id, delivered_at_ms);
 CREATE INDEX IF NOT EXISTS idx_inbox_message ON message_inbox(message_id);
 CREATE INDEX IF NOT EXISTS idx_human_requests_state ON human_requests(state);
@@ -103,6 +162,7 @@ CREATE TABLE IF NOT EXISTS assign_jobs (
   created_at_ms INTEGER NOT NULL,
   updated_at_ms INTEGER NOT NULL,
   started_at_ms INTEGER,
+  dispatched_at_ms INTEGER,
   completed_at_ms INTEGER,
   last_retry_at_ms INTEGER
 );

package/hub/server.mjs

@@ -44,6 +44,13 @@ import {
   cleanupStaleFsmonitorDaemons,
 } from "@triflux/core/hub/lib/process-utils.mjs";
 import * as spawnTrace from "@triflux/core/hub/lib/spawn-trace.mjs";
+import {
+  clampDurationMs,
+  DEFAULT_ASSIGN_TIMEOUT_MS,
+  DEFAULT_ASSIGN_TTL_MS,
+  DEFAULT_REGISTER_TIMEOUT_SEC,
+  resolveWorkerLeaseTtlMs,
+} from "@triflux/core/hub/lib/timeout-defaults.mjs";
 import {
   recordRequest,
   recordWorker,
@@ -143,7 +150,7 @@ function buildHubUrl(host, port) {
 
 function parseHubPort(value) {
   const parsed = Number.parseInt(String(value ?? ""), 10);
-  return Number.isFinite(parsed) && parsed > 0 ? parsed : null;
+  return Number.isFinite(parsed) && parsed >= 0 ? parsed : null;
 }
 
 function isPidAlive(pid, killFn = process.kill) {
@@ -963,16 +970,23 @@ export async function startHub({
   createDelegatorWorker = createDelegatorMcpWorker,
 } = {}) {
   const resolvedPort = parseHubPort(portOpt);
-  const portSpecified = Number.isFinite(resolvedPort) && resolvedPort > 0;
+  const portSpecified = Number.isFinite(resolvedPort);
+  const ephemeralPort = resolvedPort === 0;
   const livePeer = detectLivePeer();
   const livePidPort = parseHubPort(livePeer?.port);
-  const port = portSpecified
+  let port = portSpecified
     ? resolvedPort
     : resolveHubPort(process.env, {
         detectPeer: () => livePeer,
       });
 
-  if (portSpecified && livePeer.alive && livePidPort && port !== livePidPort) {
+  if (
+    portSpecified &&
+    !ephemeralPort &&
+    livePeer.alive &&
+    livePidPort &&
+    port !== livePidPort
+  ) {
     hubLog.warn(
       {
         requestedPort: port,
@@ -983,12 +997,14 @@ export async function startHub({
     );
   }
 
-  const existingHub = await tryReuseExistingHub({
-    port,
-    portSpecified,
-    host,
-    detectPeer: () => livePeer,
-  });
+  const existingHub = ephemeralPort
+    ? null
+    : await tryReuseExistingHub({
+        port,
+        portSpecified,
+        host,
+        detectPeer: () => livePeer,
+      });
   if (existingHub) return existingHub;
 
   attachBrokerDiagnostics(brokerInstance);
@@ -1026,21 +1042,23 @@ export async function startHub({
     lockHeld = false;
   };
 
-  const lockedExistingHub = await tryReuseExistingHub({
-    port,
-    portSpecified,
-    host,
-    detectPeer: () => detectLivePeer(),
-  });
+  const lockedExistingHub = ephemeralPort
+    ? null
+    : await tryReuseExistingHub({
+        port,
+        portSpecified,
+        host,
+        detectPeer: () => detectLivePeer(),
+      });
   if (lockedExistingHub) {
     releaseStartupLock();
     return lockedExistingHub;
   }
 
   const HUB_TOKEN = process.env.TFX_HUB_TOKEN?.trim() || null;
-  if (HUB_TOKEN) {
+  if (!ephemeralPort && HUB_TOKEN) {
     writeOwnedTokenFile({ token: HUB_TOKEN, port });
-  } else {
+  } else if (!ephemeralPort) {
     cleanupOwnedTokenFile({ token: HUB_TOKEN, port });
   }
 
@@ -1540,6 +1558,38 @@ export async function startHub({
         try {
           const body = await parseBody(req);
           const { sessionId } = body || {};
+          if (body?.agent_id && body?.cli) {
+            const registered = router.registerAgent({
+              agent_id: body.agent_id,
+              cli: body.cli,
+              capabilities: Array.isArray(body.capabilities)
+                ? body.capabilities
+                : [],
+              topics: Array.isArray(body.topics) ? body.topics : [],
+              heartbeat_ttl_ms: body.heartbeat_ttl_ms ?? 300000,
+              metadata:
+                body.metadata && typeof body.metadata === "object"
+                  ? body.metadata
+                  : { source: "session_start" },
+              ...(Object.hasOwn(body, "project_id")
+                ? { project_id: body.project_id }
+                : {}),
+              ...(Object.hasOwn(body, "session_id")
+                ? { session_id: body.session_id }
+                : {}),
+              ...(Object.hasOwn(body, "host_id")
+                ? { host_id: body.host_id }
+                : {}),
+              ...(Object.hasOwn(body, "transport_locators_json")
+                ? { transport_locators_json: body.transport_locators_json }
+                : {}),
+            });
+            if (!registered?.ok) {
+              throw new Error(
+                registered?.error?.message || "hub agent register failed",
+              );
+            }
+          }
           const result = synapseRegistry.register(sessionId, body);
           if (!result?.ok) {
             throw new Error(result?.reason || "register failed");
@@ -1690,10 +1740,14 @@ export async function startHub({
             const {
               agent_id,
               cli,
-              timeout_sec = 600,
+              timeout_sec = DEFAULT_REGISTER_TIMEOUT_SEC,
               topics = [],
               capabilities = [],
               metadata = {},
+              project_id,
+              session_id,
+              host_id,
+              transport_locators_json,
             } = body;
             if (!agent_id || !cli) {
               return writeJson(res, 400, {
@@ -1709,7 +1763,12 @@ export async function startHub({
               metadata?.task ||
               cli;
             const workerStartedAt = performance.now();
-            const heartbeat_ttl_ms = (timeout_sec + 120) * 1000;
+            const hasExplicitTimeout = Number.isFinite(
+              Number(body.timeout_sec),
+            );
+            const heartbeat_ttl_ms = hasExplicitTimeout
+              ? (timeout_sec + 120) * 1000
+              : Math.max((timeout_sec + 120) * 1000, resolveWorkerLeaseTtlMs());
             const result = await pipe.executeCommand("register", {
               agent_id,
               cli,
@@ -1717,6 +1776,12 @@ export async function startHub({
               topics,
               heartbeat_ttl_ms,
               metadata,
+              ...(project_id === undefined ? {} : { project_id }),
+              ...(session_id === undefined ? {} : { session_id }),
+              ...(host_id === undefined ? {} : { host_id }),
+              ...(transport_locators_json === undefined
+                ? {}
+                : { transport_locators_json }),
             });
             if (result.ok) {
               workerSpans.set(agent_id, {
@@ -1725,7 +1790,7 @@ export async function startHub({
               });
               recordWorker(agent_id, "spawned", command, 0);
             }
-            return writeJson(res, 200, result);
+            return writeJson(res, result.ok ? 200 : 409, result);
           }
 
           if (path === "/bridge/result" && req.method === "POST") {
@@ -1750,6 +1815,20 @@ export async function startHub({
             return writeJson(res, 200, result);
           }
 
+          if (path === "/bridge/heartbeat" && req.method === "POST") {
+            const { agent_id, ttl_ms } = body;
+            if (!agent_id)
+              return writeJson(res, 400, { ok: false, error: "agent_id 필수" });
+            const result = await pipe.executeCommand("heartbeat", {
+              agent_id,
+              heartbeat_ttl_ms: clampDurationMs(
+                ttl_ms,
+                resolveWorkerLeaseTtlMs(),
+              ),
+            });
+            return writeJson(res, result.ok ? 200 : 404, result);
+          }
+
           if (path === "/bridge/control" && req.method === "POST") {
             const {
               from_agent = "lead",
@@ -1881,8 +1960,8 @@ export async function startHub({
               topic = "assign.job",
               payload = {},
               priority = 5,
-              ttl_ms = 600000,
-              timeout_ms = 600000,
+              ttl_ms = DEFAULT_ASSIGN_TTL_MS,
+              timeout_ms = DEFAULT_ASSIGN_TIMEOUT_MS,
               max_retries = 0,
               trace_id,
               correlation_id,
@@ -2352,7 +2431,7 @@ export async function startHub({
     try {
       store.close();
     } catch {}
-    if (!preserveTokenFile) {
+    if (!ephemeralPort && !preserveTokenFile) {
       cleanupOwnedTokenFile({ token: HUB_TOKEN, port });
     }
     releaseStartupLock();
@@ -2426,6 +2505,9 @@ export async function startHub({
       httpServer.listen(port, host, () => {
         httpServer.off("error", onError);
         try {
+          const boundAddress = httpServer.address();
+          if (boundAddress && typeof boundAddress === "object")
+            port = boundAddress.port;
           let idleTimer = null;
           let stopPromise = null;
 
@@ -2446,25 +2528,28 @@ export async function startHub({
             idleTimeoutMs: hubIdleTimeoutMs,
           };
 
-          writeState({
-            pid: process.pid,
-            port,
-            host,
-            auth_mode: HUB_TOKEN ? "token-required" : "localhost-only",
-            url: info.url,
-            pipe_path: pipe.path,
-            pipePath: pipe.path,
-            assign_callback_pipe_path: assignCallbacks.path,
-            assignCallbackPipePath: assignCallbacks.path,
-            authMode: HUB_TOKEN ? "token-required" : "localhost-only",
-            startedAt,
-            started: startedAtMs,
-            version,
-            sessionId,
-            session_id: sessionId,
-          });
+          if (!ephemeralPort) {
+            writeState({
+              pid: process.pid,
+              port,
+              host,
+              auth_mode: HUB_TOKEN ? "token-required" : "localhost-only",
+              url: info.url,
+              pipe_path: pipe.path,
+              pipePath: pipe.path,
+              assign_callback_pipe_path: assignCallbacks.path,
+              assignCallbackPipePath: assignCallbacks.path,
+              authMode: HUB_TOKEN ? "token-required" : "localhost-only",
+              startedAt,
+              started: startedAtMs,
+              version,
+              sessionId,
+              session_id: sessionId,
+            });
+          }
           releaseStartupLock();
-          void syncHubMcpSettingsIfAvailable({ hubUrl: info.url });
+          if (!ephemeralPort)
+            void syncHubMcpSettingsIfAvailable({ hubUrl: info.url });
 
           hubLog.info(
             {
@@ -2551,8 +2636,10 @@ export async function startHub({
                 synapseRegistry.destroy();
               } catch {}
               store.close();
-              cleanupOwnedPidFile();
-              cleanupOwnedTokenFile({ token: HUB_TOKEN, port });
+              if (!ephemeralPort) {
+                cleanupOwnedPidFile();
+                cleanupOwnedTokenFile({ token: HUB_TOKEN, port });
+              }
               httpServer.closeAllConnections();
               await new Promise((resolveClose) =>
                 httpServer.close(resolveClose),