Let failed studio jobs retry instead of sitting in the queue with only Remove.

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

server/api/studio-queue/[id]/retry.post.ts

@@ -0,0 +1,7 @@
+import { retryStudioJob } from '~/server/utils/studioQueue'
+
+export default defineEventHandler(async (event) => {
+  const { owner } = assertLibraryOwner(event)
+  const id = String(getRouterParam(event, 'id') || '')
+  return await retryStudioJob(owner, id)
+})

server/utils/studioQueue.ts

@@ -492,6 +492,36 @@ export async function requestCutIn(owner: string, id: string) {
   return updated
 }
 
+export async function retryStudioJob(owner: string, id: string) {
+  const jobsNow = readJobs(owner)
+  const target = jobsNow.find(job => job.id === id)
+  if (!target) throw createError({ statusCode: 404, statusMessage: 'Queued job not found' })
+  if (target.status !== 'error' && target.status !== 'held' && target.status !== 'waiting') {
+    throw createError({ statusCode: 409, statusMessage: 'Only a failed or paused job can be retried' })
+  }
+  const updated = await mutate(owner, (jobs) => {
+    const job = jobs.find(item => item.id === id)
+    if (!job) throw createError({ statusCode: 404, statusMessage: 'Queued job not found' })
+    if (job.status !== 'error' && job.status !== 'held' && job.status !== 'waiting') {
+      throw createError({ statusCode: 409, statusMessage: 'Only a failed or paused job can be retried' })
+    }
+    job.status = 'waiting'
+    job.cutIn = true
+    job.lastError = undefined
+    job.liveJobId = undefined
+    job.pausedByUser = false
+    job.pauseAfterCurrent = false
+    job.holdForCutIn = false
+    job.updatedAt = Date.now()
+    for (const item of jobs) {
+      if (item.id !== id && item.status === 'waiting') item.cutIn = false
+    }
+    return structuredClone(job)
+  })
+  await kickStudioQueue()
+  return updated
+}
+
 function pruneDone(jobs: StudioJob[]) {
   const cutoff = Date.now() - 1000 * 60 * 60 * 24
   return jobs.filter(job => {
@@ -585,6 +615,10 @@ function pendingAlive(job: StudioJob) {
   ))
 }
 
+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 || ''))
+}
+
 function repairStaleJobs(jobs: StudioJob[]) {
   for (const job of jobs) {
     if (job.status === 'error' && /allowIdentityRefs is not defined/i.test(String(job.lastError || ''))) {
@@ -594,6 +628,13 @@ function repairStaleJobs(jobs: StudioJob[]) {
       job.updatedAt = Date.now()
       continue
     }
+    if (job.status === 'error' && isTransientComfyError(job.lastError)) {
+      job.status = 'waiting'
+      job.liveJobId = undefined
+      job.lastError = undefined
+      job.updatedAt = Date.now()
+      continue
+    }
     if (job.status !== 'running') continue
     const live = job.liveJobId ? getJob(job.liveJobId) : undefined
     const liveBusy = live && (live.status === 'queued' || live.status === 'uploading' || live.status === 'running')
@@ -1123,10 +1164,6 @@ function remainingStudioShots(job: Job) {
   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) {