Finish the current video shot list before starting the next queued video.

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

server/utils/studioQueue.ts

@@ -2,7 +2,7 @@ import { existsSync, mkdirSync, readdirSync, readFileSync, renameSync, writeFile
 import { join } from 'node:path'
 import { getJob, listJobs, emitJob, type Job } from '~/server/utils/jobs'
 import { listPendingJobs, patchPendingJob, readPendingJob, deletePendingJob } from '~/server/utils/pending'
-import { getShotQueue } from '~/server/utils/shotQueue'
+import { getShotQueue, pendingSegmentCount } from '~/server/utils/shotQueue'
 import { fetchLiveQueue } from '~/server/utils/comfy'
 import { isLtxWorkflow, isTextToVideo, isXaigenStudio, LTX_DISABLED_MESSAGE, ltxWorkflowEnabled, parseVideoWorkflow, type VideoWorkflowId } from '~/utils/videoModels'
 import { persistLoraFields } from '~/server/utils/loras'
@@ -262,17 +262,24 @@ function sweepStaleLiveJobs() {
   }
 }
 
+function liveJobOwnsGpu(job: Job) {
+  if (job.library?.stopAfterCurrent) return false
+  if (job.library?.chainContinuing) return true
+  if (job.status === 'uploading' || job.status === 'running') return true
+  if (job.status === 'queued') return Boolean(job.promptId) || jobAgeMs(job) < SUBMIT_WINDOW_MS
+  return false
+}
+
 export async function videoJobsBusy() {
   sweepStaleLiveJobs()
   if (listJobs().some(jobIsLocallySubmitting)) return true
+  // 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
+  if (listOwnersWithStudioQueues().some(owner => listStudioJobs(owner).some(job => job.status === 'running'))) return true
   const queue = await fetchLiveQueue()
   if (queue) return queue.running > 0 || queue.pending > 0
-  return listJobs().some((job) => {
-    if (job.library?.stopAfterCurrent) return false
-    if (job.status === 'uploading' || job.status === 'running') return true
-    if (job.status === 'queued') return jobAgeMs(job) < SUBMIT_WINDOW_MS
-    return false
-  }) || listPendingJobs().some((pending) => {
+  return listPendingJobs().some((pending) => {
     if (!pending.promptId || pending.stopAfterCurrent) return false
     const live = getJob(pending.jobId)
     if (live) return jobIsLocallySubmitting(live) || live.status === 'running' || live.status === 'uploading'
@@ -525,10 +532,10 @@ function pickHeld(jobs: StudioJob[]) {
     .filter(job => (
       job.status === 'held'
       && job.shotQueueId
-      && job.holdForCutIn
       && !job.pausedByUser
+      && (job.holdForCutIn === true || job.resumeAutoRun === true)
     ))
-    .sort((a, b) => b.updatedAt - a.updatedAt)[0] || null
+    .sort((a, b) => a.createdAt - b.createdAt)[0] || null
 }
 
 function pickWaiting(jobs: StudioJob[]) {
@@ -628,16 +635,17 @@ async function dispatchStudioQueue() {
       await startStudioJob(cutIn)
       return
     }
+    const held = pickHeld(store.jobs)
+    if (held?.shotQueueId) {
+      const resumed = await resumeHeldStudioJob(held)
+      if (!resumed) scheduleKickRetry()
+      return
+    }
     const waiting = pickWaiting(store.jobs)
     if (waiting) {
       await startStudioJob(waiting)
       return
     }
-    const held = pickHeld(store.jobs)
-    if (held?.shotQueueId) {
-      const resumed = await resumeHeldStudioJob(held)
-      if (resumed) return
-    }
   }
 }
 
@@ -649,6 +657,10 @@ async function resumeHeldStudioJob(item: StudioJob) {
     })
     return false
   }
+  await patchStudioJob(item.ownerKey, item.id, (job) => {
+    job.status = 'running'
+    job.lastError = undefined
+  })
   const { startQueueBurst } = await import('~/server/utils/videoChain')
   const count = item.resumeAutoRun ? 'all' as const : 1
   try {
@@ -662,10 +674,12 @@ async function resumeHeldStudioJob(item: StudioJob) {
     return true
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error)
+    if (/already processing/i.test(message)) return true
+    const transient = /offline|never answered|host agent|unreachable|asleep|Starting Comfy|ComfyUI \(1\)|instance picker/i.test(message)
     await patchStudioJob(item.ownerKey, item.id, (job) => {
-      job.status = 'error'
+      job.status = transient ? 'held' : 'error'
       job.lastError = message
-      job.holdForCutIn = false
+      job.holdForCutIn = transient
     })
     return false
   }
@@ -1031,16 +1045,29 @@ export async function startStudioJob(item: StudioJob) {
   }
 }
 
+function remainingStudioShots(job: Job) {
+  const library = job.library
+  if (!library) return 0
+  if (library.queueId) {
+    const queue = getShotQueue(library.ownerKey, library.queueId)
+    if (queue) return pendingSegmentCount(queue)
+  }
+  return Math.max(0, (library.chainTotal || 1) - (library.chainStep || 1))
+}
+
+function isTransientComfyError(error?: string) {
+  return /still busy|COMFY_BUSY|offline|never answered|host agent|unreachable|asleep|Starting Comfy|ComfyUI \(1\)|instance picker/i.test(String(error || ''))
+}
+
 export async function onLiveVideoSettled(job: Job) {
   const owner = job.library?.ownerKey
   if (!owner) {
     kickStudioQueue()
     return
   }
-  const remaining = job.library
-    ? Math.max(0, (job.library.chainTotal || 1) - (job.library.chainStep || 1))
-    : 0
-  const failed = job.status === 'error' || job.status === 'cancelled'
+  const remaining = remainingStudioShots(job)
+  const wakeFail = job.status === 'error' && remaining > 0 && isTransientComfyError(job.error)
+  const failed = (job.status === 'error' || job.status === 'cancelled') && !wakeFail
   const snapshot = readStore(owner)
   const rowNow = snapshot.jobs.find(item => item.liveJobId === job.id)
     || (job.library?.queueId
@@ -1085,6 +1112,25 @@ export async function onLiveVideoSettled(job: Job) {
       row.resumeAutoRun = row.payload.queueAutoRun === true || job.library?.queueAutoRun === true || row.resumeAutoRun === true
       row.updatedAt = Date.now()
     })
+  } else if (!failed && remaining > 0) {
+    await mutateStore(owner, (store) => {
+      const row = store.jobs.find(item => item.liveJobId === job.id)
+        || (job.library?.queueId ? store.jobs.find(item => item.shotQueueId === job.library?.queueId) : undefined)
+      if (!row) {
+        syncPausedFlag(store)
+        return
+      }
+      const auto = row.payload.queueAutoRun === true || job.library?.queueAutoRun === true || row.resumeAutoRun === true
+      row.status = 'held'
+      row.liveJobId = undefined
+      row.holdForCutIn = auto
+      row.pauseAfterCurrent = false
+      row.pausedByUser = !auto
+      row.resumeAutoRun = auto
+      row.lastError = wakeFail ? job.error : undefined
+      row.updatedAt = Date.now()
+      syncPausedFlag(store)
+    })
   } else if (userPause) {
     await mutateStore(owner, (store) => {
       const row = store.jobs.find(item => item.liveJobId === job.id)