From e1bad7240598650d2f86f16dd76ba8d9a596e888 Mon Sep 17 00:00:00 2001
From: Towsty <ianjmcarr@gmail.com>
Date: Sat, 5 Sep 2026 11:47:10 -0500
Subject: [PATCH] Fix queue race: keep the running studio slot until settle
 finishes, then start the next waiting job.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 server/utils/studioQueue.ts | 41 +++++++++++++++++--------------------
 1 file changed, 19 insertions(+), 22 deletions(-)

diff --git a/server/utils/studioQueue.ts b/server/utils/studioQueue.ts
index 3de1c2c..be9218e 100644
--- a/server/utils/studioQueue.ts
+++ b/server/utils/studioQueue.ts
@@ -259,12 +259,13 @@ function jobAgeMs(job: { startedAt: number }) {
   return Date.now() - job.startedAt
 }
 
-/** Uploading, or created just now and not yet on Comfy. Paused leftovers do not count. */
+/** Uploading, or created and not yet on Comfy (includes wake / model load). */
 export function jobIsLocallySubmitting(job: Job) {
   if (job.library?.stopAfterCurrent) return false
   if (job.status === 'uploading') return true
-  if (job.status === 'queued' && !job.promptId && jobAgeMs(job) < SUBMIT_WINDOW_MS) return true
-  if (job.status === 'running' && !job.promptId && jobAgeMs(job) < SUBMIT_WINDOW_MS) return true
+  // Must cover ensureComfyReady's busy wait (up to ~3 min). The old 45s window
+  // let the dispatcher start the next queue item while the current one was still waking Comfy.
+  if ((job.status === 'queued' || job.status === 'running') && !job.promptId && jobAgeMs(job) < QUEUED_GRACE_MS) return true
   return false
 }
 
@@ -339,12 +340,11 @@ function liveJobOwnsGpu(job: Job) {
   if (job.library?.chainContinuing) return true
   if (job.saving) return true
   if (job.status === 'uploading') return true
-  if (job.status === 'running') {
-    // Without a prompt id past the submit window, this is a zombie — do not block the queue.
-    if (!job.promptId && jobAgeMs(job) >= SUBMIT_WINDOW_MS) return false
+  if (job.status === 'running' || job.status === 'queued') {
+    // Without a prompt id past the wake/submit grace, this is a zombie — do not block forever.
+    if (!job.promptId && jobAgeMs(job) >= QUEUED_GRACE_MS) return false
     return true
   }
-  if (job.status === 'queued') return Boolean(job.promptId) || jobAgeMs(job) < SUBMIT_WINDOW_MS
   return false
 }
 
@@ -397,16 +397,11 @@ export async function videoJobsBusy() {
   // An empty Comfy queue is not idle: the 3s buffer + last-frame extract between
   // shots leaves Comfy empty while the current video still owns the GPU.
   if (listJobs().some(liveJobOwnsGpu)) return true
-  // Studio "running" only blocks when a live job still owns the GPU, or a fresh
-  // pending file exists. Orphan disk rows used to block the whole queue forever.
-  if (listOwnersWithStudioQueues().some(owner => listStudioJobs(owner).some((job) => {
-    if (job.status !== 'running') return false
-    if (job.liveJobId) {
-      const live = getJob(job.liveJobId)
-      if (live && liveJobOwnsGpu(live)) return true
-    }
-    return pendingAlive(job)
-  }))) return true
+  // Studio row still "running" owns the queue slot until onLiveVideoSettled clears it.
+  // Do not peek under the live job here — that raced settle and started the next item early.
+  if (listOwnersWithStudioQueues().some(owner => listStudioJobs(owner).some(job => job.status === 'running'))) {
+    return true
+  }
   const queue = await fetchLiveQueue()
   if (queue) {
     if (queue.running > 0) return true
@@ -825,11 +820,9 @@ function pickHeld(jobs: StudioJob[]) {
 }
 
 function pickWaiting(jobs: StudioJob[]) {
-  const waiting = jobs.filter(job => job.status === 'waiting')
-  if (!waiting.length) return null
-  if (jobs.some(job => job.pausedByUser === true)) {
-    return waiting.reduce((latest, job) => job.createdAt >= latest.createdAt ? job : latest)
-  }
+  const waiting = jobs
+    .filter(job => job.status === 'waiting')
+    .sort((a, b) => a.createdAt - b.createdAt)
   return waiting[0] || null
 }
 
@@ -927,6 +920,9 @@ function repairStaleJobs(jobs: StudioJob[]) {
     }
     if (job.status !== 'running') continue
     const live = job.liveJobId ? getJob(job.liveJobId) : undefined
+    // Live already complete: onLiveVideoSettled owns finalization. Stealing the row here
+    // started the next queue item while purge/settle was still running → next jobs failed.
+    if (live && live.status === 'complete') continue
     const liveBusy = live && (live.status === 'queued' || live.status === 'uploading' || live.status === 'running')
     if (liveBusy || pendingAlive(job)) continue
     const liveFailed = live && (live.status === 'error' || live.status === 'cancelled')
@@ -940,6 +936,7 @@ function repairStaleJobs(jobs: StudioJob[]) {
       job.updatedAt = Date.now()
       continue
     }
+    // No live job and no fresh pending — orphaned running row.
     if (job.shotQueueId) {
       job.status = 'held'
       job.liveJobId = undefined