Fix Comfy wake: launch when HTTP is down instead of treating leftover Python as already booting.

Co-authored-by: Cursor <cursoragent@cursor.com>

scripts/comfy-host-agent.mjs

@@ -16,6 +16,7 @@ const trainingUrl = String(process.env.TRAINING_CONTROL_URL || 'http://127.0.0.1
 
 let lastWorkAt = Date.now()
 let lastQueueRunning = 0
+let lastLaunchAt = 0
 let stoppedByAgent = false
 let lastHealthyPort = 0
 let lastProcessUp = false
@@ -162,12 +163,23 @@ function ensureProxy(targetPort) {
   })
 }
 
+function markAsleep() {
+  stoppedByAgent = true
+  lastQueueRunning = 0
+  lastHealthyPort = 0
+  lastProcessUp = false
+  lastLaunchAt = 0
+  proxyTarget = 0
+}
+
 async function syncProxy() {
   const healthyPort = await findHealthyPort()
   lastHealthyPort = healthyPort || 0
   if (healthyPort) {
     ensureProxy(healthyPort)
     lastProcessUp = true
+  } else {
+    lastProcessUp = false
   }
   return healthyPort
 }
@@ -493,10 +505,7 @@ async function maybeIdleStop(healthyPort) {
     http: Boolean(healthyPort)
   }))
   const killed = await stopComfyProcesses()
-  stoppedByAgent = true
+  markAsleep()
-  lastQueueRunning = 0
-  lastHealthyPort = 0
-  lastProcessUp = false
   console.log(JSON.stringify({ src: 'comfy-host-agent', event: 'stopped', reason: 'idle', killed }))
 }
 
@@ -617,14 +626,18 @@ const server = http.createServer(async (req, res) => {
   if (req.method === 'GET' && url.pathname === '/status') {
     // Answer immediately. MyMonitor allows ~500ms; probing ports + WMI here
     // made the A light stay red even while this process was running.
-    const healthyPort = lastHealthyPort || proxyTarget
+    // Do not treat a stale proxyTarget as proof Comfy is up — that made
-    const processUpNow = Boolean(lastProcessUp || healthyPort)
+    // /status report asleep+http at once, so training skipped /stop.
+    const healthyPort = lastHealthyPort || 0
+    const processUpNow = Boolean(lastProcessUp && healthyPort)
+    const asleep = !healthyPort && !processUpNow
+    if (!asleep) stoppedByAgent = false
     return json(res, 200, {
       ok: true,
       http: Boolean(healthyPort),
       process: processUpNow,
       processUp: processUpNow,
-      asleep: Boolean(stoppedByAgent || (!healthyPort && !processUpNow)),
+      asleep,
       lastActivityAt: new Date(lastWorkAt).toISOString(),
       idleMs,
       port: healthyPort || null,
@@ -647,11 +660,12 @@ const server = http.createServer(async (req, res) => {
       markWork()
       return json(res, 200, { ok: true, started: false, already: true, asleep: false, port: healthyPort, proxyPort })
     }
-    if (await pythonMainUp()) {
+    if (lastLaunchAt && Date.now() - lastLaunchAt < 60_000) {
       return json(res, 200, { ok: true, started: false, already: true, booting: true, asleep: false })
     }
     await taskkillImage('Comfy Desktop.exe')
     stoppedByAgent = false
+    lastLaunchAt = Date.now()
     const launched = await startComfy()
     if (launched.port) {
       lastProcessUp = true
@@ -671,10 +685,7 @@ const server = http.createServer(async (req, res) => {
       })
     }
     const killed = await stopComfyProcesses()
-    stoppedByAgent = true
+    markAsleep()
-    lastQueueRunning = 0
-    lastHealthyPort = 0
-    lastProcessUp = false
     console.log(JSON.stringify({ src: 'comfy-host-agent', event: 'stopped', reason: 'stop', killed }))
     return json(res, 200, { ok: true, stopped: true, asleep: true, killed })
   }

server/utils/comfyLifecycle.ts

@@ -154,7 +154,7 @@ export async function requestComfyWake() {
       statusMessage: 'COMFY_CONTROL_URL is not set. The studio cannot poke the Beast host agent.'
     })
   }
-  const remote = await controlRequest('/start', 'POST')
+  const remote = await controlRequest('/start', 'POST', 25_000)
   if (!remote) {
     throw createError({
       statusCode: 503,
@@ -263,7 +263,7 @@ async function startComfy() {
   }
   lastStartAt = now
 
-  const remote = await controlRequest('/start', 'POST')
+  const remote = await controlRequest('/start', 'POST', 25_000)
   throwIfTrainingBusy(remote)
   if (remote?.ok || remote?.started) {
     log('start-remote', { pid: remote.pid || null })

server/utils/imageComfyLifecycle.ts

@@ -88,7 +88,7 @@ export async function fetchImageQueue() {
   return fetchQueueAt(getImageComfyHost())
 }
 
-async function controlRequest(path: string, method = 'GET') {
+async function controlRequest(path: string, method = 'GET', timeoutMs = 5000) {
   const { controlUrl, controlToken } = settings()
   if (!controlUrl) return null
   try {
@@ -98,7 +98,7 @@ async function controlRequest(path: string, method = 'GET') {
         Accept: 'application/json',
         ...(controlToken ? { Authorization: `Bearer ${controlToken}` } : {})
       },
-      signal: AbortSignal.timeout(5000)
+      signal: AbortSignal.timeout(timeoutMs)
     })
     if (!res.ok) return { ok: false, status: res.status }
     return await res.json() as {
@@ -154,7 +154,7 @@ async function startImageComfy() {
     return { started: false, reason: 'recent-start' as const }
   }
   lastStartAt = now
-  const remote = await controlRequest('/start', 'POST')
+  const remote = await controlRequest('/start', 'POST', 25_000)
   if (remote?.ok || remote?.started) {
     log('start-remote', { port: remote.port || null })
     return { started: true, reason: 'remote' as const }