seaislee1209 60713ea009
Some checks failed
Build and Deploy / build-and-deploy (push) Has been cancelled
feat: v0.14.0 后端异步轮询(Celery + Redis)
- Celery 异步任务:任务提交后后端持续轮询火山 API 直到拿到终态,用户关浏览器也不会丢视频
- 渐进式轮询:前2分钟每5秒、2-10分钟每15秒、10分钟后每30秒
- 优雅降级:无 Redis 时静默跳过,不影响现有前端轮询
- K8s:新增 Redis Deployment + Service、Celery Worker Deployment
- CI/CD:deploy.yaml 自动部署 Redis/Celery,每次推代码自动重启 celery worker
- 兜底:poll_stuck_tasks management command 清理僵尸任务

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-03-25 13:27:16 +08:00

109 lines
4.4 KiB
Python

"""Management command to poll stuck tasks and update their status.
This is a fallback for when Celery workers miss tasks or aren't running.
Run via cron or K8s CronJob: python manage.py poll_stuck_tasks
"""
import logging
from django.core.management.base import BaseCommand
from django.utils import timezone
from apps.generation.models import GenerationRecord
from utils.airdrama_client import query_task, map_status, extract_video_url, ERROR_MESSAGES
logger = logging.getLogger(__name__)
class Command(BaseCommand):
help = 'Poll Volcano API for stuck queued/processing tasks and update their status.'
def handle(self, *args, **options):
stuck = GenerationRecord.objects.filter(status__in=['queued', 'processing'])
count = stuck.count()
if count == 0:
self.stdout.write('No stuck tasks found.')
return
self.stdout.write(f'Found {count} stuck task(s), polling...')
resolved = 0
for record in stuck:
ark_task_id = record.ark_task_id
# No ark_task_id means API submission failed — mark as failed
if not ark_task_id:
record.status = 'failed'
record.error_message = '任务提交失败(系统清理)'
record.completed_at = timezone.now()
record.save(update_fields=['status', 'error_message', 'completed_at'])
from apps.generation.views import _release_freeze
_release_freeze(record)
resolved += 1
self.stdout.write(f' [{record.id}] no ark_task_id -> marked failed')
continue
# Poll Volcano API
try:
ark_resp = query_task(ark_task_id)
new_status = map_status(ark_resp.get('status', ''))
except Exception as e:
self.stdout.write(f' [{record.id}] ark={ark_task_id} API error: {e}')
continue
if new_status in ('queued', 'processing'):
self.stdout.write(f' [{record.id}] ark={ark_task_id} still {new_status}')
continue
# Terminal state — process
record.status = new_status
returned_seed = ark_resp.get('seed')
if returned_seed is not None:
record.seed = returned_seed
if new_status == 'completed':
video_url = extract_video_url(ark_resp)
if video_url:
try:
from utils.tos_client import upload_from_url
record.result_url = upload_from_url(video_url, folder='results')
except Exception:
logger.exception('Failed to persist video to TOS')
record.result_url = video_url
usage = ark_resp.get('usage', {})
total_tokens = usage.get('total_tokens', 0) if isinstance(usage, dict) else 0
if total_tokens > 0:
from apps.generation.views import _settle_payment
_settle_payment(record, total_tokens)
else:
from apps.generation.views import _release_freeze
_release_freeze(record)
elif new_status == 'failed':
error = ark_resp.get('error', {})
code = error.get('code', '') if isinstance(error, dict) else ''
raw_msg = error.get('message', '') if isinstance(error, dict) else str(error)
record.error_message = ERROR_MESSAGES.get(code, raw_msg)
record.raw_error = f'{code}: {raw_msg}' if code else raw_msg
usage = ark_resp.get('usage', {})
total_tokens = usage.get('total_tokens', 0) if isinstance(usage, dict) else 0
if total_tokens > 0:
from apps.generation.views import _settle_payment
_settle_payment(record, total_tokens)
else:
from apps.generation.views import _release_freeze
_release_freeze(record)
record.completed_at = timezone.now()
record.save(update_fields=[
'status', 'result_url', 'error_message', 'raw_error',
'seed', 'completed_at',
])
resolved += 1
self.stdout.write(f' [{record.id}] ark={ark_task_id} -> {new_status}')
self.stdout.write(f'Done. Resolved {resolved}/{count} tasks.')