@triflux/remote 10.0.1 → 10.25.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (155) hide show
  1. package/hub/pipe.mjs +239 -121
  2. package/hub/public/dashboard.html +68 -7
  3. package/hub/server.mjs +2181 -701
  4. package/hub/store-adapter.mjs +101 -37
  5. package/hub/store.mjs +356 -153
  6. package/hub/team/agent-map.json +26 -7
  7. package/hub/team/ansi.mjs +245 -146
  8. package/hub/team/backend.mjs +84 -17
  9. package/hub/team/build-worker-prompt.mjs +42 -0
  10. package/hub/team/check-mcp-hub.mjs +49 -0
  11. package/hub/team/claude-daemon-control.mjs +219 -0
  12. package/hub/team/claude-native-bridge.mjs +784 -0
  13. package/hub/team/claude-session-projection.mjs +65 -0
  14. package/hub/team/cli/commands/attach.mjs +57 -37
  15. package/hub/team/cli/commands/control.mjs +29 -8
  16. package/hub/team/cli/commands/debug.mjs +95 -74
  17. package/hub/team/cli/commands/focus.mjs +80 -53
  18. package/hub/team/cli/commands/interrupt.mjs +18 -6
  19. package/hub/team/cli/commands/kill.mjs +16 -5
  20. package/hub/team/cli/commands/list.mjs +31 -24
  21. package/hub/team/cli/commands/send.mjs +19 -6
  22. package/hub/team/cli/commands/start/index.mjs +154 -33
  23. package/hub/team/cli/commands/start/parse-args.mjs +89 -12
  24. package/hub/team/cli/commands/start/start-headless.mjs +127 -37
  25. package/hub/team/cli/commands/start/start-in-process.mjs +50 -40
  26. package/hub/team/cli/commands/start/start-mux.mjs +122 -73
  27. package/hub/team/cli/commands/start/start-wt.mjs +86 -69
  28. package/hub/team/cli/commands/status.mjs +43 -14
  29. package/hub/team/cli/commands/stop.mjs +11 -4
  30. package/hub/team/cli/commands/task.mjs +8 -3
  31. package/hub/team/cli/commands/tasks.mjs +13 -13
  32. package/hub/team/cli/index.mjs +2 -2
  33. package/hub/team/cli/manifest.mjs +38 -8
  34. package/hub/team/cli/render.mjs +52 -30
  35. package/hub/team/cli/services/attach-fallback.mjs +74 -54
  36. package/hub/team/cli/services/hub-client.mjs +45 -17
  37. package/hub/team/cli/services/member-selector.mjs +37 -30
  38. package/hub/team/cli/services/native-control.mjs +144 -116
  39. package/hub/team/cli/services/runtime-mode.mjs +55 -1
  40. package/hub/team/cli/services/state-store.mjs +66 -9
  41. package/hub/team/cli/services/task-model.mjs +40 -30
  42. package/hub/team/codex-review.mjs +521 -0
  43. package/hub/team/conductor-mesh-bridge.mjs +25 -24
  44. package/hub/team/conductor-registry.mjs +121 -0
  45. package/hub/team/conductor.mjs +1379 -671
  46. package/hub/team/consensus-meta.mjs +176 -0
  47. package/hub/team/dashboard-anchor.mjs +4 -5
  48. package/hub/team/dashboard-layout.mjs +3 -1
  49. package/hub/team/dashboard-open.mjs +21 -119
  50. package/hub/team/dashboard.mjs +76 -28
  51. package/hub/team/event-log.mjs +18 -10
  52. package/hub/team/execution-mode.mjs +241 -0
  53. package/hub/team/extract-completion-payload.mjs +110 -0
  54. package/hub/team/git-preflight.mjs +286 -0
  55. package/hub/team/handoff.mjs +49 -18
  56. package/hub/team/headless-bridge-session.mjs +17 -0
  57. package/hub/team/headless.mjs +2065 -1156
  58. package/hub/team/health-probe.mjs +232 -69
  59. package/hub/team/launcher-template.mjs +23 -13
  60. package/hub/team/lead-control.mjs +91 -4
  61. package/hub/team/mcp-selector.mjs +145 -0
  62. package/hub/team/native-supervisor.mjs +90 -26
  63. package/hub/team/native.mjs +74 -35
  64. package/hub/team/nativeProxy.mjs +184 -113
  65. package/hub/team/notify.mjs +147 -78
  66. package/hub/team/orchestrator.mjs +173 -161
  67. package/hub/team/pane.mjs +43 -15
  68. package/hub/team/process-cleanup.mjs +69 -25
  69. package/hub/team/psmux.mjs +1067 -264
  70. package/hub/team/recovery-store.mjs +149 -0
  71. package/hub/team/remote-probe.mjs +69 -53
  72. package/hub/team/remote-session.mjs +131 -59
  73. package/hub/team/remote-watcher.mjs +62 -34
  74. package/hub/team/retry-state-machine.mjs +267 -0
  75. package/hub/team/routing.mjs +51 -25
  76. package/hub/team/runtime-strategy.mjs +200 -0
  77. package/hub/team/schemas/worker-completion.json +39 -0
  78. package/hub/team/sentinel-capture.mjs +127 -0
  79. package/hub/team/session-sync.mjs +153 -6
  80. package/hub/team/session.mjs +605 -611
  81. package/hub/team/shared.mjs +13 -13
  82. package/hub/team/staleState.mjs +72 -30
  83. package/hub/team/swarm-cli.mjs +390 -0
  84. package/hub/team/swarm-hypervisor.mjs +1526 -198
  85. package/hub/team/swarm-intent.mjs +216 -0
  86. package/hub/team/swarm-locks.mjs +140 -25
  87. package/hub/team/swarm-planner.mjs +121 -21
  88. package/hub/team/swarm-preflight.mjs +317 -0
  89. package/hub/team/swarm-reconciler.mjs +128 -23
  90. package/hub/team/synapse-cli.mjs +458 -0
  91. package/hub/team/synapse-http.mjs +60 -0
  92. package/hub/team/synapse-registry.mjs +329 -0
  93. package/hub/team/terminal-opener.mjs +216 -0
  94. package/hub/team/tui-core.mjs +288 -0
  95. package/hub/team/tui-lite.mjs +234 -146
  96. package/hub/team/tui-remote-adapter.mjs +21 -37
  97. package/hub/team/tui-synapse.mjs +223 -0
  98. package/hub/team/tui-viewer.mjs +104 -45
  99. package/hub/team/tui-widgets.mjs +262 -0
  100. package/hub/team/tui.mjs +936 -449
  101. package/hub/team/worker-completion-validator.mjs +84 -0
  102. package/hub/team/worker-sandbox.mjs +93 -0
  103. package/hub/team/worker-signal.mjs +263 -0
  104. package/hub/team/worker-signal.types.d.ts +52 -0
  105. package/hub/team/worktree-lifecycle.mjs +518 -64
  106. package/hub/team/wt-manager.mjs +297 -26
  107. package/hub/tools.mjs +474 -252
  108. package/hub/tray.mjs +411 -368
  109. package/hub/workers/claude-worker.mjs +162 -118
  110. package/hub/workers/codex-app-server-worker.mjs +1015 -0
  111. package/hub/workers/codex-mcp.mjs +724 -507
  112. package/hub/workers/delegator-mcp.mjs +663 -337
  113. package/hub/workers/factory.mjs +100 -21
  114. package/hub/workers/gemini-worker.mjs +179 -86
  115. package/hub/workers/interface.mjs +23 -2
  116. package/hub/workers/lib/jsonrpc-stdio.mjs +463 -0
  117. package/hub/workers/worker-utils.mjs +22 -16
  118. package/package.json +4 -3
  119. package/scripts/lib/agent-json.mjs +27 -0
  120. package/scripts/lib/async.mjs +174 -0
  121. package/scripts/lib/claudemd-scanner.mjs +223 -0
  122. package/scripts/lib/cli-agy.mjs +62 -0
  123. package/scripts/lib/cli-claude.mjs +78 -0
  124. package/scripts/lib/cli-codex.mjs +199 -0
  125. package/scripts/lib/cli-gemini.mjs +67 -0
  126. package/scripts/lib/codex-recovery.sh +50 -0
  127. package/scripts/lib/context.mjs +67 -0
  128. package/scripts/lib/cross-review-utils.mjs +59 -0
  129. package/scripts/lib/doctor-env-checks.mjs +121 -0
  130. package/scripts/lib/dynamic-route-cli.mjs +107 -0
  131. package/scripts/lib/env-probe.mjs +269 -0
  132. package/scripts/lib/env.mjs +14 -0
  133. package/scripts/lib/gemini-profiles.mjs +119 -0
  134. package/scripts/lib/handoff.mjs +187 -0
  135. package/scripts/lib/hook-utils.mjs +16 -0
  136. package/scripts/lib/hub.mjs +48 -0
  137. package/scripts/lib/keyword-rules.mjs +195 -0
  138. package/scripts/lib/logger.mjs +105 -0
  139. package/scripts/lib/mcp-filter.mjs +883 -0
  140. package/scripts/lib/mcp-gateway-servers.mjs +70 -0
  141. package/scripts/lib/mcp-guard-engine.mjs +2120 -0
  142. package/scripts/lib/mcp-health.mjs +773 -0
  143. package/scripts/lib/mcp-manifest.mjs +89 -0
  144. package/scripts/lib/mcp-server-catalog.mjs +355 -0
  145. package/scripts/lib/pid.mjs +63 -0
  146. package/scripts/lib/process-utils.mjs +26 -0
  147. package/scripts/lib/psmux-info.mjs +179 -0
  148. package/scripts/lib/quota.mjs +47 -0
  149. package/scripts/lib/remote-spawn-transfer.mjs +226 -0
  150. package/scripts/lib/skill-state.mjs +220 -0
  151. package/scripts/lib/skill-template.mjs +375 -0
  152. package/scripts/lib/team.mjs +75 -0
  153. package/scripts/lib/timeout.mjs +66 -0
  154. package/scripts/lib/tmp.mjs +26 -0
  155. package/scripts/lib/toml.mjs +113 -0
package/hub/server.mjs CHANGED
@@ -1,74 +1,130 @@
1
1
  // hub/server.mjs — HTTP MCP + REST bridge + Named Pipe 서버 진입점
2
- import { createServer as createHttpServer } from 'node:http';
3
- import { createHash, randomUUID, timingSafeEqual } from 'node:crypto';
4
- import { extname, join, resolve, sep } from 'node:path';
5
- import { homedir } from 'node:os';
6
- import { writeFileSync, unlinkSync, existsSync, mkdirSync, readFileSync } from 'node:fs';
7
- import { fileURLToPath } from 'node:url';
8
- import { execSync as execSyncHub } from 'node:child_process';
9
-
10
- import { Server } from '@modelcontextprotocol/sdk/server/index.js';
11
- import { StreamableHTTPServerTransport } from '@modelcontextprotocol/sdk/server/streamableHttp.js';
12
- import { ListToolsRequestSchema, CallToolRequestSchema } from '@modelcontextprotocol/sdk/types.js';
13
-
14
- import { createStoreAdapter } from './store-adapter.mjs';
15
- import { createRouter } from '@triflux/core/hub/router.mjs';
16
- import { createHitlManager } from '@triflux/core/hub/hitl.mjs';
17
- import { createPipeServer } from './pipe.mjs';
18
- import { createAssignCallbackServer } from '@triflux/core/hub/assign-callbacks.mjs';
19
- import { createTools } from './tools.mjs';
20
- import { DelegatorService } from '@triflux/core/hub/delegator/index.mjs';
21
- import { createDelegatorMcpWorker } from './workers/delegator-mcp.mjs';
22
- import { cleanupOrphanNodeProcesses } from '@triflux/core/hub/lib/process-utils.mjs';
23
- import { createModuleLogger } from '../scripts/lib/logger.mjs';
24
- import { wrapRequestHandler } from '@triflux/core/hub/middleware/request-logger.mjs';
25
- import { acquireLock, getVersionHash, isServerHealthy, readState, releaseLock, writeState } from '@triflux/core/hub/state.mjs';
26
- import { createAdaptiveFingerprintService } from '@triflux/core/hub/session-fingerprint.mjs';
27
- import { createAdaptiveEngine } from '@triflux/core/hub/adaptive.mjs';
28
- import { registerTeamBridge } from '@triflux/core/hub/team-bridge.mjs';
29
- import { nativeProxy } from './team/nativeProxy.mjs';
2
+
3
+ import { execSync as execSyncHub } from "node:child_process";
4
+ import { createHash, randomUUID, timingSafeEqual } from "node:crypto";
5
+ import { EventEmitter } from "node:events";
6
+ import {
7
+ existsSync,
8
+ mkdirSync,
9
+ readFileSync,
10
+ unlinkSync,
11
+ writeFileSync,
12
+ } from "node:fs";
13
+ import { createServer as createHttpServer } from "node:http";
14
+ import { homedir } from "node:os";
15
+ import { extname, join, resolve, sep } from "node:path";
16
+ import { performance } from "node:perf_hooks";
17
+ import { fileURLToPath } from "node:url";
18
+
19
+ import { Server } from "@modelcontextprotocol/sdk/server/index.js";
20
+ import { StreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/streamableHttp.js";
21
+ import {
22
+ CallToolRequestSchema,
23
+ ListToolsRequestSchema,
24
+ } from "@modelcontextprotocol/sdk/types.js";
25
+ import { createModuleLogger } from "../scripts/lib/logger.mjs";
26
+ import { broker as brokerInstance, reloadBroker } from "@triflux/core/hub/account-broker.mjs";
27
+ import { createAdaptiveEngine } from "@triflux/core/hub/adaptive.mjs";
28
+ import { createAssignCallbackServer } from "@triflux/core/hub/assign-callbacks.mjs";
29
+ import { DelegatorService } from "@triflux/core/hub/delegator/index.mjs";
30
+ import { createHitlManager } from "@triflux/core/hub/hitl.mjs";
31
+ import {
32
+ cleanupOrphanNodeProcesses,
33
+ cleanupOrphanRuntimeProcesses,
34
+ cleanupStaleFsmonitorDaemons,
35
+ } from "@triflux/core/hub/lib/process-utils.mjs";
36
+ import * as spawnTrace from "@triflux/core/hub/lib/spawn-trace.mjs";
37
+ import {
38
+ recordRequest,
39
+ recordWorker,
40
+ snapshot as traceSnapshot,
41
+ } from "@triflux/core/hub/lib/trace-recorder.mjs";
42
+ import { logQuotaRefreshFailures } from "@triflux/core/hub/middleware/quota-middleware.mjs";
43
+ import { wrapRequestHandler } from "@triflux/core/hub/middleware/request-logger.mjs";
44
+ import { createPipeServer } from "./pipe.mjs";
45
+ import { createRouter } from "@triflux/core/hub/router.mjs";
46
+ import { createAdaptiveFingerprintService } from "@triflux/core/hub/session-fingerprint.mjs";
47
+ import {
48
+ acquireLock,
49
+ getVersionHash,
50
+ isServerHealthy,
51
+ readState,
52
+ releaseLock,
53
+ writeState,
54
+ } from "@triflux/core/hub/state.mjs";
55
+ import { createStoreAdapter } from "./store-adapter.mjs";
56
+ import { createGitPreflight } from "./team/git-preflight.mjs";
57
+ import { nativeProxy } from "./team/nativeProxy.mjs";
58
+ import { createSwarmLocks } from "./team/swarm-locks.mjs";
59
+ import { createSynapseRegistry } from "./team/synapse-registry.mjs";
60
+ import { registerTeamBridge } from "@triflux/core/hub/team-bridge.mjs";
61
+ import { createTools } from "./tools.mjs";
62
+ import { createDelegatorMcpWorker } from "./workers/delegator-mcp.mjs";
30
63
 
31
64
  registerTeamBridge(nativeProxy);
32
65
 
33
- const hubLog = createModuleLogger('hub');
66
+ const hubLog = createModuleLogger("hub");
67
+ const brokerDiagnosticsAttached = new WeakSet();
34
68
 
35
69
  const MAX_BODY_SIZE = 1024 * 1024;
36
- const PUBLIC_PATHS = new Set(['/', '/status', '/health', '/healthz']);
37
- const RATE_LIMIT_MAX = 100; // requests per window
70
+ const PUBLIC_PATHS = new Set(["/", "/status", "/health", "/healthz"]);
71
+ const RATE_LIMIT_MAX = 100; // requests per window
38
72
  const RATE_LIMIT_WINDOW_MS = 60 * 1000; // 1 minute sliding window
39
- const LOOPBACK_REMOTE_ADDRESSES = new Set(['127.0.0.1', '::1', '::ffff:127.0.0.1']);
40
- const ALLOWED_ORIGIN_RE = /^https?:\/\/(localhost|127\.0\.0\.1|\[::1\])(:\d+)?$/i;
41
- const PROJECT_ROOT = fileURLToPath(new URL('..', import.meta.url));
42
- const PUBLIC_DIR = resolve(join(PROJECT_ROOT, 'hub', 'public'));
43
- const CACHE_DIR = join(homedir(), '.claude', 'cache');
44
- const BATCH_EVENTS_PATH = join(CACHE_DIR, 'batch-events.jsonl');
45
- const SV_ACCUMULATOR_PATH = join(CACHE_DIR, 'sv-accumulator.json');
46
- const CODEX_RATE_LIMITS_CACHE_PATH = join(CACHE_DIR, 'codex-rate-limits-cache.json');
47
- const GEMINI_QUOTA_CACHE_PATH = join(CACHE_DIR, 'gemini-quota-cache.json');
48
- const CLAUDE_USAGE_CACHE_PATH = join(CACHE_DIR, 'claude-usage-cache.json');
73
+ const LOOPBACK_REMOTE_ADDRESSES = new Set([
74
+ "127.0.0.1",
75
+ "::1",
76
+ "::ffff:127.0.0.1",
77
+ ]);
78
+ const ALLOWED_ORIGIN_RE =
79
+ /^https?:\/\/(localhost|127\.0\.0\.1|\[::1\])(:\d+)?$/i;
80
+ const PROJECT_ROOT = fileURLToPath(new URL("..", import.meta.url));
81
+ const PUBLIC_DIR = resolve(join(PROJECT_ROOT, "hub", "public"));
82
+ const CACHE_DIR = join(homedir(), ".claude", "cache");
83
+ const HUB_DEFAULT_PORT = 27888;
84
+ const BATCH_EVENTS_PATH = join(CACHE_DIR, "batch-events.jsonl");
85
+ const SV_ACCUMULATOR_PATH = join(CACHE_DIR, "sv-accumulator.json");
86
+ const CODEX_RATE_LIMITS_CACHE_PATH = join(
87
+ CACHE_DIR,
88
+ "codex-rate-limits-cache.json",
89
+ );
90
+ const GEMINI_QUOTA_CACHE_PATH = join(CACHE_DIR, "gemini-quota-cache.json");
91
+ const CLAUDE_USAGE_CACHE_PATH = join(CACHE_DIR, "claude-usage-cache.json");
49
92
  const AIMD_WINDOW_MS = 30 * 60 * 1000;
50
93
  const AIMD_INITIAL_BATCH_SIZE = 3;
51
94
  const AIMD_MIN_BATCH_SIZE = 1;
52
95
  const AIMD_MAX_BATCH_SIZE = 10;
53
- const HUB_IDLE_TIMEOUT_DEFAULT_MS = 10 * 60 * 1000;
96
+ const SYNAPSE_VALID_OPS = new Set([
97
+ "checkout",
98
+ "rebase",
99
+ "cherry-pick",
100
+ "reset",
101
+ "stash-pop",
102
+ "worktree-remove",
103
+ ]);
104
+ const HUB_IDLE_TIMEOUT_DEFAULT_MS = 0; // 0 = 영구 실행 (idle shutdown 비활성). TFX_HUB_IDLE_TIMEOUT_MS 환경변수로 오버라이드 가능
54
105
  const HUB_IDLE_SWEEP_DEFAULT_MS = 60 * 1000;
55
106
  const STATIC_CONTENT_TYPES = Object.freeze({
56
- '.html': 'text/html',
57
- '.css': 'text/css',
58
- '.js': 'application/javascript',
59
- '.png': 'image/png',
107
+ ".html": "text/html",
108
+ ".css": "text/css",
109
+ ".js": "application/javascript",
110
+ ".png": "image/png",
60
111
  });
61
112
 
62
113
  // IP-based sliding window rate limiter (in-memory, no external deps)
63
114
  // Each entry is an array of request timestamps within the current window.
64
115
  const rateLimitMap = new Map();
65
116
 
66
- function formatHostForUrl(host = '127.0.0.1') {
67
- return String(host).includes(':') ? `[${host}]` : host;
117
+ function formatHostForUrl(host = "127.0.0.1") {
118
+ return String(host).includes(":") ? `[${host}]` : host;
68
119
  }
69
120
 
70
121
  function buildHubUrl(host, port) {
71
- return `http://${formatHostForUrl(host || '127.0.0.1')}:${port}/mcp`;
122
+ return `http://${formatHostForUrl(host || "127.0.0.1")}:${port}/mcp`;
123
+ }
124
+
125
+ function parseHubPort(value) {
126
+ const parsed = Number.parseInt(String(value ?? ""), 10);
127
+ return Number.isFinite(parsed) && parsed > 0 ? parsed : null;
72
128
  }
73
129
 
74
130
  function isPidAlive(pid, killFn = process.kill) {
@@ -82,24 +138,45 @@ function isPidAlive(pid, killFn = process.kill) {
82
138
  }
83
139
  }
84
140
 
85
- async function tryReuseExistingHub({
141
+ export async function tryReuseExistingHub({
86
142
  port,
87
- host = '127.0.0.1',
143
+ portSpecified = Number.isFinite(Number(port)) && Number(port) > 0,
144
+ host = "127.0.0.1",
88
145
  readCurrentState = readState,
89
146
  readInfo = getHubInfo,
90
147
  checkHealth = isServerHealthy,
91
148
  killFn = process.kill,
149
+ detectPeer = detectLivePeer,
150
+ log = hubLog,
92
151
  } = {}) {
93
152
  const existing = readCurrentState();
94
153
  const existingPort = Number(existing?.port);
95
- if (!isPidAlive(existing?.pid, killFn) || !Number.isFinite(existingPort) || existingPort <= 0) {
154
+ const requestedPort = parseHubPort(port) ?? HUB_DEFAULT_PORT;
155
+ const livePeer = detectPeer();
156
+ const livePidPort = parseHubPort(livePeer?.port);
157
+ if (
158
+ !isPidAlive(existing?.pid, killFn) ||
159
+ !Number.isFinite(existingPort) ||
160
+ existingPort <= 0
161
+ ) {
162
+ return null;
163
+ }
164
+ if (existingPort !== requestedPort) {
165
+ log.warn(
166
+ {
167
+ requestedPort,
168
+ existingPort,
169
+ pid: livePeer.pid,
170
+ livePidPort,
171
+ },
172
+ "hub.port_mismatch_not_reusing_live_pid",
173
+ );
96
174
  return null;
97
175
  }
98
- if (Number.isFinite(Number(port)) && existingPort !== Number(port)) return null;
99
176
  if (!(await checkHealth(existingPort))) return null;
100
177
 
101
178
  const info = readInfo() ?? existing;
102
- const infoHost = typeof info?.host === 'string' ? info.host : host;
179
+ const infoHost = typeof info?.host === "string" ? info.host : host;
103
180
  return {
104
181
  reused: true,
105
182
  external: true,
@@ -125,80 +202,330 @@ function checkRateLimit(ip) {
125
202
  }
126
203
 
127
204
  function isInitializeRequest(body) {
128
- if (body?.method === 'initialize') return true;
129
- if (Array.isArray(body)) return body.some((message) => message.method === 'initialize');
205
+ if (body?.method === "initialize") return true;
206
+ if (Array.isArray(body))
207
+ return body.some((message) => message.method === "initialize");
130
208
  return false;
131
209
  }
132
210
 
211
+ function parseTfxClientHeader(value) {
212
+ const raw = Array.isArray(value) ? value[0] : value;
213
+ const client = String(raw || "unknown").toLowerCase();
214
+ return ["codex", "claude", "gemini"].includes(client) ? client : "unknown";
215
+ }
216
+
133
217
  async function parseBody(req) {
134
218
  const chunks = [];
135
219
  let size = 0;
136
220
  for await (const chunk of req) {
137
221
  size += chunk.length;
138
222
  if (size > MAX_BODY_SIZE) {
139
- throw Object.assign(new Error('Body too large'), { statusCode: 413 });
223
+ throw Object.assign(new Error("Body too large"), { statusCode: 413 });
140
224
  }
141
225
  chunks.push(chunk);
142
226
  }
143
227
  return JSON.parse(Buffer.concat(chunks).toString());
144
228
  }
145
229
 
146
- const PID_DIR = join(homedir(), '.claude', 'cache', 'tfx-hub');
147
- const PID_FILE = join(PID_DIR, 'hub.pid');
148
- const TOKEN_FILE = join(homedir(), '.claude', '.tfx-hub-token');
230
+ const PID_DIR = join(homedir(), ".claude", "cache", "tfx-hub");
231
+ const PID_FILE = join(PID_DIR, "hub.pid");
232
+ const TOKEN_FILE = join(homedir(), ".claude", ".tfx-hub-token");
233
+
234
+ function readHubPidFile(
235
+ pidFilePath = PID_FILE,
236
+ { exists = existsSync, readFile = readFileSync } = {},
237
+ ) {
238
+ if (!exists(pidFilePath)) {
239
+ return { exists: false, info: null, error: null };
240
+ }
241
+
242
+ try {
243
+ return {
244
+ exists: true,
245
+ info: JSON.parse(readFile(pidFilePath, "utf8")),
246
+ error: null,
247
+ };
248
+ } catch (error) {
249
+ return { exists: true, info: null, error };
250
+ }
251
+ }
252
+
253
+ export function resolveHubPort(env = process.env, opts = {}) {
254
+ void opts;
255
+ const envPort = parseHubPort(env?.TFX_HUB_PORT);
256
+ if (envPort) return envPort;
257
+ return HUB_DEFAULT_PORT;
258
+ }
259
+
260
+ export function detectLivePeer(
261
+ pidFilePath = PID_FILE,
262
+ { exists = existsSync, readFile = readFileSync, killFn = process.kill } = {},
263
+ ) {
264
+ const state = readHubPidFile(pidFilePath, { exists, readFile });
265
+ const info = state.info ?? null;
266
+ const pid = Number(info?.pid);
267
+ const port = Number(info?.port);
268
+ const base = {
269
+ alive: false,
270
+ pid: Number.isFinite(pid) && pid > 0 ? pid : undefined,
271
+ port: Number.isFinite(port) && port > 0 ? port : undefined,
272
+ version: typeof info?.version === "string" ? info.version : undefined,
273
+ host: typeof info?.host === "string" ? info.host : undefined,
274
+ url: typeof info?.url === "string" ? info.url : undefined,
275
+ reason: "missing",
276
+ };
277
+
278
+ if (!state.exists) return base;
279
+ if (state.error) return { ...base, reason: "invalid_json" };
280
+ if (!Number.isFinite(pid) || pid <= 0) {
281
+ return { ...base, reason: "invalid_pid" };
282
+ }
283
+
284
+ if (isPidAlive(pid, killFn)) {
285
+ return {
286
+ alive: true,
287
+ pid,
288
+ port: base.port,
289
+ version: base.version,
290
+ host: base.host,
291
+ url: base.url,
292
+ reason: "alive",
293
+ };
294
+ }
295
+
296
+ return { ...base, reason: "dead" };
297
+ }
298
+
299
+ export function cleanStaleHubPid(
300
+ pidFilePath = PID_FILE,
301
+ {
302
+ exists = existsSync,
303
+ readFile = readFileSync,
304
+ unlink = unlinkSync,
305
+ killFn = process.kill,
306
+ } = {},
307
+ ) {
308
+ const peer = detectLivePeer(pidFilePath, { exists, readFile, killFn });
309
+ if (peer.alive) {
310
+ return { cleaned: false, reason: "alive", pid: peer.pid };
311
+ }
312
+ if (peer.reason === "missing") {
313
+ return { cleaned: false, reason: "missing" };
314
+ }
315
+
316
+ try {
317
+ unlink(pidFilePath);
318
+ return {
319
+ cleaned: true,
320
+ reason: peer.reason === "dead" ? "stale_pid" : peer.reason,
321
+ pid: peer.pid,
322
+ };
323
+ } catch (error) {
324
+ return {
325
+ cleaned: false,
326
+ reason: "unlink_failed",
327
+ pid: peer.pid,
328
+ error,
329
+ };
330
+ }
331
+ }
332
+
333
+ async function syncHubMcpSettingsIfAvailable({ hubUrl }) {
334
+ try {
335
+ const mod = await import(
336
+ new URL("../scripts/sync-hub-mcp-settings.mjs", import.meta.url)
337
+ );
338
+ if (typeof mod?.syncHubMcpSettings === "function") {
339
+ await mod.syncHubMcpSettings({ hubUrl });
340
+ } else {
341
+ hubLog.warn({ hubUrl }, "hub.mcp_sync_missing_export");
342
+ }
343
+ if (typeof mod?.syncCodexHubUrl === "function") {
344
+ await mod.syncCodexHubUrl({ hubUrl });
345
+ }
346
+ const allowProjectSync =
347
+ process.env.TFX_PROJECT_MCP_SYNC === "1" ||
348
+ !(
349
+ process.env.NODE_ENV === "test" ||
350
+ process.env.CI === "true" ||
351
+ process.env.TFX_TEST === "1"
352
+ );
353
+ if (allowProjectSync && typeof mod?.syncProjectMcpJson === "function") {
354
+ await mod.syncProjectMcpJson({ hubUrl, projectRoot: PROJECT_ROOT });
355
+ }
356
+ } catch (error) {
357
+ const message = error?.message || String(error);
358
+ if (error?.code === "ERR_MODULE_NOT_FOUND") {
359
+ hubLog.warn({ hubUrl, err: message }, "hub.mcp_sync_missing");
360
+ return;
361
+ }
362
+ hubLog.warn({ hubUrl, err: message }, "hub.mcp_sync_skipped");
363
+ }
364
+ }
365
+
366
+ function findListeningPidByPort(
367
+ port,
368
+ { platform = process.platform, execSync = execSyncHub } = {},
369
+ ) {
370
+ const targetPort = parseHubPort(port);
371
+ if (!targetPort) return null;
372
+
373
+ const execOptions = {
374
+ encoding: "utf8",
375
+ timeout: 5000,
376
+ stdio: ["ignore", "pipe", "ignore"],
377
+ windowsHide: true,
378
+ };
379
+
380
+ try {
381
+ if (platform === "win32") {
382
+ const output = execSync("netstat -ano -p tcp", execOptions);
383
+ for (const line of output.split(/\r?\n/)) {
384
+ const match = line
385
+ .trim()
386
+ .match(/^(TCP|UDP)\s+(\S+)\s+(\S+)\s+(\S+)\s+(\d+)$/i);
387
+ if (!match) continue;
388
+ const [, protocol, localAddress, , state, pidText] = match;
389
+ if (protocol.toUpperCase() !== "TCP") continue;
390
+ if (state.toUpperCase() !== "LISTENING") continue;
391
+ if (localAddress.split(":").pop() !== String(targetPort)) continue;
392
+ const pid = Number.parseInt(pidText, 10);
393
+ if (Number.isFinite(pid) && pid > 0) return pid;
394
+ }
395
+ return null;
396
+ }
397
+
398
+ try {
399
+ const output = execSync(
400
+ `lsof -nP -iTCP:${targetPort} -sTCP:LISTEN -t`,
401
+ execOptions,
402
+ );
403
+ const pid = Number.parseInt(output.trim().split(/\r?\n/)[0] ?? "", 10);
404
+ return Number.isFinite(pid) && pid > 0 ? pid : null;
405
+ } catch {
406
+ const output = execSync(`ss -ltnp 'sport = :${targetPort}'`, execOptions);
407
+ const match = output.match(/pid=(\d+)/);
408
+ const pid = Number.parseInt(match?.[1] ?? "", 10);
409
+ return Number.isFinite(pid) && pid > 0 ? pid : null;
410
+ }
411
+ } catch {
412
+ return null;
413
+ }
414
+ }
415
+
416
+ async function resolvePortInUse({
417
+ port,
418
+ host,
419
+ version,
420
+ pidFilePath = PID_FILE,
421
+ detectPeer = detectLivePeer,
422
+ cleanPid = cleanStaleHubPid,
423
+ lookupListeningPid = findListeningPidByPort,
424
+ } = {}) {
425
+ const peer = detectPeer(pidFilePath);
426
+ const ownerPid = lookupListeningPid(port);
427
+ const ownerHint = `pid=${ownerPid ?? "unknown"}`;
428
+ if (peer.alive) {
429
+ const peerPort = Number(peer.port);
430
+ const sameHub =
431
+ peerPort === Number(port) &&
432
+ typeof peer.version === "string" &&
433
+ peer.version === version;
434
+ if (sameHub) {
435
+ return {
436
+ action: "reuse",
437
+ peer,
438
+ url: peer.url ?? buildHubUrl(peer.host || host, peerPort),
439
+ };
440
+ }
441
+
442
+ const peerDetails = [
443
+ `hub.pid pid=${peer.pid ?? "unknown"}`,
444
+ `hub.pid port=${peerPort || "unknown"}`,
445
+ `hub.pid version=${peer.version ?? "unknown"}`,
446
+ ownerHint,
447
+ ].join(", ");
448
+ return {
449
+ action: "error",
450
+ message: `Hub 포트 ${port}이 이미 사용 중입니다 (${peerDetails}). fallback port는 허용되지 않습니다. \`tfx hub stop\` 후 재시도하세요.`,
451
+ };
452
+ }
453
+
454
+ const cleaned = cleanPid(pidFilePath);
455
+ if (cleaned.cleaned) {
456
+ return { action: "retry", cleaned, peer };
457
+ }
458
+
459
+ return {
460
+ action: "error",
461
+ message: `Hub 포트 ${port}이 이미 사용 중입니다 (${ownerHint}). fallback port는 허용되지 않습니다. \`tfx hub stop\` 후 재시도하세요. (PID file: ${pidFilePath})`,
462
+ };
463
+ }
149
464
 
150
465
  function isPublicPath(path) {
151
- return PUBLIC_PATHS.has(path)
152
- || path === '/dashboard'
153
- || path === '/api/qos-stats'
154
- || path.startsWith('/public/');
466
+ return (
467
+ PUBLIC_PATHS.has(path) ||
468
+ path === "/dashboard" ||
469
+ path === "/api/qos-stats" ||
470
+ path.startsWith("/public/")
471
+ );
155
472
  }
156
473
 
157
474
  function isAllowedOrigin(origin) {
158
475
  return origin && ALLOWED_ORIGIN_RE.test(origin);
159
476
  }
160
477
 
161
- function getRequestPath(url = '/') {
478
+ function getRequestPath(url = "/") {
162
479
  try {
163
- return new URL(url, 'http://127.0.0.1').pathname;
480
+ return new URL(url, "http://127.0.0.1").pathname;
164
481
  } catch {
165
- return String(url).replace(/\?.*/, '') || '/';
482
+ return String(url).replace(/\?.*/, "") || "/";
166
483
  }
167
484
  }
168
485
 
169
486
  function isLoopbackRemoteAddress(remoteAddress) {
170
- return typeof remoteAddress === 'string' && LOOPBACK_REMOTE_ADDRESSES.has(remoteAddress);
487
+ return (
488
+ typeof remoteAddress === "string" &&
489
+ LOOPBACK_REMOTE_ADDRESSES.has(remoteAddress)
490
+ );
171
491
  }
172
492
 
173
493
  function extractBearerToken(req) {
174
- const authHeader = typeof req.headers.authorization === 'string' ? req.headers.authorization : '';
175
- return authHeader.startsWith('Bearer ') ? authHeader.slice(7).trim() : '';
494
+ const authHeader =
495
+ typeof req.headers.authorization === "string"
496
+ ? req.headers.authorization
497
+ : "";
498
+ return authHeader.startsWith("Bearer ") ? authHeader.slice(7).trim() : "";
176
499
  }
177
500
 
178
501
  function writeJson(res, statusCode, body, headers = {}) {
179
502
  res.writeHead(statusCode, {
180
- 'Content-Type': 'application/json',
503
+ "Content-Type": "application/json",
181
504
  ...headers,
182
505
  });
183
506
  res.end(JSON.stringify(body));
184
507
  }
185
508
 
186
509
  function applyCorsHeaders(req, res) {
187
- const origin = typeof req.headers.origin === 'string' ? req.headers.origin : '';
510
+ const origin =
511
+ typeof req.headers.origin === "string" ? req.headers.origin : "";
188
512
  if (origin) {
189
- res.setHeader('Vary', 'Origin');
513
+ res.setHeader("Vary", "Origin");
190
514
  }
191
515
  if (!isAllowedOrigin(origin)) return false;
192
516
 
193
- res.setHeader('Access-Control-Allow-Origin', origin);
194
- res.setHeader('Access-Control-Allow-Methods', 'GET, POST, DELETE, OPTIONS');
195
- res.setHeader('Access-Control-Allow-Headers', 'Content-Type, Authorization, mcp-session-id, Last-Event-ID');
517
+ res.setHeader("Access-Control-Allow-Origin", origin);
518
+ res.setHeader("Access-Control-Allow-Methods", "GET, POST, DELETE, OPTIONS");
519
+ res.setHeader(
520
+ "Access-Control-Allow-Headers",
521
+ "Content-Type, Authorization, mcp-session-id, Last-Event-ID",
522
+ );
196
523
  return true;
197
524
  }
198
525
 
199
526
  function safeTokenCompare(a, b) {
200
- const ha = createHash('sha256').update(a).digest();
201
- const hb = createHash('sha256').update(b).digest();
527
+ const ha = createHash("sha256").update(a).digest();
528
+ const hb = createHash("sha256").update(b).digest();
202
529
  return timingSafeEqual(ha, hb);
203
530
  }
204
531
 
@@ -215,30 +542,83 @@ function isAuthorizedRequest(req, path, hubToken) {
215
542
  function resolveTeamStatusCode(result) {
216
543
  if (result?.ok) return 200;
217
544
  const code = result?.error?.code;
218
- if (code === 'TEAM_NOT_FOUND' || code === 'TASK_NOT_FOUND' || code === 'TASKS_DIR_NOT_FOUND') return 404;
219
- if (code === 'CLAIM_CONFLICT' || code === 'MTIME_CONFLICT') return 409;
220
- if (code === 'INVALID_TEAM_NAME' || code === 'INVALID_TASK_ID' || code === 'INVALID_TEXT' || code === 'INVALID_FROM' || code === 'INVALID_STATUS') return 400;
545
+ if (
546
+ code === "TEAM_NOT_FOUND" ||
547
+ code === "TASK_NOT_FOUND" ||
548
+ code === "TASKS_DIR_NOT_FOUND"
549
+ )
550
+ return 404;
551
+ if (code === "CLAIM_CONFLICT" || code === "MTIME_CONFLICT") return 409;
552
+ if (
553
+ code === "INVALID_TEAM_NAME" ||
554
+ code === "INVALID_TASK_ID" ||
555
+ code === "INVALID_TEXT" ||
556
+ code === "INVALID_FROM" ||
557
+ code === "INVALID_STATUS"
558
+ )
559
+ return 400;
221
560
  return 500;
222
561
  }
223
562
 
224
563
  function resolvePipelineStatusCode(result) {
225
564
  if (result?.ok) return 200;
226
- if (result?.error === 'pipeline_not_found') return 404;
227
- if (result?.error === 'hub_db_not_found') return 503;
565
+ if (result?.error === "pipeline_not_found") return 404;
566
+ if (result?.error === "hub_db_not_found") return 503;
567
+ return 400;
568
+ }
569
+
570
+ function resolveSendInputStatusCode(result) {
571
+ if (result?.ok) return 200;
572
+ const code = result?.error?.code;
573
+ if (code === "CONDUCTOR_REGISTRY_NOT_AVAILABLE") return 503;
574
+ if (code === "CONDUCTOR_SESSION_NOT_FOUND") return 404;
575
+ if (code === "SEND_INPUT_FAILED") return 409;
228
576
  return 400;
229
577
  }
230
578
 
579
+ function normalizeBridgePayload(payload) {
580
+ if (!payload || typeof payload !== "object" || Array.isArray(payload)) {
581
+ return {};
582
+ }
583
+ return payload;
584
+ }
585
+
586
+ function normalizeHandoffBody(body) {
587
+ const payload = normalizeBridgePayload(body?.payload);
588
+ return {
589
+ ...payload,
590
+ ...body,
591
+ from: body?.from,
592
+ to: body?.to,
593
+ payload,
594
+ };
595
+ }
596
+
597
+ function normalizePublishBody(body) {
598
+ const payload = normalizeBridgePayload(body?.payload);
599
+ const type = body?.type || body?.message_type || "event";
600
+ return {
601
+ ...payload,
602
+ ...body,
603
+ from: body?.from,
604
+ to: body?.to,
605
+ type,
606
+ message_type: type,
607
+ payload,
608
+ };
609
+ }
610
+
231
611
  function safeReadJsonFile(filePath) {
232
612
  try {
233
613
  if (!existsSync(filePath)) return null;
234
- return JSON.parse(readFileSync(filePath, 'utf8'));
614
+ return JSON.parse(readFileSync(filePath, "utf8"));
235
615
  } catch {
236
616
  return null;
237
617
  }
238
618
  }
239
619
 
240
620
  function parsePositiveInt(value, fallback) {
241
- const parsed = Number.parseInt(String(value ?? ''), 10);
621
+ const parsed = Number.parseInt(String(value ?? ""), 10);
242
622
  return Number.isFinite(parsed) && parsed > 0 ? parsed : fallback;
243
623
  }
244
624
 
@@ -246,7 +626,7 @@ function readRecentAimdEvents(now = Date.now()) {
246
626
  try {
247
627
  if (!existsSync(BATCH_EVENTS_PATH)) return [];
248
628
  const cutoff = now - AIMD_WINDOW_MS;
249
- return readFileSync(BATCH_EVENTS_PATH, 'utf8')
629
+ return readFileSync(BATCH_EVENTS_PATH, "utf8")
250
630
  .split(/\r?\n/)
251
631
  .filter(Boolean)
252
632
  .map((line) => {
@@ -270,9 +650,9 @@ function calculateAimdBatchSize(events) {
270
650
 
271
651
  for (const event of events) {
272
652
  const result = event?.result;
273
- if (result === 'success' || result === 'success_with_warnings') {
653
+ if (result === "success" || result === "success_with_warnings") {
274
654
  batchSize = Math.min(AIMD_MAX_BATCH_SIZE, batchSize + 1);
275
- } else if (result === 'failed' || result === 'timeout') {
655
+ } else if (result === "failed" || result === "timeout") {
276
656
  batchSize = Math.max(AIMD_MIN_BATCH_SIZE, batchSize * 0.5);
277
657
  }
278
658
  }
@@ -294,18 +674,107 @@ function getQosStatsPayload() {
294
674
  };
295
675
  }
296
676
 
677
+ export function sanitizeBrokerDiagnosticEvent(event = {}) {
678
+ const diagnostic = {
679
+ accountId: event.accountId ?? event.id,
680
+ provider: event.provider,
681
+ direction: event.direction,
682
+ reason: event.reason,
683
+ error: event.error,
684
+ };
685
+ return Object.fromEntries(
686
+ Object.entries(diagnostic).filter(([, value]) => value !== undefined),
687
+ );
688
+ }
689
+
690
+ export function attachBrokerDiagnostics(currentBroker, logger = hubLog) {
691
+ if (
692
+ !currentBroker ||
693
+ typeof currentBroker.on !== "function" ||
694
+ brokerDiagnosticsAttached.has(currentBroker)
695
+ ) {
696
+ return false;
697
+ }
698
+
699
+ currentBroker.on("securityViolation", (event) => {
700
+ logger.warn(
701
+ { brokerEvent: sanitizeBrokerDiagnosticEvent(event) },
702
+ "broker.security_violation",
703
+ );
704
+ });
705
+ currentBroker.on("authSyncError", (event) => {
706
+ logger.warn(
707
+ { brokerEvent: sanitizeBrokerDiagnosticEvent(event) },
708
+ "broker.auth_sync_error",
709
+ );
710
+ });
711
+
712
+ brokerDiagnosticsAttached.add(currentBroker);
713
+ return true;
714
+ }
715
+
716
+ function syncBrokerAuthCache(currentBroker, logger = hubLog) {
717
+ if (
718
+ !currentBroker?.snapshot ||
719
+ typeof currentBroker.syncAuthFromSource !== "function"
720
+ ) {
721
+ return [];
722
+ }
723
+
724
+ const authAccounts = currentBroker
725
+ .snapshot()
726
+ .filter(
727
+ (account) => account.provider === "codex" && account.mode === "auth",
728
+ );
729
+
730
+ return authAccounts.map((account) => {
731
+ try {
732
+ const result = currentBroker.syncAuthFromSource(account.id);
733
+ if (result?.copied) {
734
+ logger.info(
735
+ {
736
+ accountId: account.id,
737
+ reason: result.reason,
738
+ sourcePath: result.sourcePath,
739
+ cachePath: result.cachePath,
740
+ },
741
+ "broker.auth_sync_from_source",
742
+ );
743
+ }
744
+ return result;
745
+ } catch (error) {
746
+ logger.warn(
747
+ { accountId: account.id, err: error?.message || String(error) },
748
+ "broker.auth_sync_from_source_failed",
749
+ );
750
+ return {
751
+ ok: false,
752
+ accountId: account.id,
753
+ reason: error?.message || String(error),
754
+ };
755
+ }
756
+ });
757
+ }
758
+
759
+ function getBrokerPublicSnapshot(currentBroker = brokerInstance) {
760
+ if (!currentBroker || typeof currentBroker.publicSnapshot !== "function") {
761
+ return [];
762
+ }
763
+ return currentBroker.publicSnapshot();
764
+ }
765
+
297
766
  function resolvePublicFilePath(path) {
298
767
  let relativePath = null;
299
- if (path === '/dashboard') {
300
- relativePath = 'dashboard.html';
301
- } else if (path.startsWith('/public/')) {
302
- relativePath = path.slice('/public/'.length);
768
+ if (path === "/dashboard") {
769
+ relativePath = "dashboard.html";
770
+ } else if (path.startsWith("/public/")) {
771
+ relativePath = path.slice("/public/".length);
303
772
  }
304
773
 
305
774
  if (!relativePath) return null;
306
775
 
307
776
  try {
308
- relativePath = decodeURIComponent(relativePath).replace(/^[/\\]+/, '');
777
+ relativePath = decodeURIComponent(relativePath).replace(/^[/\\]+/, "");
309
778
  } catch {
310
779
  return null;
311
780
  }
@@ -324,21 +793,23 @@ function servePublicFile(res, path) {
324
793
 
325
794
  mkdirSync(PUBLIC_DIR, { recursive: true });
326
795
  if (!existsSync(filePath)) {
327
- hubLog.warn({ filePath }, 'static.not_found');
796
+ hubLog.warn({ filePath }, "static.not_found");
328
797
  res.writeHead(404);
329
- res.end('Not Found (static file missing)');
798
+ res.end("Not Found (static file missing)");
330
799
  return true;
331
800
  }
332
801
 
333
802
  try {
334
803
  const body = readFileSync(filePath);
335
804
  res.writeHead(200, {
336
- 'Content-Type': STATIC_CONTENT_TYPES[extname(filePath).toLowerCase()] || 'application/octet-stream',
805
+ "Content-Type":
806
+ STATIC_CONTENT_TYPES[extname(filePath).toLowerCase()] ||
807
+ "application/octet-stream",
337
808
  });
338
809
  res.end(body);
339
810
  } catch {
340
811
  res.writeHead(404);
341
- res.end('Not Found');
812
+ res.end("Not Found");
342
813
  }
343
814
  return true;
344
815
  }
@@ -355,16 +826,46 @@ function servePublicFile(res, path) {
355
826
  export async function startHub({
356
827
  port: portOpt,
357
828
  dbPath,
358
- host = '127.0.0.1',
829
+ host = "127.0.0.1",
359
830
  sessionId = process.pid,
360
831
  createDelegatorWorker = createDelegatorMcpWorker,
361
832
  } = {}) {
362
- const port = portOpt ?? parseInt(process.env.TFX_HUB_PORT || '27888', 10);
833
+ const resolvedPort = parseHubPort(portOpt);
834
+ const portSpecified = Number.isFinite(resolvedPort) && resolvedPort > 0;
835
+ const livePeer = detectLivePeer();
836
+ const livePidPort = parseHubPort(livePeer?.port);
837
+ const port = portSpecified
838
+ ? resolvedPort
839
+ : resolveHubPort(process.env, {
840
+ detectPeer: () => livePeer,
841
+ });
842
+
843
+ if (portSpecified && livePeer.alive && livePidPort && port !== livePidPort) {
844
+ hubLog.warn(
845
+ {
846
+ requestedPort: port,
847
+ livePid: livePeer.pid,
848
+ livePidPort,
849
+ },
850
+ "hub.port_override_pid_mismatch",
851
+ );
852
+ }
363
853
 
364
- const existingHub = await tryReuseExistingHub({ port, host });
854
+ const existingHub = await tryReuseExistingHub({
855
+ port,
856
+ portSpecified,
857
+ host,
858
+ detectPeer: () => livePeer,
859
+ });
365
860
  if (existingHub) return existingHub;
366
861
 
367
- const hubIdleTimeoutMs = parsePositiveInt(process.env.TFX_HUB_IDLE_TIMEOUT_MS, HUB_IDLE_TIMEOUT_DEFAULT_MS);
862
+ attachBrokerDiagnostics(brokerInstance);
863
+ syncBrokerAuthCache(brokerInstance);
864
+
865
+ const hubIdleTimeoutMs = parsePositiveInt(
866
+ process.env.TFX_HUB_IDLE_TIMEOUT_MS,
867
+ HUB_IDLE_TIMEOUT_DEFAULT_MS,
868
+ );
368
869
  const hubIdleSweepMs = parsePositiveInt(
369
870
  process.env.TFX_HUB_IDLE_SWEEP_MS,
370
871
  Math.min(HUB_IDLE_SWEEP_DEFAULT_MS, hubIdleTimeoutMs),
@@ -378,9 +879,9 @@ export async function startHub({
378
879
  // DB를 npm 패키지 밖에 저장하여 npm update 시 EBUSY 방지
379
880
  // 기존: PROJECT_ROOT/.tfx/state/state.db (패키지 내부 → 락 충돌)
380
881
  // 변경: ~/.claude/cache/tfx-hub/state.db (패키지 외부 → 안전)
381
- const hubCacheDir = join(homedir(), '.claude', 'cache', 'tfx-hub');
882
+ const hubCacheDir = join(homedir(), ".claude", "cache", "tfx-hub");
382
883
  mkdirSync(hubCacheDir, { recursive: true });
383
- dbPath = join(hubCacheDir, 'state.db');
884
+ dbPath = join(hubCacheDir, "state.db");
384
885
  }
385
886
 
386
887
  mkdirSync(PUBLIC_DIR, { recursive: true });
@@ -396,7 +897,12 @@ export async function startHub({
396
897
  lockHeld = false;
397
898
  };
398
899
 
399
- const lockedExistingHub = await tryReuseExistingHub({ port, host });
900
+ const lockedExistingHub = await tryReuseExistingHub({
901
+ port,
902
+ portSpecified,
903
+ host,
904
+ detectPeer: () => detectLivePeer(),
905
+ });
400
906
  if (lockedExistingHub) {
401
907
  releaseStartupLock();
402
908
  return lockedExistingHub;
@@ -404,10 +910,12 @@ export async function startHub({
404
910
 
405
911
  const HUB_TOKEN = process.env.TFX_HUB_TOKEN?.trim() || null;
406
912
  if (HUB_TOKEN) {
407
- mkdirSync(join(homedir(), '.claude'), { recursive: true });
913
+ mkdirSync(join(homedir(), ".claude"), { recursive: true });
408
914
  writeFileSync(TOKEN_FILE, HUB_TOKEN, { mode: 0o600 });
409
915
  } else {
410
- try { unlinkSync(TOKEN_FILE); } catch {}
916
+ try {
917
+ unlinkSync(TOKEN_FILE);
918
+ } catch {}
411
919
  }
412
920
 
413
921
  const store = await createStoreAdapter(dbPath);
@@ -415,9 +923,51 @@ export async function startHub({
415
923
  const fingerprintService = createAdaptiveFingerprintService({ store });
416
924
 
417
925
  // Neural Memory adaptive engine 초기화
418
- const adaptiveEngine = createAdaptiveEngine({ repoRoot: PROJECT_ROOT, fingerprintService });
926
+ const adaptiveEngine = createAdaptiveEngine({
927
+ repoRoot: PROJECT_ROOT,
928
+ fingerprintService,
929
+ });
419
930
  adaptiveEngine.startSession();
420
931
 
932
+ // safety-guard → reflexion 패널티 승격 + 적응형 규칙 유지보수
933
+ const projectSlug = PROJECT_ROOT.split(/[\\/]/).pop();
934
+ try {
935
+ const { promotePenalties } = await import("./promote-penalties.mjs");
936
+ const result = promotePenalties(store, { projectSlug });
937
+ if (result.promoted > 0) {
938
+ console.log(
939
+ `[reflexion] ${result.promoted} penalties promoted to adaptive rules`,
940
+ );
941
+ }
942
+ } catch {
943
+ /* promote-penalties 실패는 Hub 시작을 막지 않음 */
944
+ }
945
+
946
+ // stale adaptive_rules 정리 (30일 초과 + confidence 0.2 미만)
947
+ try {
948
+ const pruned = store.pruneStaleRules();
949
+ if (pruned > 0)
950
+ console.log(`[reflexion] ${pruned} stale adaptive rules pruned`);
951
+ } catch {
952
+ /* prune 실패 무시 */
953
+ }
954
+
955
+ // adaptive rule confidence decay (7일 이상 미관측 규칙 -0.1 감소)
956
+ try {
957
+ const { decayRules } = await import("@triflux/core/hub/reflexion.mjs");
958
+ const decay = decayRules(
959
+ store,
960
+ adaptiveEngine.sessionCount?.() || 1,
961
+ projectSlug,
962
+ );
963
+ if (decay.deleted.length > 0)
964
+ console.log(
965
+ `[reflexion] ${decay.deleted.length} low-confidence rules removed`,
966
+ );
967
+ } catch {
968
+ /* decay 실패 무시 */
969
+ }
970
+
421
971
  // Delegator MCP resident service 초기화
422
972
  const delegatorWorker = createDelegatorWorker({ cwd: PROJECT_ROOT });
423
973
  try {
@@ -428,518 +978,997 @@ export async function startHub({
428
978
  }
429
979
  const delegatorService = new DelegatorService({ worker: delegatorWorker });
430
980
 
431
- const pipe = createPipeServer({ router, store, sessionId, delegatorService });
432
- const assignCallbacks = createAssignCallbackServer({ store, sessionId });
981
+ // Synapse Layer 4: session registry + git preflight + swarm locks
982
+ const synapseEmitter = new EventEmitter();
983
+ synapseEmitter.setMaxListeners(50);
984
+ const synapseRegistry = createSynapseRegistry({
985
+ persistPath: join(CACHE_DIR, "tfx-hub", "synapse-sessions.json"),
986
+ emitter: synapseEmitter,
987
+ });
988
+ const swarmLocks = createSwarmLocks({
989
+ repoRoot: PROJECT_ROOT,
990
+ persistPath: join(CACHE_DIR, "tfx-hub", "swarm-locks.json"),
991
+ });
992
+ const gitPreflight = createGitPreflight({
993
+ registry: synapseRegistry,
994
+ locks: swarmLocks,
995
+ });
996
+
997
+ // Synapse Layer 5: emitter subscribers — bridge events to hub logging
998
+ synapseEmitter.on("synapse.session.started", ({ sessionId }) => {
999
+ hubLog.info({ sessionId }, "synapse.session.started");
1000
+ });
1001
+ synapseEmitter.on("synapse.session.heartbeat", ({ sessionId }) => {
1002
+ hubLog.debug({ sessionId }, "synapse.session.heartbeat");
1003
+ });
1004
+ synapseEmitter.on("synapse.session.stale", ({ sessionId }) => {
1005
+ hubLog.warn({ sessionId }, "synapse.session.stale");
1006
+ });
1007
+ synapseEmitter.on("synapse.session.removed", ({ sessionId }) => {
1008
+ hubLog.info({ sessionId }, "synapse.session.removed");
1009
+ });
1010
+
433
1011
  const hitl = createHitlManager(store, router);
1012
+ const pipe = createPipeServer({
1013
+ router,
1014
+ store,
1015
+ sessionId,
1016
+ delegatorService,
1017
+ hitlManager: hitl,
1018
+ onActivity: markRequestActivity,
1019
+ });
1020
+ const assignCallbacks = createAssignCallbackServer({ store, sessionId });
434
1021
  const tools = createTools(store, router, hitl, pipe);
435
1022
  const transports = new Map();
1023
+ const workerSpans = new Map();
1024
+
1025
+ async function closeMcpTransportSession(sid, session, reason) {
1026
+ if (!sid) return;
1027
+ if (!session) {
1028
+ transports.delete(sid);
1029
+ return;
1030
+ }
1031
+ if (session.closing) {
1032
+ transports.delete(sid);
1033
+ return;
1034
+ }
1035
+
1036
+ session.closing = true;
1037
+ transports.delete(sid);
1038
+
1039
+ try {
1040
+ await session.mcp.close();
1041
+ } catch (error) {
1042
+ hubLog.debug(
1043
+ { sid, reason, err: String(error?.message || error) },
1044
+ "hub.mcp_session_close_mcp_failed",
1045
+ );
1046
+ }
1047
+ try {
1048
+ await session.transport.close();
1049
+ } catch (error) {
1050
+ hubLog.debug(
1051
+ { sid, reason, err: String(error?.message || error) },
1052
+ "hub.mcp_session_close_transport_failed",
1053
+ );
1054
+ }
1055
+ }
1056
+
1057
+ function getTraceClient(clientRef) {
1058
+ return clientRef?.client || "unknown";
1059
+ }
436
1060
 
437
- function createMcpForSession() {
1061
+ function finishWorkerTrace(workerId, kind) {
1062
+ if (!workerId || typeof workerId !== "string") return;
1063
+ const span = workerSpans.get(workerId);
1064
+ if (!span) return;
1065
+ workerSpans.delete(workerId);
1066
+ recordWorker(
1067
+ workerId,
1068
+ "exited",
1069
+ span.command || kind,
1070
+ performance.now() - span.startedAt,
1071
+ );
1072
+ }
1073
+
1074
+ function createMcpForSession(clientRef = { client: "unknown" }) {
438
1075
  const mcp = new Server(
439
- { name: 'tfx-hub', version: '1.0.0' },
1076
+ { name: "tfx-hub", version: "1.0.0" },
440
1077
  { capabilities: { tools: {} } },
441
1078
  );
442
1079
 
443
- mcp.setRequestHandler(
444
- ListToolsRequestSchema,
445
- async () => ({
446
- tools: tools.map((tool) => ({
447
- name: tool.name,
448
- description: tool.description,
449
- inputSchema: tool.inputSchema,
450
- })),
451
- }),
452
- );
1080
+ mcp.setRequestHandler(ListToolsRequestSchema, async () => {
1081
+ const started = performance.now();
1082
+ try {
1083
+ return {
1084
+ tools: tools.map((tool) => ({
1085
+ name: tool.name,
1086
+ description: tool.description,
1087
+ inputSchema: tool.inputSchema,
1088
+ })),
1089
+ };
1090
+ } finally {
1091
+ recordRequest(
1092
+ "tools.list",
1093
+ performance.now() - started,
1094
+ getTraceClient(clientRef),
1095
+ "handler",
1096
+ );
1097
+ }
1098
+ });
453
1099
 
454
- mcp.setRequestHandler(
455
- CallToolRequestSchema,
456
- async (request) => {
457
- const { name, arguments: args } = request.params;
1100
+ mcp.setRequestHandler(CallToolRequestSchema, async (request) => {
1101
+ const { name, arguments: args } = request.params;
1102
+ const category = `tools.call.${name || "unknown"}`;
1103
+ const started = performance.now();
1104
+ try {
458
1105
  const tool = tools.find((candidate) => candidate.name === name);
459
1106
  if (!tool) {
460
1107
  return {
461
- content: [{ type: 'text', text: JSON.stringify({ ok: false, error: { code: 'UNKNOWN_TOOL', message: `도구 없음: ${name}` } }) }],
1108
+ content: [
1109
+ {
1110
+ type: "text",
1111
+ text: JSON.stringify({
1112
+ ok: false,
1113
+ error: {
1114
+ code: "UNKNOWN_TOOL",
1115
+ message: `도구 없음: ${name}`,
1116
+ },
1117
+ }),
1118
+ },
1119
+ ],
462
1120
  isError: true,
463
1121
  };
464
1122
  }
465
1123
  return tool.handler(args || {});
466
- },
467
- );
1124
+ } finally {
1125
+ recordRequest(
1126
+ category,
1127
+ performance.now() - started,
1128
+ getTraceClient(clientRef),
1129
+ "handler",
1130
+ );
1131
+ }
1132
+ });
468
1133
 
469
1134
  return mcp;
470
1135
  }
471
1136
 
472
- const httpServer = createHttpServer(wrapRequestHandler(async (req, res) => {
473
- markRequestActivity();
474
- res.setHeader('X-Content-Type-Options', 'nosniff');
475
- res.setHeader('X-Frame-Options', 'DENY');
476
- const path = getRequestPath(req.url);
477
- const corsAllowed = applyCorsHeaders(req, res);
478
-
479
- if (req.method === 'OPTIONS') {
480
- const localOnlyMode = !HUB_TOKEN;
481
- const isLoopbackRequest = isLoopbackRemoteAddress(req.socket.remoteAddress);
482
- res.writeHead(corsAllowed && (!localOnlyMode || isLoopbackRequest) ? 204 : 403);
483
- return res.end();
484
- }
1137
+ const httpServer = createHttpServer(
1138
+ wrapRequestHandler(async (req, res) => {
1139
+ markRequestActivity();
1140
+ res.setHeader("X-Content-Type-Options", "nosniff");
1141
+ res.setHeader("X-Frame-Options", "DENY");
1142
+ const path = getRequestPath(req.url);
1143
+ const corsAllowed = applyCorsHeaders(req, res);
1144
+
1145
+ if (req.method === "OPTIONS") {
1146
+ const localOnlyMode = !HUB_TOKEN;
1147
+ const isLoopbackRequest = isLoopbackRemoteAddress(
1148
+ req.socket.remoteAddress,
1149
+ );
1150
+ res.writeHead(
1151
+ corsAllowed && (!localOnlyMode || isLoopbackRequest) ? 204 : 403,
1152
+ );
1153
+ return res.end();
1154
+ }
485
1155
 
486
- const clientIp = req.socket.remoteAddress || 'unknown';
487
- if (!isLoopbackRemoteAddress(clientIp)) {
488
- const rateCheck = checkRateLimit(clientIp);
489
- if (!rateCheck.allowed) {
1156
+ const clientIp = req.socket.remoteAddress || "unknown";
1157
+ if (!isLoopbackRemoteAddress(clientIp)) {
1158
+ const rateCheck = checkRateLimit(clientIp);
1159
+ if (!rateCheck.allowed) {
1160
+ return writeJson(
1161
+ res,
1162
+ 429,
1163
+ { ok: false, error: "Too Many Requests" },
1164
+ { "Retry-After": String(rateCheck.retryAfterSec) },
1165
+ );
1166
+ }
1167
+ }
1168
+
1169
+ if (!isAuthorizedRequest(req, path, HUB_TOKEN)) {
1170
+ if (!HUB_TOKEN) {
1171
+ return writeJson(res, 403, {
1172
+ ok: false,
1173
+ error: "Forbidden: localhost only",
1174
+ });
1175
+ }
490
1176
  return writeJson(
491
1177
  res,
492
- 429,
493
- { ok: false, error: 'Too Many Requests' },
494
- { 'Retry-After': String(rateCheck.retryAfterSec) },
1178
+ 401,
1179
+ { ok: false, error: "Unauthorized" },
1180
+ { "WWW-Authenticate": 'Bearer realm="tfx-hub"' },
495
1181
  );
496
1182
  }
497
- }
498
1183
 
499
- if (!isAuthorizedRequest(req, path, HUB_TOKEN)) {
500
- if (!HUB_TOKEN) {
501
- return writeJson(res, 403, { ok: false, error: 'Forbidden: localhost only' });
1184
+ if (path === "/" || path === "/status") {
1185
+ const status = router.getStatus("hub").data;
1186
+ const includeMetrics =
1187
+ new URL(req.url, `http://${host}`).searchParams.get(
1188
+ "include_metrics",
1189
+ ) === "1";
1190
+ const body = {
1191
+ ...status,
1192
+ sessions: transports.size,
1193
+ pid: process.pid,
1194
+ port,
1195
+ auth_mode: HUB_TOKEN ? "token-required" : "localhost-only",
1196
+ idle_timeout_ms: hubIdleTimeoutMs,
1197
+ last_request_at: new Date(lastRequestAt).toISOString(),
1198
+ pipe_path: pipe.path,
1199
+ pipe: pipe.getStatus(),
1200
+ assign_callback_pipe_path: assignCallbacks.path,
1201
+ assign_callback_pipe: assignCallbacks.getStatus(),
1202
+ spawn_trace: {
1203
+ max_per_sec: spawnTrace.getMaxSpawnPerSec(),
1204
+ max_total_descendants: spawnTrace.MAX_TOTAL_DESCENDANTS,
1205
+ },
1206
+ version,
1207
+ };
1208
+ if (includeMetrics) {
1209
+ body.metrics = traceSnapshot();
1210
+ }
1211
+ return writeJson(res, 200, body);
502
1212
  }
503
- return writeJson(
504
- res,
505
- 401,
506
- { ok: false, error: 'Unauthorized' },
507
- { 'WWW-Authenticate': 'Bearer realm="tfx-hub"' },
508
- );
509
- }
510
1213
 
511
- if (path === '/' || path === '/status') {
512
- const status = router.getStatus('hub').data;
513
- return writeJson(res, 200, {
514
- ...status,
515
- sessions: transports.size,
516
- pid: process.pid,
517
- port,
518
- auth_mode: HUB_TOKEN ? 'token-required' : 'localhost-only',
519
- idle_timeout_ms: hubIdleTimeoutMs,
520
- last_request_at: new Date(lastRequestAt).toISOString(),
521
- pipe_path: pipe.path,
522
- pipe: pipe.getStatus(),
523
- assign_callback_pipe_path: assignCallbacks.path,
524
- assign_callback_pipe: assignCallbacks.getStatus(),
525
- });
526
- }
1214
+ if (path === "/health" || path === "/healthz") {
1215
+ const status = router.getStatus("hub").data;
1216
+ const healthy = status?.hub?.state === "healthy";
1217
+ return writeJson(res, healthy ? 200 : 503, {
1218
+ ok: healthy,
1219
+ version,
1220
+ platform: process.platform,
1221
+ uptime_s: Math.max(0, Math.floor((Date.now() - startedAtMs) / 1000)),
1222
+ node: process.version,
1223
+ sessions: transports.size,
1224
+ store: store.type || "sqlite",
1225
+ idle_timeout_ms: hubIdleTimeoutMs,
1226
+ idle_ms: Math.max(0, Date.now() - lastRequestAt),
1227
+ fingerprint: fingerprintService.getHealth(),
1228
+ });
1229
+ }
527
1230
 
528
- if (path === '/health' || path === '/healthz') {
529
- const status = router.getStatus('hub').data;
530
- const healthy = status?.hub?.state === 'healthy';
531
- return writeJson(res, healthy ? 200 : 503, {
532
- ok: healthy,
533
- version,
534
- platform: process.platform,
535
- uptime_s: Math.max(0, Math.floor((Date.now() - startedAtMs) / 1000)),
536
- node: process.version,
537
- sessions: transports.size,
538
- store: store.type || 'sqlite',
539
- idle_timeout_ms: hubIdleTimeoutMs,
540
- idle_ms: Math.max(0, Date.now() - lastRequestAt),
541
- fingerprint: fingerprintService.getHealth(),
542
- });
543
- }
1231
+ if (path === "/api/qos-stats" && req.method === "GET") {
1232
+ return writeJson(res, 200, getQosStatsPayload());
1233
+ }
544
1234
 
545
- if (path === '/api/qos-stats' && req.method === 'GET') {
546
- return writeJson(res, 200, getQosStatsPayload());
547
- }
1235
+ if (path === "/broker/snapshot" && req.method === "GET") {
1236
+ const snap = getBrokerPublicSnapshot(brokerInstance);
1237
+ return writeJson(res, 200, {
1238
+ ok: true,
1239
+ accounts: snap,
1240
+ ts: Date.now(),
1241
+ });
1242
+ }
548
1243
 
549
- if (path.startsWith('/bridge')) {
550
- const isBridgeStatusGet = path === '/bridge/status' && req.method === 'GET';
551
- if (req.method !== 'POST' && req.method !== 'DELETE' && !isBridgeStatusGet) {
552
- return writeJson(res, 405, { ok: false, error: 'Method Not Allowed' });
1244
+ if (path === "/broker/dashboard" && req.method === "GET") {
1245
+ res.writeHead(200, { "Content-Type": "text/html; charset=utf-8" });
1246
+ res.end(renderBrokerDashboard());
1247
+ return;
553
1248
  }
554
1249
 
555
- try {
556
- const body = req.method === 'POST' ? await parseBody(req) : {};
557
- const requestUrl = new URL(req.url || path, 'http://127.0.0.1');
558
-
559
- if (path === '/bridge/status' && req.method === 'GET') {
560
- const scope = requestUrl.searchParams.get('scope') || 'hub';
561
- const include_metrics = requestUrl.searchParams.get('include_metrics') !== '0';
562
- const agent_id = requestUrl.searchParams.get('agent_id') || undefined;
563
- const trace_id = requestUrl.searchParams.get('trace_id') || undefined;
564
- const result = await pipe.executeQuery('status', {
565
- scope,
566
- include_metrics,
567
- agent_id,
568
- trace_id,
1250
+ if (
1251
+ path === "/broker/quota-refresh" &&
1252
+ (req.method === "POST" || req.method === "GET")
1253
+ ) {
1254
+ try {
1255
+ const results = await refreshAllAccountQuotas();
1256
+ logQuotaRefreshFailures(hubLog, results);
1257
+ return writeJson(res, 200, { ok: true, results, ts: Date.now() });
1258
+ } catch (err) {
1259
+ hubLog.error(
1260
+ { err: String(err?.message || err) },
1261
+ "broker.quota_refresh_error",
1262
+ );
1263
+ return writeJson(res, 200, {
1264
+ ok: false,
1265
+ error: String(err?.message || err),
569
1266
  });
570
- return writeJson(res, 200, result);
571
1267
  }
1268
+ }
572
1269
 
573
- if (path === '/bridge/register' && req.method === 'POST') {
574
- const { agent_id, cli, timeout_sec = 600, topics = [], capabilities = [], metadata = {} } = body;
575
- if (!agent_id || !cli) {
576
- return writeJson(res, 400, { ok: false, error: 'agent_id, cli 필수' });
577
- }
578
-
579
- const heartbeat_ttl_ms = (timeout_sec + 120) * 1000;
580
- const result = await pipe.executeCommand('register', {
581
- agent_id,
582
- cli,
583
- capabilities,
584
- topics,
585
- heartbeat_ttl_ms,
586
- metadata,
587
- });
588
- return writeJson(res, 200, result);
1270
+ if (path === "/broker/reload" && req.method === "POST") {
1271
+ const result = reloadBroker();
1272
+ if (!result.ok) {
1273
+ return writeJson(res, 200, { ok: false, error: result.error });
589
1274
  }
1275
+ attachBrokerDiagnostics(result.broker);
1276
+ syncBrokerAuthCache(result.broker);
1277
+ const accounts = getBrokerPublicSnapshot(result.broker).length;
1278
+ return writeJson(res, 200, { ok: true, accounts });
1279
+ }
590
1280
 
591
- if (path === '/bridge/result' && req.method === 'POST') {
592
- const { agent_id, topic = 'task.result', payload = {}, trace_id, correlation_id } = body;
593
- if (!agent_id) {
594
- return writeJson(res, 400, { ok: false, error: 'agent_id 필수' });
595
- }
1281
+ if (path === "/spawn-trace/reload" && req.method === "POST") {
1282
+ return writeJson(res, 200, {
1283
+ ok: true,
1284
+ max_spawn_per_sec: spawnTrace.reload(),
1285
+ });
1286
+ }
596
1287
 
597
- const result = await pipe.executeCommand('result', {
598
- agent_id,
599
- topic,
600
- payload,
601
- trace_id,
602
- correlation_id,
603
- });
604
- return writeJson(res, 200, result);
605
- }
1288
+ // ── Synapse Layer 5: session registry + locks + preflight routes ──
1289
+ if (path === "/synapse/sessions" && req.method === "GET") {
1290
+ return writeJson(res, 200, {
1291
+ ok: true,
1292
+ ...synapseRegistry.snapshot(),
1293
+ ts: Date.now(),
1294
+ });
1295
+ }
606
1296
 
607
- if (path === '/bridge/control' && req.method === 'POST') {
608
- const {
609
- from_agent = 'lead',
610
- to_agent,
611
- command,
612
- reason = '',
613
- payload = {},
614
- trace_id,
615
- correlation_id,
616
- ttl_ms = 3600000,
617
- } = body;
618
-
619
- if (!to_agent || !command) {
620
- return writeJson(res, 400, { ok: false, error: 'to_agent, command 필수' });
1297
+ if (path === "/synapse/register" && req.method === "POST") {
1298
+ try {
1299
+ const body = await parseBody(req);
1300
+ const { sessionId } = body || {};
1301
+ const result = synapseRegistry.register(sessionId, body);
1302
+ if (!result?.ok) {
1303
+ throw new Error(result?.reason || "register failed");
621
1304
  }
622
-
623
- const result = await pipe.executeCommand('control', {
624
- from_agent,
625
- to_agent,
626
- command,
627
- reason,
628
- payload,
629
- ttl_ms,
630
- trace_id,
631
- correlation_id,
1305
+ return writeJson(res, 200, {
1306
+ ok: true,
1307
+ sessionId: result.sessionId || sessionId,
1308
+ });
1309
+ } catch (err) {
1310
+ return writeJson(res, 400, {
1311
+ ok: false,
1312
+ error: String(err?.message || err),
632
1313
  });
633
-
634
- return writeJson(res, 200, result);
635
1314
  }
1315
+ }
636
1316
 
637
- if (path === '/bridge/status' && req.method === 'POST') {
638
- const {
639
- scope = 'hub',
640
- agent_id,
641
- status,
642
- include_metrics = true,
643
- trace_id,
644
- } = body;
645
-
646
- if (agent_id && status) {
647
- const normalizedAgentId = String(agent_id || '').trim();
648
- const normalizedStatus = String(status || '').trim().toLowerCase();
649
- if (!normalizedAgentId || !normalizedStatus) {
650
- return writeJson(res, 400, { ok: false, error: 'agent_id, status 필수' });
651
- }
652
- const statusForStore = new Set(['online', 'stale', 'offline']).has(normalizedStatus)
653
- ? normalizedStatus
654
- : 'online';
655
- router.updateAgentStatus(normalizedAgentId, statusForStore);
656
- const snapshot = await pipe.executeQuery('status', {
657
- scope: 'agent',
658
- agent_id: normalizedAgentId,
659
- include_metrics: false,
660
- });
661
- return writeJson(res, 200, {
662
- ok: true,
663
- data: {
664
- agent_id: normalizedAgentId,
665
- status: statusForStore,
666
- reported_status: normalizedStatus,
667
- reported_at_ms: Date.now(),
668
- snapshot: snapshot?.data?.agent || null,
669
- },
670
- });
1317
+ if (path === "/synapse/heartbeat" && req.method === "POST") {
1318
+ try {
1319
+ const body = await parseBody(req);
1320
+ const { sessionId, partial } = body || {};
1321
+ const ok = synapseRegistry.heartbeat(sessionId, partial);
1322
+ if (!ok) {
1323
+ throw new Error("heartbeat failed");
671
1324
  }
672
-
673
- const result = await pipe.executeQuery('status', {
674
- scope,
675
- agent_id,
676
- include_metrics,
677
- trace_id,
1325
+ return writeJson(res, 200, { ok: true });
1326
+ } catch (err) {
1327
+ return writeJson(res, 400, {
1328
+ ok: false,
1329
+ error: String(err?.message || err),
678
1330
  });
679
- return writeJson(res, 200, result);
680
1331
  }
1332
+ }
681
1333
 
682
- if (path === '/bridge/assign/async' && req.method === 'POST') {
683
- const {
684
- supervisor_agent,
685
- worker_agent,
686
- task,
687
- topic = 'assign.job',
688
- payload = {},
689
- priority = 5,
690
- ttl_ms = 600000,
691
- timeout_ms = 600000,
692
- max_retries = 0,
693
- trace_id,
694
- correlation_id,
695
- } = body;
696
-
697
- if (!supervisor_agent || !worker_agent || !task) {
698
- return writeJson(res, 400, { ok: false, error: 'supervisor_agent, worker_agent, task 필수' });
1334
+ if (path === "/synapse/unregister" && req.method === "POST") {
1335
+ try {
1336
+ const body = await parseBody(req);
1337
+ const { sessionId } = body || {};
1338
+ const ok = synapseRegistry.unregister(sessionId);
1339
+ if (!ok) {
1340
+ throw new Error("unregister failed");
699
1341
  }
700
-
701
- const result = await pipe.executeCommand('assign', {
702
- supervisor_agent,
703
- worker_agent,
704
- task,
705
- topic,
706
- payload,
707
- priority,
708
- ttl_ms,
709
- timeout_ms,
710
- max_retries,
711
- trace_id,
712
- correlation_id,
1342
+ return writeJson(res, 200, { ok: true });
1343
+ } catch (err) {
1344
+ return writeJson(res, 400, {
1345
+ ok: false,
1346
+ error: String(err?.message || err),
713
1347
  });
714
- return writeJson(res, result.ok ? 200 : 400, result);
715
1348
  }
1349
+ }
716
1350
 
717
- if (path === '/bridge/assign/result' && req.method === 'POST') {
718
- const {
719
- job_id,
720
- worker_agent,
721
- status,
722
- attempt,
723
- result: assignResult,
724
- error: assignError,
725
- payload = {},
726
- metadata = {},
727
- } = body;
728
-
729
- if (!job_id || !status) {
730
- return writeJson(res, 400, { ok: false, error: 'job_id, status 필수' });
731
- }
1351
+ if (path === "/synapse/locks" && req.method === "GET") {
1352
+ return writeJson(res, 200, {
1353
+ ok: true,
1354
+ locks: swarmLocks.snapshot(),
1355
+ ts: Date.now(),
1356
+ });
1357
+ }
732
1358
 
733
- const result = await pipe.executeCommand('assign_result', {
734
- job_id,
735
- worker_agent,
736
- status,
737
- attempt,
738
- result: assignResult,
739
- error: assignError,
740
- payload,
741
- metadata,
1359
+ if (path === "/synapse/preflight" && req.method === "POST") {
1360
+ try {
1361
+ const body = await parseBody(req);
1362
+ const { op, args = {}, sessionContext = {} } = body;
1363
+ if (!op || typeof op !== "string") {
1364
+ return writeJson(res, 400, { ok: false, error: "op 필수" });
1365
+ }
1366
+ if (!SYNAPSE_VALID_OPS.has(op)) {
1367
+ return writeJson(res, 400, {
1368
+ ok: false,
1369
+ error: `invalid op: ${op}`,
1370
+ });
1371
+ }
1372
+ const result = gitPreflight.check(op, args, sessionContext);
1373
+ return writeJson(res, 200, { ok: true, ...result });
1374
+ } catch (err) {
1375
+ return writeJson(res, 400, {
1376
+ ok: false,
1377
+ error: String(err?.message || err),
742
1378
  });
743
- return writeJson(res, result.ok ? 200 : 409, result);
744
1379
  }
1380
+ }
745
1381
 
746
- if (path === '/bridge/assign/status' && req.method === 'POST') {
747
- const result = await pipe.executeQuery('assign_status', body);
748
- const statusCode = result.ok ? 200 : (result.error?.code === 'ASSIGN_NOT_FOUND' ? 404 : 400);
749
- return writeJson(res, statusCode, result);
1382
+ if (path.startsWith("/bridge")) {
1383
+ const isBridgeStatusGet =
1384
+ path === "/bridge/status" && req.method === "GET";
1385
+ const isBridgeHitlPendingGet =
1386
+ path === "/bridge/hitl/pending" && req.method === "GET";
1387
+ if (
1388
+ req.method !== "POST" &&
1389
+ req.method !== "DELETE" &&
1390
+ !isBridgeStatusGet &&
1391
+ !isBridgeHitlPendingGet
1392
+ ) {
1393
+ return writeJson(res, 405, {
1394
+ ok: false,
1395
+ error: "Method Not Allowed",
1396
+ });
750
1397
  }
751
1398
 
752
- if (path === '/bridge/assign/retry' && req.method === 'POST') {
753
- const { job_id, reason, requested_by } = body;
754
- if (!job_id) {
755
- return writeJson(res, 400, { ok: false, error: 'job_id 필수' });
1399
+ try {
1400
+ const body = req.method === "POST" ? await parseBody(req) : {};
1401
+ const requestUrl = new URL(req.url || path, "http://127.0.0.1");
1402
+
1403
+ if (path === "/bridge/status" && req.method === "GET") {
1404
+ const scope = requestUrl.searchParams.get("scope") || "hub";
1405
+ const include_metrics =
1406
+ requestUrl.searchParams.get("include_metrics") !== "0";
1407
+ const agent_id =
1408
+ requestUrl.searchParams.get("agent_id") || undefined;
1409
+ const trace_id =
1410
+ requestUrl.searchParams.get("trace_id") || undefined;
1411
+ const result = await pipe.executeQuery("status", {
1412
+ scope,
1413
+ include_metrics,
1414
+ agent_id,
1415
+ trace_id,
1416
+ });
1417
+ return writeJson(res, 200, result);
756
1418
  }
757
1419
 
758
- const result = await pipe.executeCommand('assign_retry', {
759
- job_id,
760
- reason,
761
- requested_by,
762
- });
763
- const statusCode = result.ok ? 200
764
- : result.error?.code === 'ASSIGN_NOT_FOUND' ? 404
765
- : result.error?.code === 'ASSIGN_RETRY_EXHAUSTED' ? 409
766
- : 400;
767
- return writeJson(res, statusCode, result);
768
- }
1420
+ if (path === "/bridge/hitl/pending" && req.method === "GET") {
1421
+ const result = { ok: true, data: hitl.getPendingRequests() };
1422
+ return writeJson(res, result.ok ? 200 : 400, result);
1423
+ }
1424
+
1425
+ if (path === "/bridge/register" && req.method === "POST") {
1426
+ const {
1427
+ agent_id,
1428
+ cli,
1429
+ timeout_sec = 600,
1430
+ topics = [],
1431
+ capabilities = [],
1432
+ metadata = {},
1433
+ } = body;
1434
+ if (!agent_id || !cli) {
1435
+ return writeJson(res, 400, {
1436
+ ok: false,
1437
+ error: "agent_id, cli 필수",
1438
+ });
1439
+ }
769
1440
 
770
- if (req.method === 'POST') {
771
- let teamResult = null;
772
- if (path === '/bridge/team/info' || path === '/bridge/team-info') {
773
- teamResult = await pipe.executeQuery('team_info', body);
774
- } else if (path === '/bridge/team/task-list' || path === '/bridge/team-task-list') {
775
- teamResult = await pipe.executeQuery('team_task_list', body);
776
- } else if (path === '/bridge/team/task-update' || path === '/bridge/team-task-update') {
777
- teamResult = await pipe.executeCommand('team_task_update', body);
778
- } else if (path === '/bridge/team/send-message' || path === '/bridge/team-send-message') {
779
- teamResult = await pipe.executeCommand('team_send_message', body);
1441
+ const command =
1442
+ body.command ||
1443
+ metadata?.command ||
1444
+ metadata?.cmd ||
1445
+ metadata?.task ||
1446
+ cli;
1447
+ const workerStartedAt = performance.now();
1448
+ const heartbeat_ttl_ms = (timeout_sec + 120) * 1000;
1449
+ const result = await pipe.executeCommand("register", {
1450
+ agent_id,
1451
+ cli,
1452
+ capabilities,
1453
+ topics,
1454
+ heartbeat_ttl_ms,
1455
+ metadata,
1456
+ });
1457
+ if (result.ok) {
1458
+ workerSpans.set(agent_id, {
1459
+ command,
1460
+ startedAt: workerStartedAt,
1461
+ });
1462
+ recordWorker(agent_id, "spawned", command, 0);
1463
+ }
1464
+ return writeJson(res, 200, result);
780
1465
  }
781
1466
 
782
- if (teamResult) {
783
- return writeJson(res, resolveTeamStatusCode(teamResult), teamResult);
1467
+ if (path === "/bridge/result" && req.method === "POST") {
1468
+ const {
1469
+ agent_id,
1470
+ topic = "task.result",
1471
+ payload = {},
1472
+ trace_id,
1473
+ correlation_id,
1474
+ } = body;
1475
+ if (!agent_id) {
1476
+ return writeJson(res, 400, { ok: false, error: "agent_id 필수" });
1477
+ }
1478
+
1479
+ const result = await pipe.executeCommand("result", {
1480
+ agent_id,
1481
+ topic,
1482
+ payload,
1483
+ trace_id,
1484
+ correlation_id,
1485
+ });
1486
+ return writeJson(res, 200, result);
784
1487
  }
785
1488
 
786
- if (path.startsWith('/bridge/team')) {
787
- return writeJson(res, 404, { ok: false, error: `Unknown team endpoint: ${path}` });
1489
+ if (path === "/bridge/control" && req.method === "POST") {
1490
+ const {
1491
+ from_agent = "lead",
1492
+ to_agent,
1493
+ command,
1494
+ reason = "",
1495
+ payload = {},
1496
+ trace_id,
1497
+ correlation_id,
1498
+ ttl_ms = 3600000,
1499
+ } = body;
1500
+
1501
+ if (!to_agent || !command) {
1502
+ return writeJson(res, 400, {
1503
+ ok: false,
1504
+ error: "to_agent, command 필수",
1505
+ });
1506
+ }
1507
+
1508
+ const result = await pipe.executeCommand("control", {
1509
+ from_agent,
1510
+ to_agent,
1511
+ command,
1512
+ reason,
1513
+ payload,
1514
+ ttl_ms,
1515
+ trace_id,
1516
+ correlation_id,
1517
+ });
1518
+
1519
+ return writeJson(res, 200, result);
788
1520
  }
789
1521
 
790
- // ── 파이프라인 엔드포인트 ──
791
- if (path === '/bridge/pipeline/state' && req.method === 'POST') {
792
- const result = await pipe.executeQuery('pipeline_state', body);
793
- return writeJson(res, resolvePipelineStatusCode(result), result);
1522
+ if (path === "/bridge/handoff" && req.method === "POST") {
1523
+ const result = router.handleHandoff(normalizeHandoffBody(body));
1524
+ return writeJson(res, result.ok ? 200 : 400, result);
794
1525
  }
795
1526
 
796
- if (path === '/bridge/pipeline/advance' && req.method === 'POST') {
797
- const result = await pipe.executeCommand('pipeline_advance', body);
798
- return writeJson(res, resolvePipelineStatusCode(result), result);
1527
+ if (path === "/bridge/publish" && req.method === "POST") {
1528
+ const normalized = normalizePublishBody(body);
1529
+ const result = router.handlePublish(normalized);
1530
+ if (result.ok) {
1531
+ finishWorkerTrace(
1532
+ normalized.from ||
1533
+ normalized.agent_id ||
1534
+ normalized.worker_agent ||
1535
+ normalized.from_agent,
1536
+ "publish",
1537
+ );
1538
+ }
1539
+ return writeJson(res, result.ok ? 200 : 400, result);
799
1540
  }
800
1541
 
801
- if (path === '/bridge/pipeline/init' && req.method === 'POST') {
802
- const result = await pipe.executeCommand('pipeline_init', body);
803
- return writeJson(res, resolvePipelineStatusCode(result), result);
1542
+ if (path === "/bridge/send-input" && req.method === "POST") {
1543
+ const { session_id, text } = body;
1544
+ if (!session_id || typeof text !== "string" || text.length === 0) {
1545
+ return writeJson(res, 400, {
1546
+ ok: false,
1547
+ error: "session_id, text 필수",
1548
+ });
1549
+ }
1550
+
1551
+ const result = await pipe.executeCommand("send_input", {
1552
+ session_id,
1553
+ text,
1554
+ });
1555
+ return writeJson(res, resolveSendInputStatusCode(result), result);
804
1556
  }
805
1557
 
806
- if (path === '/bridge/pipeline/list' && req.method === 'POST') {
807
- const result = await pipe.executeQuery('pipeline_list', body);
808
- return writeJson(res, resolvePipelineStatusCode(result), result);
1558
+ if (path === "/bridge/status" && req.method === "POST") {
1559
+ const {
1560
+ scope = "hub",
1561
+ agent_id,
1562
+ status,
1563
+ include_metrics = true,
1564
+ trace_id,
1565
+ } = body;
1566
+
1567
+ if (agent_id && status) {
1568
+ const normalizedAgentId = String(agent_id || "").trim();
1569
+ const normalizedStatus = String(status || "")
1570
+ .trim()
1571
+ .toLowerCase();
1572
+ if (!normalizedAgentId || !normalizedStatus) {
1573
+ return writeJson(res, 400, {
1574
+ ok: false,
1575
+ error: "agent_id, status 필수",
1576
+ });
1577
+ }
1578
+ const statusForStore = new Set([
1579
+ "online",
1580
+ "stale",
1581
+ "offline",
1582
+ ]).has(normalizedStatus)
1583
+ ? normalizedStatus
1584
+ : "online";
1585
+ router.updateAgentStatus(normalizedAgentId, statusForStore);
1586
+ const snapshot = await pipe.executeQuery("status", {
1587
+ scope: "agent",
1588
+ agent_id: normalizedAgentId,
1589
+ include_metrics: false,
1590
+ });
1591
+ return writeJson(res, 200, {
1592
+ ok: true,
1593
+ data: {
1594
+ agent_id: normalizedAgentId,
1595
+ status: statusForStore,
1596
+ reported_status: normalizedStatus,
1597
+ reported_at_ms: Date.now(),
1598
+ snapshot: snapshot?.data?.agent || null,
1599
+ },
1600
+ });
1601
+ }
1602
+
1603
+ const result = await pipe.executeQuery("status", {
1604
+ scope,
1605
+ agent_id,
1606
+ include_metrics,
1607
+ trace_id,
1608
+ });
1609
+ return writeJson(res, 200, result);
809
1610
  }
810
1611
 
811
- // ── Delegator 엔드포인트 ──
812
- if (path === '/bridge/delegator/delegate' && req.method === 'POST') {
813
- const result = await pipe.executeCommand('delegator_delegate', body);
1612
+ if (path === "/bridge/assign/async" && req.method === "POST") {
1613
+ const {
1614
+ supervisor_agent,
1615
+ worker_agent,
1616
+ task,
1617
+ topic = "assign.job",
1618
+ payload = {},
1619
+ priority = 5,
1620
+ ttl_ms = 600000,
1621
+ timeout_ms = 600000,
1622
+ max_retries = 0,
1623
+ trace_id,
1624
+ correlation_id,
1625
+ } = body;
1626
+
1627
+ if (!supervisor_agent || !worker_agent || !task) {
1628
+ return writeJson(res, 400, {
1629
+ ok: false,
1630
+ error: "supervisor_agent, worker_agent, task 필수",
1631
+ });
1632
+ }
1633
+
1634
+ const result = await pipe.executeCommand("assign", {
1635
+ supervisor_agent,
1636
+ worker_agent,
1637
+ task,
1638
+ topic,
1639
+ payload,
1640
+ priority,
1641
+ ttl_ms,
1642
+ timeout_ms,
1643
+ max_retries,
1644
+ trace_id,
1645
+ correlation_id,
1646
+ });
814
1647
  return writeJson(res, result.ok ? 200 : 400, result);
815
1648
  }
816
1649
 
817
- if (path === '/bridge/delegator/reply' && req.method === 'POST') {
818
- const result = await pipe.executeCommand('delegator_reply', body);
819
- return writeJson(res, result.ok ? 200 : 400, result);
1650
+ if (path === "/bridge/assign/result" && req.method === "POST") {
1651
+ const {
1652
+ job_id,
1653
+ worker_agent,
1654
+ status,
1655
+ attempt,
1656
+ result: assignResult,
1657
+ error: assignError,
1658
+ payload = {},
1659
+ metadata = {},
1660
+ } = body;
1661
+
1662
+ if (!job_id || !status) {
1663
+ return writeJson(res, 400, {
1664
+ ok: false,
1665
+ error: "job_id, status 필수",
1666
+ });
1667
+ }
1668
+
1669
+ const result = await pipe.executeCommand("assign_result", {
1670
+ job_id,
1671
+ worker_agent,
1672
+ status,
1673
+ attempt,
1674
+ result: assignResult,
1675
+ error: assignError,
1676
+ payload,
1677
+ metadata,
1678
+ });
1679
+ return writeJson(res, result.ok ? 200 : 409, result);
820
1680
  }
821
1681
 
822
- if (path === '/bridge/delegator/status' && req.method === 'POST') {
823
- const result = await pipe.executeQuery('delegator_status', body);
824
- return writeJson(res, result.ok ? 200 : 400, result);
1682
+ if (path === "/bridge/assign/status" && req.method === "POST") {
1683
+ const result = await pipe.executeQuery("assign_status", body);
1684
+ const statusCode = result.ok
1685
+ ? 200
1686
+ : result.error?.code === "ASSIGN_NOT_FOUND"
1687
+ ? 404
1688
+ : 400;
1689
+ return writeJson(res, statusCode, result);
825
1690
  }
826
- }
827
1691
 
828
- if (path === '/bridge/context' && req.method === 'POST') {
829
- const { agent_id, topics, max_messages = 10, auto_ack = true } = body;
830
- if (!agent_id) {
831
- return writeJson(res, 400, { ok: false, error: 'agent_id 필수' });
1692
+ if (path === "/bridge/assign/retry" && req.method === "POST") {
1693
+ const { job_id, reason, requested_by } = body;
1694
+ if (!job_id) {
1695
+ return writeJson(res, 400, { ok: false, error: "job_id 필수" });
1696
+ }
1697
+
1698
+ const result = await pipe.executeCommand("assign_retry", {
1699
+ job_id,
1700
+ reason,
1701
+ requested_by,
1702
+ });
1703
+ const statusCode = result.ok
1704
+ ? 200
1705
+ : result.error?.code === "ASSIGN_NOT_FOUND"
1706
+ ? 404
1707
+ : result.error?.code === "ASSIGN_RETRY_EXHAUSTED"
1708
+ ? 409
1709
+ : 400;
1710
+ return writeJson(res, statusCode, result);
832
1711
  }
833
1712
 
834
- const result = await pipe.executeQuery('drain', {
835
- agent_id,
836
- topics,
837
- max_messages,
838
- auto_ack,
839
- });
840
- return writeJson(res, 200, result);
841
- }
1713
+ if (req.method === "POST") {
1714
+ let teamResult = null;
1715
+ if (path === "/bridge/team/info" || path === "/bridge/team-info") {
1716
+ teamResult = await pipe.executeQuery("team_info", body);
1717
+ } else if (
1718
+ path === "/bridge/team/task-list" ||
1719
+ path === "/bridge/team-task-list"
1720
+ ) {
1721
+ teamResult = await pipe.executeQuery("team_task_list", body);
1722
+ } else if (
1723
+ path === "/bridge/team/task-update" ||
1724
+ path === "/bridge/team-task-update"
1725
+ ) {
1726
+ teamResult = await pipe.executeCommand("team_task_update", body);
1727
+ } else if (
1728
+ path === "/bridge/team/send-message" ||
1729
+ path === "/bridge/team-send-message"
1730
+ ) {
1731
+ teamResult = await pipe.executeCommand("team_send_message", body);
1732
+ }
1733
+
1734
+ if (teamResult) {
1735
+ return writeJson(
1736
+ res,
1737
+ resolveTeamStatusCode(teamResult),
1738
+ teamResult,
1739
+ );
1740
+ }
842
1741
 
843
- if (path === '/bridge/deregister' && req.method === 'POST') {
844
- const { agent_id } = body;
845
- if (!agent_id) {
846
- return writeJson(res, 400, { ok: false, error: 'agent_id 필수' });
1742
+ if (path.startsWith("/bridge/team")) {
1743
+ return writeJson(res, 404, {
1744
+ ok: false,
1745
+ error: `Unknown team endpoint: ${path}`,
1746
+ });
1747
+ }
1748
+
1749
+ // ── 파이프라인 엔드포인트 ──
1750
+ if (path === "/bridge/pipeline/state" && req.method === "POST") {
1751
+ const result = await pipe.executeQuery("pipeline_state", body);
1752
+ return writeJson(res, resolvePipelineStatusCode(result), result);
1753
+ }
1754
+
1755
+ if (path === "/bridge/pipeline/advance" && req.method === "POST") {
1756
+ const result = await pipe.executeCommand(
1757
+ "pipeline_advance",
1758
+ body,
1759
+ );
1760
+ return writeJson(res, resolvePipelineStatusCode(result), result);
1761
+ }
1762
+
1763
+ if (path === "/bridge/pipeline/init" && req.method === "POST") {
1764
+ const result = await pipe.executeCommand("pipeline_init", body);
1765
+ return writeJson(res, resolvePipelineStatusCode(result), result);
1766
+ }
1767
+
1768
+ if (path === "/bridge/pipeline/list" && req.method === "POST") {
1769
+ const result = await pipe.executeQuery("pipeline_list", body);
1770
+ return writeJson(res, resolvePipelineStatusCode(result), result);
1771
+ }
1772
+
1773
+ // ── Delegator 엔드포인트 ──
1774
+ if (
1775
+ path === "/bridge/delegator/delegate" &&
1776
+ req.method === "POST"
1777
+ ) {
1778
+ const result = await pipe.executeCommand(
1779
+ "delegator_delegate",
1780
+ body,
1781
+ );
1782
+ return writeJson(res, result.ok ? 200 : 400, result);
1783
+ }
1784
+
1785
+ if (path === "/bridge/delegator/reply" && req.method === "POST") {
1786
+ const result = await pipe.executeCommand("delegator_reply", body);
1787
+ return writeJson(res, result.ok ? 200 : 400, result);
1788
+ }
1789
+
1790
+ if (path === "/bridge/delegator/status" && req.method === "POST") {
1791
+ const result = await pipe.executeQuery("delegator_status", body);
1792
+ return writeJson(res, result.ok ? 200 : 400, result);
1793
+ }
1794
+
1795
+ if (path === "/bridge/hitl/request" && req.method === "POST") {
1796
+ const result = hitl.requestHumanInput(body);
1797
+ return writeJson(res, result.ok ? 200 : 400, result);
1798
+ }
1799
+
1800
+ if (path === "/bridge/hitl/submit" && req.method === "POST") {
1801
+ const result = hitl.submitHumanInput(body);
1802
+ return writeJson(res, result.ok ? 200 : 400, result);
1803
+ }
847
1804
  }
848
- const result = await pipe.executeCommand('deregister', { agent_id });
849
- return writeJson(res, 200, result);
850
- }
851
1805
 
852
- return writeJson(res, 404, { ok: false, error: 'Unknown bridge endpoint' });
853
- } catch (error) {
854
- if (!res.headersSent) {
855
- console.error('[tfx-hub] bridge error:', error);
856
- writeJson(res, 500, { ok: false, error: 'Internal server error' });
1806
+ if (path === "/bridge/context" && req.method === "POST") {
1807
+ const {
1808
+ agent_id,
1809
+ topics,
1810
+ max_messages = 10,
1811
+ auto_ack = true,
1812
+ } = body;
1813
+ if (!agent_id) {
1814
+ return writeJson(res, 400, { ok: false, error: "agent_id 필수" });
1815
+ }
1816
+
1817
+ const result = await pipe.executeQuery("drain", {
1818
+ agent_id,
1819
+ topics,
1820
+ max_messages,
1821
+ auto_ack,
1822
+ });
1823
+ return writeJson(res, 200, result);
1824
+ }
1825
+
1826
+ if (path === "/bridge/deregister" && req.method === "POST") {
1827
+ const { agent_id } = body;
1828
+ if (!agent_id) {
1829
+ return writeJson(res, 400, { ok: false, error: "agent_id 필수" });
1830
+ }
1831
+ const result = await pipe.executeCommand("deregister", {
1832
+ agent_id,
1833
+ });
1834
+ if (result.ok) {
1835
+ finishWorkerTrace(agent_id, "deregister");
1836
+ }
1837
+ return writeJson(res, 200, result);
1838
+ }
1839
+
1840
+ return writeJson(res, 404, {
1841
+ ok: false,
1842
+ error: "Unknown bridge endpoint",
1843
+ });
1844
+ } catch (error) {
1845
+ if (!res.headersSent) {
1846
+ console.error("[tfx-hub] bridge error:", error);
1847
+ writeJson(res, 500, { ok: false, error: "Internal server error" });
1848
+ }
1849
+ return;
857
1850
  }
858
- return;
859
1851
  }
860
- }
861
1852
 
862
- if (req.method === 'GET' && servePublicFile(res, path)) {
863
- return;
864
- }
1853
+ if (req.method === "GET" && servePublicFile(res, path)) {
1854
+ return;
1855
+ }
865
1856
 
866
- if (path !== '/mcp') {
867
- res.writeHead(404);
868
- return res.end('Not Found');
869
- }
1857
+ if (path !== "/mcp") {
1858
+ res.writeHead(404);
1859
+ return res.end("Not Found");
1860
+ }
870
1861
 
871
- try {
872
- const sessionIdHeader = req.headers['mcp-session-id'];
873
-
874
- if (req.method === 'POST') {
875
- const body = await parseBody(req);
876
-
877
- if (sessionIdHeader && transports.has(sessionIdHeader)) {
878
- const session = transports.get(sessionIdHeader);
879
- session.transport._lastActivity = Date.now();
880
- await session.transport.handleRequest(req, res, body);
881
- } else if (!sessionIdHeader && isInitializeRequest(body)) {
882
- const transport = new StreamableHTTPServerTransport({
883
- sessionIdGenerator: () => randomUUID(),
884
- onsessioninitialized: (sid) => {
885
- transport._lastActivity = Date.now();
886
- transports.set(sid, { transport, mcp });
887
- },
888
- });
889
- transport.onclose = () => {
890
- if (transport.sessionId) {
891
- const session = transports.get(transport.sessionId);
892
- if (session) {
893
- try { session.mcp.close(); } catch {}
894
- }
895
- transports.delete(transport.sessionId);
1862
+ try {
1863
+ const sessionIdHeader = req.headers["mcp-session-id"];
1864
+
1865
+ if (req.method === "POST") {
1866
+ const body = await parseBody(req);
1867
+
1868
+ if (sessionIdHeader && transports.has(sessionIdHeader)) {
1869
+ const session = transports.get(sessionIdHeader);
1870
+ session.transport._lastActivity = Date.now();
1871
+ await session.transport.handleRequest(req, res, body);
1872
+ } else if (!sessionIdHeader && isInitializeRequest(body)) {
1873
+ const client = parseTfxClientHeader(req.headers["x-tfx-client"]);
1874
+ const clientRef = { client };
1875
+ const initializeStarted = performance.now();
1876
+ const transport = new StreamableHTTPServerTransport({
1877
+ sessionIdGenerator: () => randomUUID(),
1878
+ onsessioninitialized: (sid) => {
1879
+ transport._lastActivity = Date.now();
1880
+ transports.set(sid, {
1881
+ transport,
1882
+ mcp,
1883
+ closing: false,
1884
+ client,
1885
+ createdAt: Date.now(),
1886
+ });
1887
+ },
1888
+ });
1889
+ transport.onclose = () => {
1890
+ const sid = transport.sessionId;
1891
+ if (sid)
1892
+ void closeMcpTransportSession(
1893
+ sid,
1894
+ transports.get(sid),
1895
+ "transport.onclose",
1896
+ );
1897
+ };
1898
+ const mcp = createMcpForSession(clientRef);
1899
+ await mcp.connect(transport);
1900
+ try {
1901
+ await transport.handleRequest(req, res, body);
1902
+ } finally {
1903
+ recordRequest(
1904
+ "initialize",
1905
+ performance.now() - initializeStarted,
1906
+ client,
1907
+ "ingress",
1908
+ );
896
1909
  }
897
- };
898
- const mcp = createMcpForSession();
899
- await mcp.connect(transport);
900
- await transport.handleRequest(req, res, body);
901
- } else {
902
- res.writeHead(400, { 'Content-Type': 'application/json' });
903
- res.end(JSON.stringify({
904
- jsonrpc: '2.0',
905
- error: { code: -32000, message: 'Bad Request: No valid session ID' },
906
- id: null,
907
- }));
908
- }
909
- } else if (req.method === 'GET') {
910
- if (sessionIdHeader && transports.has(sessionIdHeader)) {
911
- await transports.get(sessionIdHeader).transport.handleRequest(req, res);
1910
+ } else {
1911
+ res.writeHead(400, { "Content-Type": "application/json" });
1912
+ res.end(
1913
+ JSON.stringify({
1914
+ jsonrpc: "2.0",
1915
+ error: {
1916
+ code: -32000,
1917
+ message: "Bad Request: No valid session ID",
1918
+ },
1919
+ id: null,
1920
+ }),
1921
+ );
1922
+ }
1923
+ } else if (req.method === "GET") {
1924
+ if (sessionIdHeader && transports.has(sessionIdHeader)) {
1925
+ await transports
1926
+ .get(sessionIdHeader)
1927
+ .transport.handleRequest(req, res);
1928
+ } else {
1929
+ res.writeHead(400);
1930
+ res.end("Invalid or missing session ID");
1931
+ }
1932
+ } else if (req.method === "DELETE") {
1933
+ if (sessionIdHeader && transports.has(sessionIdHeader)) {
1934
+ await transports
1935
+ .get(sessionIdHeader)
1936
+ .transport.handleRequest(req, res);
1937
+ } else {
1938
+ res.writeHead(400);
1939
+ res.end("Invalid or missing session ID");
1940
+ }
912
1941
  } else {
913
- res.writeHead(400);
914
- res.end('Invalid or missing session ID');
1942
+ res.writeHead(405);
1943
+ res.end("Method Not Allowed");
915
1944
  }
916
- } else if (req.method === 'DELETE') {
917
- if (sessionIdHeader && transports.has(sessionIdHeader)) {
918
- await transports.get(sessionIdHeader).transport.handleRequest(req, res);
919
- } else {
920
- res.writeHead(400);
921
- res.end('Invalid or missing session ID');
1945
+ } catch (error) {
1946
+ hubLog.error({ err: error }, "http.error");
1947
+ if (!res.headersSent) {
1948
+ const code =
1949
+ error.statusCode === 413
1950
+ ? 413
1951
+ : error instanceof SyntaxError
1952
+ ? 400
1953
+ : 500;
1954
+ const message =
1955
+ code === 413
1956
+ ? "Body too large"
1957
+ : code === 400
1958
+ ? "Invalid JSON"
1959
+ : "Internal server error";
1960
+ res.writeHead(code, { "Content-Type": "application/json" });
1961
+ res.end(
1962
+ JSON.stringify({
1963
+ jsonrpc: "2.0",
1964
+ error: { code: code === 500 ? -32603 : -32700, message },
1965
+ id: null,
1966
+ }),
1967
+ );
922
1968
  }
923
- } else {
924
- res.writeHead(405);
925
- res.end('Method Not Allowed');
926
- }
927
- } catch (error) {
928
- hubLog.error({ err: error }, 'http.error');
929
- if (!res.headersSent) {
930
- const code = error.statusCode === 413 ? 413
931
- : error instanceof SyntaxError ? 400 : 500;
932
- const message = code === 413 ? 'Body too large'
933
- : code === 400 ? 'Invalid JSON' : 'Internal server error';
934
- res.writeHead(code, { 'Content-Type': 'application/json' });
935
- res.end(JSON.stringify({
936
- jsonrpc: '2.0',
937
- error: { code: code === 500 ? -32603 : -32700, message },
938
- id: null,
939
- }));
940
1969
  }
941
- }
942
- }));
1970
+ }),
1971
+ );
943
1972
 
944
1973
  httpServer.requestTimeout = 30000;
945
1974
  httpServer.headersTimeout = 10000;
@@ -947,41 +1976,74 @@ export async function startHub({
947
1976
  router.startSweeper();
948
1977
 
949
1978
  const hitlTimer = setInterval(() => {
950
- try { hitl.checkTimeouts(); } catch (err) { hubLog.warn({ err }, 'hitl.timeout_check_failed'); }
1979
+ try {
1980
+ hitl.checkTimeouts();
1981
+ } catch (err) {
1982
+ hubLog.warn({ err }, "hitl.timeout_check_failed");
1983
+ }
951
1984
  }, 10000);
952
1985
  hitlTimer.unref();
953
1986
 
954
1987
  // MCP session TTL: sessions idle for SESSION_TTL_MS are closed automatically.
955
1988
  // Configurable via SESSION_TTL_MS (default 30 minutes). The sweep runs every 60 s.
956
- const SESSION_TTL_MS = parseInt(process.env.TFX_SESSION_TTL_MS || '', 10) || 30 * 60 * 1000;
1989
+ const SESSION_TTL_MS =
1990
+ parseInt(process.env.TFX_SESSION_TTL_MS || "", 10) || 30 * 60 * 1000;
957
1991
  const sessionTimer = setInterval(() => {
958
1992
  const now = Date.now();
959
1993
  for (const [sid, session] of transports) {
960
- if (now - (session.transport._lastActivity || 0) <= SESSION_TTL_MS) continue;
961
- try { session.mcp.close(); } catch {}
962
- try { session.transport.close(); } catch {}
963
- transports.delete(sid);
1994
+ if (now - (session.transport._lastActivity || 0) <= SESSION_TTL_MS)
1995
+ continue;
1996
+ void closeMcpTransportSession(sid, session, "session_ttl");
964
1997
  }
965
1998
  }, 60000);
966
1999
  sessionTimer.unref();
967
2000
 
968
2001
  // 고아 node.exe 프로세스 + stale spawn 세션 주기적 정리 (5분마다)
969
- const orphanCleanupTimer = setInterval(() => {
970
- try {
971
- const { killed } = cleanupOrphanNodeProcesses();
972
- if (killed > 0) {
973
- hubLog.info({ killed }, 'hub.orphan_cleanup');
974
- }
975
- } catch {}
2002
+ // TFX_DISABLE_ORPHAN_CLEANUP=1 로 비활성 (active SSH/swarm 세션 보호 회피용 hotfix gate)
2003
+ const orphanCleanupTimer = setInterval(
2004
+ () => {
2005
+ if (process.env.TFX_DISABLE_ORPHAN_CLEANUP === "1") return;
2006
+ try {
2007
+ const { killed, killedProcesses = [] } = cleanupOrphanNodeProcesses();
2008
+ const {
2009
+ killed: runtimeKilled,
2010
+ killedProcesses: runtimeKilledProcesses = [],
2011
+ } = cleanupOrphanRuntimeProcesses();
2012
+ const totalKilled = killed + runtimeKilled;
2013
+ if (totalKilled > 0) {
2014
+ hubLog.info(
2015
+ {
2016
+ killed: totalKilled,
2017
+ processes: [...killedProcesses, ...runtimeKilledProcesses],
2018
+ caller: "timer",
2019
+ },
2020
+ "hub.orphan_cleanup",
2021
+ );
2022
+ }
976
2023
 
977
- // stale tfx-spawn-* psmux 세션 정리 (30분 이상 idle)
978
- try {
979
- const staleKilled = cleanupStaleSpawnSessions(hubLog);
980
- if (staleKilled > 0) {
981
- hubLog.info({ killed: staleKilled }, 'hub.stale_spawn_cleanup');
982
- }
983
- } catch {}
984
- }, 5 * 60 * 1000);
2024
+ const { killed: fsmonitorKilled, stale } = cleanupStaleFsmonitorDaemons(
2025
+ {
2026
+ minAgeMs: 24 * 60 * 60 * 1000,
2027
+ },
2028
+ );
2029
+ if (fsmonitorKilled > 0) {
2030
+ hubLog.info(
2031
+ { killed: fsmonitorKilled, stale: stale.length },
2032
+ "hub.fsmonitor_cleanup",
2033
+ );
2034
+ }
2035
+ } catch {}
2036
+
2037
+ // stale tfx-spawn-* psmux 세션 정리 (30분 이상 idle)
2038
+ try {
2039
+ const staleKilled = cleanupStaleSpawnSessions(hubLog);
2040
+ if (staleKilled > 0) {
2041
+ hubLog.info({ killed: staleKilled }, "hub.stale_spawn_cleanup");
2042
+ }
2043
+ } catch {}
2044
+ },
2045
+ 5 * 60 * 1000,
2046
+ );
985
2047
  orphanCleanupTimer.unref();
986
2048
 
987
2049
  // Evict stale rate-limit buckets once per minute to bound memory usage.
@@ -1000,38 +2062,27 @@ export async function startHub({
1000
2062
 
1001
2063
  mkdirSync(PID_DIR, { recursive: true });
1002
2064
 
1003
- // Stale PID 파일 정리 — 이전 Hub 프로세스가 비정상 종료된 경우
1004
- if (existsSync(PID_FILE)) {
2065
+ const cleanupStartupFailure = async ({ preserveTokenFile = false } = {}) => {
1005
2066
  try {
1006
- const prevInfo = JSON.parse(readFileSync(PID_FILE, 'utf8'));
1007
- const prevPid = Number(prevInfo?.pid);
1008
- if (Number.isFinite(prevPid) && prevPid > 0) {
1009
- try {
1010
- process.kill(prevPid, 0); // alive 체크만
1011
- // 프로세스가 살아있으면 포트 충돌 가능성 — 기존 Hub 재사용 안내
1012
- if (Number(prevInfo.port) === Number(port)) {
1013
- hubLog.warn({ prevPid, port }, 'hub.stale_pid: previous hub still alive on same port');
1014
- }
1015
- } catch {
1016
- // 프로세스 죽음 → stale PID 파일 삭제
1017
- try { unlinkSync(PID_FILE); } catch {}
1018
- hubLog.info({ prevPid }, 'hub.stale_pid_cleaned');
1019
- }
1020
- } else {
1021
- try { unlinkSync(PID_FILE); } catch {}
1022
- }
1023
- } catch {
1024
- try { unlinkSync(PID_FILE); } catch {}
2067
+ router.stopSweeper();
2068
+ } catch {}
2069
+ try {
2070
+ await pipe.stop();
2071
+ } catch {}
2072
+ try {
2073
+ await assignCallbacks.stop();
2074
+ } catch {}
2075
+ try {
2076
+ await delegatorWorker.stop();
2077
+ } catch {}
2078
+ try {
2079
+ store.close();
2080
+ } catch {}
2081
+ if (!preserveTokenFile) {
2082
+ try {
2083
+ unlinkSync(TOKEN_FILE);
2084
+ } catch {}
1025
2085
  }
1026
- }
1027
-
1028
- const cleanupStartupFailure = async () => {
1029
- try { router.stopSweeper(); } catch {}
1030
- try { await pipe.stop(); } catch {}
1031
- try { await assignCallbacks.stop(); } catch {}
1032
- try { await delegatorWorker.stop(); } catch {}
1033
- try { store.close(); } catch {}
1034
- try { unlinkSync(TOKEN_FILE); } catch {}
1035
2086
  releaseStartupLock();
1036
2087
  };
1037
2088
 
@@ -1044,120 +2095,231 @@ export async function startHub({
1044
2095
  }
1045
2096
 
1046
2097
  return await new Promise((resolveHub, reject) => {
1047
- httpServer.listen(port, host, () => {
1048
- try {
1049
- let idleTimer = null;
1050
- let stopPromise = null;
2098
+ let bindAttempts = 0;
2099
+
2100
+ const listenOnce = () => {
2101
+ const onError = (err) => {
2102
+ void (async () => {
2103
+ if (err.code !== "EADDRINUSE") {
2104
+ await cleanupStartupFailure();
2105
+ reject(err);
2106
+ return;
2107
+ }
1051
2108
 
1052
- const info = {
1053
- port,
1054
- host,
1055
- dbPath,
1056
- pid: process.pid,
1057
- hubToken: HUB_TOKEN,
1058
- authMode: HUB_TOKEN ? 'token-required' : 'localhost-only',
1059
- url: buildHubUrl(host, port),
1060
- pipe_path: pipe.path,
1061
- pipePath: pipe.path,
1062
- assign_callback_pipe_path: assignCallbacks.path,
1063
- assignCallbackPipePath: assignCallbacks.path,
1064
- version,
1065
- storeType: store.type || 'sqlite',
1066
- idleTimeoutMs: hubIdleTimeoutMs,
1067
- };
2109
+ const conflict = await resolvePortInUse({
2110
+ port,
2111
+ host,
2112
+ version,
2113
+ });
2114
+ if (conflict.action === "retry" && bindAttempts < 1) {
2115
+ bindAttempts += 1;
2116
+ hubLog.warn(
2117
+ { port, reason: conflict.cleaned?.reason },
2118
+ "hub.port_in_use_retry_after_stale_pid_cleanup",
2119
+ );
2120
+ listenOnce();
2121
+ return;
2122
+ }
1068
2123
 
1069
- writeState({
1070
- pid: process.pid,
1071
- port,
1072
- host,
1073
- auth_mode: HUB_TOKEN ? 'token-required' : 'localhost-only',
1074
- url: info.url,
1075
- pipe_path: pipe.path,
1076
- pipePath: pipe.path,
1077
- assign_callback_pipe_path: assignCallbacks.path,
1078
- assignCallbackPipePath: assignCallbacks.path,
1079
- authMode: HUB_TOKEN ? 'token-required' : 'localhost-only',
1080
- startedAt,
1081
- started: startedAtMs,
1082
- version,
1083
- sessionId,
1084
- session_id: sessionId,
1085
- });
1086
- releaseStartupLock();
1087
-
1088
- hubLog.info({ url: info.url, pipePath: pipe.path, assignCallbackPath: assignCallbacks.path, pid: process.pid, storeType: info.storeType, version }, 'hub.started');
1089
- hubLog.debug({ publicDir: PUBLIC_DIR, exists: existsSync(PUBLIC_DIR), hasDashboard: existsSync(resolve(PUBLIC_DIR, 'dashboard.html')) }, 'hub.public_dir');
1090
-
1091
- const stopFn = async () => {
1092
- if (stopPromise) return stopPromise;
1093
-
1094
- stopPromise = (async () => {
1095
- router.stopSweeper();
1096
- clearInterval(hitlTimer);
1097
- clearInterval(sessionTimer);
1098
- clearInterval(rateLimitTimer);
1099
- clearInterval(orphanCleanupTimer);
1100
- if (idleTimer) {
1101
- clearInterval(idleTimer);
1102
- }
1103
- for (const [, session] of transports) {
1104
- try { await session.mcp.close(); } catch {}
1105
- try { await session.transport.close(); } catch {}
1106
- }
1107
- transports.clear();
1108
- await pipe.stop();
1109
- await assignCallbacks.stop();
1110
- await delegatorWorker.stop().catch(() => {});
1111
- store.close();
1112
- try { unlinkSync(PID_FILE); } catch {}
1113
- try { unlinkSync(TOKEN_FILE); } catch {}
1114
- httpServer.closeAllConnections();
1115
- await new Promise((resolveClose) => httpServer.close(resolveClose));
1116
- })().catch((error) => {
1117
- stopPromise = null;
1118
- throw error;
2124
+ await cleanupStartupFailure({ preserveTokenFile: true });
2125
+ if (conflict.action === "reuse") {
2126
+ hubLog.info(
2127
+ {
2128
+ port,
2129
+ pid: conflict.peer?.pid,
2130
+ version: conflict.peer?.version,
2131
+ url: conflict.url,
2132
+ },
2133
+ "hub.already_running",
2134
+ );
2135
+ resolveHub({
2136
+ reused: true,
2137
+ external: true,
2138
+ port,
2139
+ pid: conflict.peer?.pid,
2140
+ url: conflict.url,
2141
+ stop: async () => false,
2142
+ });
2143
+ return;
2144
+ }
2145
+
2146
+ hubLog.error({ port, host }, "hub.port_in_use");
2147
+ const wrapped = new Error(conflict.message);
2148
+ wrapped.code = "EADDRINUSE";
2149
+ reject(wrapped);
2150
+ })();
2151
+ };
2152
+
2153
+ httpServer.once("error", onError);
2154
+ httpServer.listen(port, host, () => {
2155
+ httpServer.off("error", onError);
2156
+ try {
2157
+ let idleTimer = null;
2158
+ let stopPromise = null;
2159
+
2160
+ const info = {
2161
+ port,
2162
+ host,
2163
+ dbPath,
2164
+ pid: process.pid,
2165
+ hubToken: HUB_TOKEN,
2166
+ authMode: HUB_TOKEN ? "token-required" : "localhost-only",
2167
+ url: buildHubUrl(host, port),
2168
+ pipe_path: pipe.path,
2169
+ pipePath: pipe.path,
2170
+ assign_callback_pipe_path: assignCallbacks.path,
2171
+ assignCallbackPipePath: assignCallbacks.path,
2172
+ version,
2173
+ storeType: store.type || "sqlite",
2174
+ idleTimeoutMs: hubIdleTimeoutMs,
2175
+ };
2176
+
2177
+ writeState({
2178
+ pid: process.pid,
2179
+ port,
2180
+ host,
2181
+ auth_mode: HUB_TOKEN ? "token-required" : "localhost-only",
2182
+ url: info.url,
2183
+ pipe_path: pipe.path,
2184
+ pipePath: pipe.path,
2185
+ assign_callback_pipe_path: assignCallbacks.path,
2186
+ assignCallbackPipePath: assignCallbacks.path,
2187
+ authMode: HUB_TOKEN ? "token-required" : "localhost-only",
2188
+ startedAt,
2189
+ started: startedAtMs,
2190
+ version,
2191
+ sessionId,
2192
+ session_id: sessionId,
1119
2193
  });
2194
+ releaseStartupLock();
2195
+ void syncHubMcpSettingsIfAvailable({ hubUrl: info.url });
2196
+
2197
+ hubLog.info(
2198
+ {
2199
+ url: info.url,
2200
+ pipePath: pipe.path,
2201
+ assignCallbackPath: assignCallbacks.path,
2202
+ pid: process.pid,
2203
+ storeType: info.storeType,
2204
+ version,
2205
+ },
2206
+ "hub.started",
2207
+ );
2208
+ hubLog.debug(
2209
+ {
2210
+ publicDir: PUBLIC_DIR,
2211
+ exists: existsSync(PUBLIC_DIR),
2212
+ hasDashboard: existsSync(resolve(PUBLIC_DIR, "dashboard.html")),
2213
+ },
2214
+ "hub.public_dir",
2215
+ );
2216
+
2217
+ /**
2218
+ * Hub 서버 정지 함수.
2219
+ *
2220
+ * Trade-off (F01 — 영구 poisoning 허용):
2221
+ * 첫 정지 호출에서 cleanup 파이프라인이 실패하면 stopPromise는 실패 상태로
2222
+ * 고정되고, 이후 모든 호출은 동일한 실패 promise를 반환합니다. stopPromise를
2223
+ * null로 리셋하면 재시도가 가능하지만, router sweeper / transports / pipe 등이
2224
+ * 이미 부분 해제된 상태에서 두 번째 close가 실행되면 use-after-close 및
2225
+ * race condition을 유발합니다. 실패한 stopFn은 프로세스 전체 재시작으로 복구해야
2226
+ * 하며, 이 동작은 의도된 설계입니다.
2227
+ */
2228
+ const stopFn = async () => {
2229
+ if (stopPromise) return stopPromise;
2230
+
2231
+ stopPromise = (async () => {
2232
+ router.stopSweeper();
2233
+ clearInterval(hitlTimer);
2234
+ clearInterval(sessionTimer);
2235
+ clearInterval(rateLimitTimer);
2236
+ clearInterval(orphanCleanupTimer);
2237
+ if (idleTimer) {
2238
+ clearInterval(idleTimer);
2239
+ }
2240
+ for (const [sid, session] of Array.from(transports)) {
2241
+ await closeMcpTransportSession(sid, session, "hub.stop");
2242
+ }
2243
+ transports.clear();
2244
+ await pipe.stop();
2245
+ await assignCallbacks.stop();
2246
+ await delegatorWorker.stop().catch(() => {});
2247
+ try {
2248
+ synapseRegistry.destroy();
2249
+ } catch {}
2250
+ store.close();
2251
+ try {
2252
+ unlinkSync(PID_FILE);
2253
+ } catch {}
2254
+ try {
2255
+ unlinkSync(TOKEN_FILE);
2256
+ } catch {}
2257
+ httpServer.closeAllConnections();
2258
+ await new Promise((resolveClose) =>
2259
+ httpServer.close(resolveClose),
2260
+ );
2261
+ })().catch((error) => {
2262
+ hubLog.error(
2263
+ { err: String(error?.message || error) },
2264
+ "hub.stop_error",
2265
+ );
2266
+ // stopPromise를 null로 리셋하지 않음 — double-close 방지
2267
+ });
1120
2268
 
1121
- return stopPromise;
1122
- };
2269
+ return stopPromise;
2270
+ };
2271
+
2272
+ if (hubIdleTimeoutMs > 0) {
2273
+ idleTimer = setInterval(() => {
2274
+ const idleMs = Date.now() - lastRequestAt;
2275
+ if (idleMs < hubIdleTimeoutMs) return;
2276
+ hubLog.warn(
2277
+ { idleMs, idleTimeoutMs: hubIdleTimeoutMs, port },
2278
+ "hub.idle_timeout_shutdown",
2279
+ );
2280
+ void stopFn().catch((error) => {
2281
+ hubLog.error(
2282
+ {
2283
+ err: error,
2284
+ idleMs,
2285
+ idleTimeoutMs: hubIdleTimeoutMs,
2286
+ port,
2287
+ },
2288
+ "hub.idle_timeout_shutdown_failed",
2289
+ );
2290
+ });
2291
+ }, hubIdleSweepMs);
2292
+ idleTimer.unref();
2293
+ }
1123
2294
 
1124
- idleTimer = setInterval(() => {
1125
- const idleMs = Date.now() - lastRequestAt;
1126
- if (idleMs < hubIdleTimeoutMs) return;
1127
- hubLog.warn({ idleMs, idleTimeoutMs: hubIdleTimeoutMs, port }, 'hub.idle_timeout_shutdown');
1128
- void stopFn().catch((error) => {
1129
- hubLog.error({ err: error, idleMs, idleTimeoutMs: hubIdleTimeoutMs, port }, 'hub.idle_timeout_shutdown_failed');
2295
+ resolveHub({
2296
+ reused: false,
2297
+ external: false,
2298
+ ...info,
2299
+ httpServer,
2300
+ store,
2301
+ router,
2302
+ hitl,
2303
+ pipe,
2304
+ assignCallbacks,
2305
+ delegatorService,
2306
+ delegatorWorker,
2307
+ getMcpSessions: () =>
2308
+ Array.from(transports.entries()).map(([id, session]) => ({
2309
+ id,
2310
+ client: session.client || "unknown",
2311
+ createdAt: session.createdAt ?? null,
2312
+ closing: Boolean(session.closing),
2313
+ })),
2314
+ stop: stopFn,
1130
2315
  });
1131
- }, hubIdleSweepMs);
1132
- idleTimer.unref();
1133
-
1134
- resolveHub({
1135
- reused: false,
1136
- external: false,
1137
- ...info,
1138
- httpServer,
1139
- store,
1140
- router,
1141
- hitl,
1142
- pipe,
1143
- assignCallbacks,
1144
- delegatorService,
1145
- delegatorWorker,
1146
- stop: stopFn,
1147
- });
1148
- } catch (error) {
1149
- void cleanupStartupFailure().finally(() => reject(error));
1150
- }
1151
- });
1152
- httpServer.on('error', (err) => {
1153
- void cleanupStartupFailure();
1154
- if (err.code === 'EADDRINUSE') {
1155
- hubLog.error({ port, host }, 'hub.port_in_use: port already occupied — check for existing hub or other service');
1156
- reject(new Error(`Hub 포트 ${port}이(가) 이미 사용 중입니다. 기존 Hub 프로세스를 확인하세요. (PID file: ${PID_FILE})`));
1157
- } else {
1158
- reject(err);
1159
- }
1160
- });
2316
+ } catch (error) {
2317
+ void cleanupStartupFailure().finally(() => reject(error));
2318
+ }
2319
+ });
2320
+ };
2321
+
2322
+ listenOnce();
1161
2323
  });
1162
2324
  }
1163
2325
 
@@ -1198,8 +2360,14 @@ export async function getOrCreateServer(opts = {}) {
1198
2360
  */
1199
2361
  function cleanupStaleSpawnSessions(log) {
1200
2362
  const MAX_AGE_MS = 30 * 60 * 1000;
1201
- const IDLE_PROMPT_RE = /^(PS\s|[$%>#]\s*$|\w+@[\w.-]+[:\s]|╰─|╭─|[fb]wd-i-search:|client_loop:\s|Connection\s+(reset|closed))/;
1202
- const execOpts = { encoding: "utf8", timeout: 5000, stdio: ["pipe", "pipe", "pipe"], windowsHide: true };
2363
+ const IDLE_PROMPT_RE =
2364
+ /^(PS\s|[$%>#]\s*$|\w+@[\w.-]+[:\s]|╰─|╭─|[fb]wd-i-search:|client_loop:\s|Connection\s+(reset|closed))/;
2365
+ const execOpts = {
2366
+ encoding: "utf8",
2367
+ timeout: 5000,
2368
+ stdio: ["pipe", "pipe", "pipe"],
2369
+ windowsHide: true,
2370
+ };
1203
2371
 
1204
2372
  let killed = 0;
1205
2373
  let raw;
@@ -1211,7 +2379,9 @@ function cleanupStaleSpawnSessions(log) {
1211
2379
 
1212
2380
  const now = Date.now();
1213
2381
  for (const line of raw.split(/\r?\n/)) {
1214
- const match = line.match(/^(tfx-spawn-[^:]+):\s+\d+\s+windows?\s+\(created\s+(.+)\)/);
2382
+ const match = line.match(
2383
+ /^(tfx-spawn-[^:]+):\s+\d+\s+windows?\s+\(created\s+(.+)\)/,
2384
+ );
1215
2385
  if (!match) continue;
1216
2386
 
1217
2387
  const [, sessionName, createdStr] = match;
@@ -1221,8 +2391,14 @@ function cleanupStaleSpawnSessions(log) {
1221
2391
 
1222
2392
  // pane 내용 확인 — 마지막 3줄 중 idle 쉘 프롬프트가 있는지
1223
2393
  try {
1224
- const pane = execSyncHub(`psmux capture-pane -t "${sessionName}:0.0" -p`, execOpts);
1225
- const tailLines = pane.split(/\r?\n/).filter((l) => l.trim()).slice(-3);
2394
+ const pane = execSyncHub(
2395
+ `psmux capture-pane -t "${sessionName}:0.0" -p`,
2396
+ execOpts,
2397
+ );
2398
+ const tailLines = pane
2399
+ .split(/\r?\n/)
2400
+ .filter((l) => l.trim())
2401
+ .slice(-3);
1226
2402
  const hasIdleLine = tailLines.some((l) => IDLE_PROMPT_RE.test(l.trim()));
1227
2403
  if (!hasIdleLine) continue; // 아직 활성 — 건드리지 않음
1228
2404
  } catch {
@@ -1233,32 +2409,336 @@ function cleanupStaleSpawnSessions(log) {
1233
2409
  try {
1234
2410
  execSyncHub(`psmux kill-session -t "${sessionName}"`, execOpts);
1235
2411
  killed++;
1236
- if (log) log.info({ session: sessionName, ageMin: Math.round((now - created) / 60000) }, "hub.stale_spawn_killed");
2412
+ if (log)
2413
+ log.info(
2414
+ { session: sessionName, ageMin: Math.round((now - created) / 60000) },
2415
+ "hub.stale_spawn_killed",
2416
+ );
1237
2417
  } catch {}
1238
2418
  }
1239
2419
 
1240
2420
  return killed;
1241
2421
  }
1242
2422
 
1243
- const selfRun = process.argv[1]?.replace(/\\/g, '/').endsWith('hub/server.mjs');
2423
+ // ── Quota check for all accounts ───────────────────────────────
2424
+
2425
+ const QUOTA_CACHE_PATH = join(CACHE_DIR, "broker-quota-cache.json");
2426
+
2427
+ async function checkSingleAccountQuota(acct) {
2428
+ try {
2429
+ const authPath = join(PID_DIR, acct.authFile);
2430
+ if (!authPath.startsWith(PID_DIR + sep))
2431
+ return { id: acct.id, status: "path_blocked" };
2432
+ if (!existsSync(authPath)) return { id: acct.id, status: "no_auth" };
2433
+ const auth = JSON.parse(readFileSync(authPath, "utf8"));
2434
+ if (acct.provider === "codex") {
2435
+ const token = auth.tokens?.access_token || auth.OPENAI_API_KEY || "";
2436
+ if (!token) return { id: acct.id, status: "no_token" };
2437
+ const r = await fetch("https://api.openai.com/v1/chat/completions", {
2438
+ method: "POST",
2439
+ signal: AbortSignal.timeout(8000),
2440
+ headers: {
2441
+ Authorization: "Bearer " + token,
2442
+ "Content-Type": "application/json",
2443
+ },
2444
+ body: JSON.stringify({
2445
+ model: "gpt-4.1-nano",
2446
+ messages: [{ role: "user", content: "hi" }],
2447
+ max_tokens: 1,
2448
+ }),
2449
+ });
2450
+ const hdrs = Object.fromEntries(
2451
+ [...r.headers.entries()].filter(([k]) => /ratelimit/i.test(k)),
2452
+ );
2453
+ if (r.status === 429)
2454
+ return { id: acct.id, status: "quota_hit", http: 429, headers: hdrs };
2455
+ if (r.status === 401) {
2456
+ // 401이어도 ratelimit 헤더가 있으면 쿼터 정보 추출 가능
2457
+ return { id: acct.id, status: "auth_error", http: 401, headers: hdrs };
2458
+ }
2459
+ return {
2460
+ id: acct.id,
2461
+ status: r.ok ? "ok" : "error",
2462
+ http: r.status,
2463
+ headers: hdrs,
2464
+ };
2465
+ }
2466
+ // gemini — OAuth token refresh needed for accurate check
2467
+ return { id: acct.id, status: "oauth_check_needed" };
2468
+ } catch (e) {
2469
+ return {
2470
+ id: acct.id,
2471
+ status: "error",
2472
+ message: e.message?.substring(0, 60),
2473
+ };
2474
+ }
2475
+ }
2476
+
2477
+ async function refreshAllAccountQuotas() {
2478
+ const snap = brokerInstance?.snapshot() || [];
2479
+ const checks = snap
2480
+ .filter((a) => a.authFile)
2481
+ .map((a) => checkSingleAccountQuota(a));
2482
+ const settled = await Promise.allSettled(checks);
2483
+ const results = settled.map((s, i) =>
2484
+ s.status === "fulfilled"
2485
+ ? s.value
2486
+ : {
2487
+ id: snap.filter((a) => a.authFile)[i]?.id ?? "unknown",
2488
+ status: "error",
2489
+ message: String(s.reason?.message || s.reason).substring(0, 60),
2490
+ },
2491
+ );
2492
+ // 캐시 저장
2493
+ try {
2494
+ writeFileSync(
2495
+ QUOTA_CACHE_PATH,
2496
+ JSON.stringify({ ts: Date.now(), results }),
2497
+ );
2498
+ } catch {
2499
+ /* best-effort */
2500
+ }
2501
+ return results;
2502
+ }
2503
+
2504
+ function _loadQuotaCache() {
2505
+ try {
2506
+ if (!existsSync(QUOTA_CACHE_PATH)) return null;
2507
+ return JSON.parse(readFileSync(QUOTA_CACHE_PATH, "utf8"));
2508
+ } catch {
2509
+ return null;
2510
+ }
2511
+ }
2512
+
2513
+ // ── Broker Dashboard HTML ──────────────────────────────────────
2514
+
2515
+ function renderBrokerDashboard() {
2516
+ return `<!DOCTYPE html><html lang="ko"><head><meta charset="utf-8">
2517
+ <title>Account Broker Dashboard</title>
2518
+ <style>
2519
+ *{margin:0;padding:0;box-sizing:border-box}
2520
+ body{background:#0a0a0b;color:#e8e8ec;font:14px/1.6 'SF Mono',Consolas,monospace;padding:24px}
2521
+ h1{font-size:18px;color:#a8b1ff;margin-bottom:8px}
2522
+ .toolbar{margin-bottom:16px;display:flex;gap:8px;align-items:center}
2523
+ .toolbar button{background:#1a1a2e;color:#a8b1ff;border:1px solid rgba(255,255,255,0.1);border-radius:4px;padding:4px 12px;cursor:pointer;font:12px inherit}
2524
+ .toolbar button:hover{background:#252547}
2525
+ .toolbar .status-text{font-size:11px;color:#555}
2526
+ .grid{display:grid;grid-template-columns:repeat(auto-fill,minmax(380px,1fr));gap:12px}
2527
+ .card{background:#141416;border:1px solid rgba(255,255,255,0.06);border-radius:8px;padding:16px}
2528
+ .card.available{border-left:3px solid #34d399}
2529
+ .card.busy{border-left:3px solid #fbbf24}
2530
+ .card.cooldown{border-left:3px solid #f87171}
2531
+ .card.circuit-open{border-left:3px solid #ef4444;opacity:0.7}
2532
+ .provider{font-size:11px;text-transform:uppercase;letter-spacing:1px;color:#888}
2533
+ .id{font-size:13px;color:#c8c8d0;margin:4px 0}
2534
+ .status{font-size:12px;font-weight:600;padding:2px 8px;border-radius:4px;display:inline-block}
2535
+ .status.available{background:#064e3b;color:#34d399}
2536
+ .status.busy{background:#451a03;color:#fbbf24}
2537
+ .status.cooldown{background:#450a0a;color:#f87171}
2538
+ .status.circuit-open{background:#450a0a;color:#ef4444}
2539
+ .gauge{margin-top:8px}
2540
+ .gauge-row{display:flex;align-items:center;gap:8px;margin:3px 0}
2541
+ .gauge-label{font-size:10px;color:#888;width:28px;text-align:right}
2542
+ .gauge-bar{flex:1;height:8px;background:#1a1a1e;border-radius:4px;overflow:hidden}
2543
+ .gauge-fill{height:100%;border-radius:4px;transition:width .3s}
2544
+ .gauge-fill.green{background:linear-gradient(90deg,#34d399,#059669)}
2545
+ .gauge-fill.yellow{background:linear-gradient(90deg,#fbbf24,#d97706)}
2546
+ .gauge-fill.red{background:linear-gradient(90deg,#f87171,#dc2626)}
2547
+ .gauge-pct{font-size:10px;color:#999;width:32px}
2548
+ .gauge-reset{font-size:9px;color:#555}
2549
+ .meta{margin-top:6px;font-size:11px;color:#666;line-height:1.8}
2550
+ .meta b{color:#999}
2551
+ .timer{color:#f87171;font-weight:600}
2552
+ .refresh-bar{color:#555;font-size:11px;margin-top:16px}
2553
+ </style></head><body>
2554
+ <h1>Account Broker Dashboard</h1>
2555
+ <div class="toolbar">
2556
+ <button onclick="checkQuotas()">Check All Quotas</button>
2557
+ <button onclick="reloadBroker()">Reload Broker</button>
2558
+ <span id="toolbar-status" class="status-text"></span>
2559
+ </div>
2560
+ <div id="grid" class="grid"></div>
2561
+ <p class="refresh-bar">auto-refresh: 10s | <a href="/broker/snapshot" style="color:#666">JSON</a> | <a href="/broker/quota-refresh" style="color:#666">Quota API</a></p>
2562
+ <script>
2563
+ let quotaData={};
2564
+ function fmt(ms){if(ms<=0)return'-';const s=Math.floor(ms/1000),m=Math.floor(s/60),h=Math.floor(m/60),d=Math.floor(h/24);if(d>0)return d+'d '+h%24+'h';if(h>0)return h+'h '+m%60+'m';if(m>0)return m+'m '+s%60+'s';return s+'s'}
2565
+
2566
+ function gaugeColor(pct){return pct<60?'green':pct<85?'yellow':'red'}
2567
+
2568
+ function renderGauge(label,pct,resetTs){
2569
+ if(pct==null)return '';
2570
+ const now=Date.now()/1000;
2571
+ const resetIn=resetTs>now?fmt((resetTs-now)*1000):'';
2572
+ return '<div class="gauge-row"><span class="gauge-label">'+label+'</span>'+
2573
+ '<div class="gauge-bar"><div class="gauge-fill '+gaugeColor(pct)+'" style="width:'+pct+'%"></div></div>'+
2574
+ '<span class="gauge-pct">'+pct+'%</span>'+
2575
+ (resetIn?'<span class="gauge-reset">reset '+resetIn+'</span>':'')+
2576
+ '</div>';
2577
+ }
2578
+
2579
+ function renderQuotaGauges(acctId){
2580
+ const q=quotaData[acctId];
2581
+ if(!q)return '<div class="gauge"><div class="gauge-row"><span style="font-size:10px;color:#444">quota: click Check All Quotas</span></div></div>';
2582
+ if(q.status==='quota_hit')return '<div class="gauge"><div class="gauge-row"><span style="font-size:10px;color:#f87171">QUOTA EXHAUSTED</span></div></div>';
2583
+ if(q.status==='auth_error'||q.status==='no_token'||q.status==='oauth_check_needed'){
2584
+ // 401이어도 ratelimit 헤더가 있으면 표시
2585
+ const h=q.headers||{};
2586
+ const limReq=h['x-ratelimit-limit-requests'],remReq=h['x-ratelimit-remaining-requests'];
2587
+ const limTok=h['x-ratelimit-limit-tokens'],remTok=h['x-ratelimit-remaining-tokens'];
2588
+ if(limReq){
2589
+ const pctReq=Math.round((1-remReq/limReq)*100);
2590
+ const pctTok=limTok?Math.round((1-remTok/limTok)*100):null;
2591
+ return '<div class="gauge">'+renderGauge('req',pctReq,0)+(pctTok!=null?renderGauge('tok',pctTok,0):'')+'</div>';
2592
+ }
2593
+ return '<div class="gauge"><div class="gauge-row"><span style="font-size:10px;color:#fbbf24">'+(q.status==='oauth_check_needed'?'OAuth refresh needed':'auth: token refresh needed')+'</span></div></div>';
2594
+ }
2595
+ if(q.status==='ok'||q.status==='error'){
2596
+ const h=q.headers||{};
2597
+ const limReq=h['x-ratelimit-limit-requests'],remReq=h['x-ratelimit-remaining-requests'];
2598
+ const limTok=h['x-ratelimit-limit-tokens'],remTok=h['x-ratelimit-remaining-tokens'];
2599
+ const resetReq=h['x-ratelimit-reset-requests'],resetTok=h['x-ratelimit-reset-tokens'];
2600
+ if(!limReq)return '';
2601
+ const pctReq=Math.round((1-remReq/limReq)*100);
2602
+ const pctTok=limTok?Math.round((1-remTok/limTok)*100):null;
2603
+ return '<div class="gauge">'+renderGauge('req',pctReq,0)+(pctTok!=null?renderGauge('tok',pctTok,0):'')+'</div>';
2604
+ }
2605
+ return '';
2606
+ }
2607
+
2608
+ async function refresh(){
2609
+ try{
2610
+ const r=await fetch('/broker/snapshot');
2611
+ const d=await r.json();
2612
+ if(!d.ok)return;
2613
+ const now=d.ts;
2614
+ const grid=document.getElementById('grid');
2615
+ grid.innerHTML=d.accounts.map(a=>{
2616
+ const cd=a.cooldownUntil>now?a.cooldownUntil-now:0;
2617
+ const rm=a.remainingMs||0;
2618
+ let st='available',sl='Available';
2619
+ if(a.circuitState==='open'){st='circuit-open';sl='Circuit Open'}
2620
+ else if(cd>0){st='cooldown';sl='Cooldown'}
2621
+ else if(a.busy){st='busy';sl='Busy ('+fmt(rm)+')'}
2622
+ return '<div class="card '+st+'">'+
2623
+ '<div class="provider">'+a.provider+' / '+(a.tier||'unknown')+'</div>'+
2624
+ '<div class="id">'+a.id+'</div>'+
2625
+ '<span class="status '+st+'">'+sl+'</span>'+
2626
+ (cd>0?'<span class="timer" style="margin-left:8px">'+fmt(cd)+' remaining</span>':'')+
2627
+ renderQuotaGauges(a.id)+
2628
+ '<div class="meta">'+
2629
+ '<b>Sessions:</b> '+a.totalSessions+
2630
+ (a.lastUsedAt?' | <b>Last:</b> '+new Date(a.lastUsedAt).toLocaleTimeString():'')+
2631
+ (a.failureTimestamps?.length?' | <b>Failures:</b> '+a.failureTimestamps.length:'')+
2632
+ (a.circuitState!=='closed'?' | <b>Circuit:</b> '+a.circuitState:'')+
2633
+ '</div></div>'
2634
+ }).join('');
2635
+ }catch(e){console.error('refresh failed',e)}
2636
+ }
2637
+
2638
+ async function checkQuotas(){
2639
+ const el=document.getElementById('toolbar-status');
2640
+ el.textContent='checking quotas...';
2641
+ try{
2642
+ const r=await fetch('/broker/quota-refresh',{method:'POST'});
2643
+ const d=await r.json();
2644
+ if(d.ok){
2645
+ d.results.forEach(q=>{quotaData[q.id]=q});
2646
+ el.textContent='updated '+d.results.length+' accounts ('+new Date().toLocaleTimeString()+')';
2647
+ refresh();
2648
+ }else{el.textContent='error';}
2649
+ }catch(e){el.textContent='failed: '+e.message;}
2650
+ }
2651
+
2652
+ async function reloadBroker(){
2653
+ await fetch('/broker/reload',{method:'POST'});
2654
+ document.getElementById('toolbar-status').textContent='broker reloaded';
2655
+ refresh();
2656
+ }
2657
+
2658
+ // 캐시에서 초기 로드
2659
+ fetch('/broker/quota-refresh').catch(()=>{});
2660
+ refresh();setInterval(refresh,10000);
2661
+ </script></body></html>`;
2662
+ }
2663
+
2664
+ const selfRun = process.argv[1]?.replace(/\\/g, "/").endsWith("hub/server.mjs");
1244
2665
  if (selfRun) {
1245
- const port = parseInt(process.env.TFX_HUB_PORT || '27888', 10);
2666
+ const port = resolveHubPort(process.env);
1246
2667
  const dbPath = process.env.TFX_HUB_DB || undefined;
1247
2668
 
1248
- startHub({ port, dbPath }).then((info) => {
1249
- const shutdown = async (signal) => {
1250
- hubLog.info({ signal }, 'hub.stopping');
1251
- try { cleanupOrphanNodeProcesses(); } catch {}
1252
- try { cleanupStaleSpawnSessions(hubLog); } catch {}
1253
- await info.stop();
1254
- process.exit(0);
1255
- };
1256
- process.on('SIGINT', () => shutdown('SIGINT'));
1257
- process.on('SIGTERM', () => shutdown('SIGTERM'));
1258
- }).catch((error) => {
1259
- hubLog.fatal({ err: error }, 'hub.start_failed');
2669
+ const cleanupPidFile = () => {
2670
+ try {
2671
+ unlinkSync(PID_FILE);
2672
+ } catch {}
2673
+ };
2674
+
2675
+ process.on("unhandledRejection", (err) => {
2676
+ hubLog.fatal({ err }, "hub.unhandledRejection");
2677
+ cleanupPidFile();
1260
2678
  process.exit(1);
1261
2679
  });
2680
+ process.on("uncaughtException", (err) => {
2681
+ hubLog.fatal({ err }, "hub.uncaughtException");
2682
+ cleanupPidFile();
2683
+ process.exit(1);
2684
+ });
2685
+
2686
+ startHub({ port, dbPath })
2687
+ .then((info) => {
2688
+ if (info?.reused && info?.external) {
2689
+ hubLog.info(
2690
+ { pid: info.pid, port: info.port, url: info.url },
2691
+ "hub.already_running_exit",
2692
+ );
2693
+ process.exit(0);
2694
+ return;
2695
+ }
2696
+ const shutdown = async (signal) => {
2697
+ hubLog.info({ signal }, "hub.stopping");
2698
+ try {
2699
+ if (process.env.TFX_DISABLE_ORPHAN_CLEANUP !== "1") {
2700
+ const { killed: orphanKilled, killedProcesses = [] } =
2701
+ cleanupOrphanNodeProcesses();
2702
+ const {
2703
+ killed: runtimeKilled,
2704
+ killedProcesses: runtimeKilledProcesses = [],
2705
+ } = cleanupOrphanRuntimeProcesses();
2706
+ const totalKilled = orphanKilled + runtimeKilled;
2707
+ if (totalKilled > 0) {
2708
+ hubLog.info(
2709
+ {
2710
+ killed: totalKilled,
2711
+ processes: [...killedProcesses, ...runtimeKilledProcesses],
2712
+ caller: "shutdown",
2713
+ },
2714
+ "hub.orphan_cleanup",
2715
+ );
2716
+ }
2717
+ }
2718
+ const { killed, stale } = cleanupStaleFsmonitorDaemons({
2719
+ minAgeMs: 24 * 60 * 60 * 1000,
2720
+ });
2721
+ if (killed > 0) {
2722
+ hubLog.info(
2723
+ { killed, stale: stale.length },
2724
+ "hub.fsmonitor_cleanup",
2725
+ );
2726
+ }
2727
+ } catch {}
2728
+ try {
2729
+ cleanupStaleSpawnSessions(hubLog);
2730
+ } catch {}
2731
+ await info.stop();
2732
+ process.exit(0);
2733
+ };
2734
+ process.on("SIGINT", () => shutdown("SIGINT"));
2735
+ process.on("SIGTERM", () => shutdown("SIGTERM"));
2736
+ })
2737
+ .catch((error) => {
2738
+ hubLog.fatal({ err: error }, "hub.start_failed");
2739
+ cleanupPidFile();
2740
+ process.exit(1);
2741
+ });
1262
2742
  }
1263
2743
 
1264
2744
  export { startHub as createServer };