feat: retry lost jobs on startup

app/shared/celery.py

@@ -4,6 +4,6 @@ from app.shared.config import settings
 
 
 def get_celery_binding() -> Celery:
-    celery = Celery("tasks")
+    celery = Celery()
     celery.conf.broker_url = settings.BROKER_URL
     return celery

app/web/main.py

@@ -1,8 +1,10 @@
+from asyncio.log import logger
 from typing import Dict, List, Optional
 from uuid import UUID
 
 from fastapi import APIRouter, Depends, FastAPI, HTTPException, Path
 from pydantic import AnyHttpUrl, BaseModel
+from sqlalchemy import or_
 from sqlalchemy.orm import Session
 
 import app.shared.db.dtos as dtos
@@ -17,6 +19,15 @@ celery = get_celery_binding()
 api_router = APIRouter(prefix="/api/v1", dependencies=[Depends(authenticate_api_key)])
 
 
+def queue_task(job: models.Job) -> None:
+    # queue an async transcription task.
+    # we use a signature here to allow full separation of
+    # worker processes and dependencies.
+    transcribe = celery.signature("app.worker.main.transcribe")
+    # TODO: catch delivery errors.
+    transcribe.delay(job.id)
+
+
 @api_router.get("/")
 def api_root() -> Dict:
     return {}
@@ -99,3 +110,23 @@ def delete_transcript(
 
 
 app.include_router(api_router)
+
+# TODO:
+# we could use `acks_late` to handle this scenario within celery itself.
+# the reason this does not work well in our case is that `visibility_timeout`
+# needs to be very high since whisper workers can be long running.
+# doing this application-side bears the risk of poison pilling the worker though,
+# implement a workaround with an acceptable trade-off. (=> retry only once?)
+@app.on_event("startup")
+def on_startup() -> None:
+    session = get_session().__next__()
+
+    jobs = (
+        session.query(models.Job)
+            .filter(or_(models.Job.status == dtos.JobStatus.processing, models.Job.status == dtos.JobStatus.create))
+            .order_by(models.Job.created_at)
+    ).all()
+
+    logger.info(f"Re-queueing {len(jobs)} jobs.")
+    for job in jobs:
+        queue_task(job)

app/web/start.sh

@@ -6,4 +6,4 @@ set -e
 alembic upgrade head
 
 # start app
-uvicorn app.web.main:app --reload --host ${HOST:-0.0.0.0} --port ${PORT:-80}
+uvicorn app.web.main:app --reload --host ${HOST:-0.0.0.0} --port ${PORT:-80} --log-level info

app/worker/main.py

@@ -1,6 +1,7 @@
 from asyncio.log import logger
 from uuid import UUID
 
+from celery import Task
 from sqlalchemy.orm import Session
 
 import app.shared.db.dtos as dtos
@@ -12,18 +13,24 @@ from app.worker.strategies.local import LocalStrategy
 celery = get_celery_binding()
 
 
-def update_job_status(db: Session, job: models.Job, status: dtos.JobStatus) -> None:
-    job.status = status
-    db.commit()
-
-
-@celery.task()
-def transcribe(job_id: UUID) -> None:
+@celery.task(
+    bind=True,
+    soft_time_limit=2 * 60 * 60 # TODO: make configurable
+)
+def transcribe(self: Task, job_id: UUID) -> None:
     try:
         db: Session = SessionLocal()
         job = db.query(models.Job).filter(models.Job.id == job_id).one()
 
-        update_job_status(db, job, dtos.JobStatus.processing)
+        if job.status == dtos.JobStatus.error or job.status == dtos.JobStatus.success:
+            logger.warn("[{job.id}]: Received job that has already been processed, abort.")
+            return
+
+        job.meta = {"task_id": self.request.id}
+        job.status = dtos.JobStatus.processing
+        db.commit()
+
+        logger.info(f"[{job.id}]: set task to status processing.")
 
         # pick a transcription strategy.
         # currently only `local` is supported.
@@ -36,20 +43,30 @@ def transcribe(job_id: UUID) -> None:
         # currently only `transcribe` is supported.
         if job.type == dtos.JobType.transcript:
             result = strategy.transcribe()
+            logger.info(f"[{job.id}]: successfully transcribed audio.")
         elif job.type == dtos.JobType.translation:
             result = strategy.translate()
+            logger.info(f"[{job.id}]: successfully translated audio.")
         else:
             result = strategy.detect_language()
 
         artifact = models.Artifact(
             job_id=job.id, data=result, type=dtos.ArtifactType.raw_transcript
         )
+
         db.add(artifact)
         db.commit()
+        logger.info(f"[{job.id}]: successfully stored artifact.")
 
-        update_job_status(db, job, dtos.JobStatus.success)
+        job.status = dtos.JobStatus.success
+        db.commit()
+
+        logger.info(f"[{job.id}]: set task to status success.")
     except Exception as e:
-        logger.error(e)
-        update_job_status(db, job, dtos.JobStatus.error)
+        if job and db:
+            job.meta = { **job.meta, "error": str(e) }
+            job.status = dtos.JobStatus.error
+            db.commit()
+            raise(e)
     finally:
         db.close()

app/worker/strategies/local.py

@@ -1,6 +1,6 @@
 import os
-import shutil
 import tempfile
+from asyncio.log import logger
 from os import path
 from typing import Any, List, Literal, Optional
 from uuid import UUID
@@ -26,27 +26,21 @@ class LocalStrategy:
         self.job_id = job_id
         self.url = url
         self.config = config
+        logger.info(f"[{self.job_id}]: initialized local strategy.")
 
     def transcribe(self) -> List[Any]:
-        result = self.run_whisper(self._download(), "transcribe")
-        self._cleanup()
-        return result
+        return self.run_whisper(self._download(), "transcribe")
 
     def translate(self) -> List[Any]:
-        result = self.run_whisper(self._download(), "translate")
-        self._cleanup()
-        return result
+        return self.run_whisper(self._download(), "translate")
 
     def detect_language(self) -> List[Any]:
         raise NotImplementedError("detect_language has not been implemented yet.")
 
     def _download(self) -> str:
-        dirname = self._get_tmp_dir()
-        filename = path.join(dirname, "media.mp3")
-
         # re-create folder.
-        shutil.rmtree(dirname, ignore_errors=True)
-        os.makedirs(dirname)
+        filename = self._get_tmp_file()
+        self._cleanup()
 
         # stream media to disk.
         with requests.get(self.url, stream=True) as r:
@@ -58,21 +52,29 @@ class LocalStrategy:
         return filename
 
     def run_whisper(self, filepath: str, task: str) -> List[Any]:
-        language = self.config.language if self.config else None
-        decode_opts = DecodeOptions(task=task, language=language)
-        model = load_model("small", download_root="/models")
+        try:
+            language = self.config.language if self.config else None
+            model = load_model("small", download_root="/models")
 
-        result = model.transcribe(
-            filepath, condition_on_previous_text=False, **decode_opts.dict()
-        )
+            result = model.transcribe(
+                filepath,
+                condition_on_previous_text=False,
+                **DecodeOptions(task=task, language=language).dict(),
+            )
 
-        return result["segments"]
+            return result["segments"]
+        finally:
+            self._cleanup()
 
-    def _get_tmp_dir(self) -> str:
-        return path.join(tempfile.gettempdir(), str(self.job_id))
+    def _get_tmp_file(self) -> str:
+        tmp = tempfile.gettempdir()
+        return path.join(tmp, str(self.job_id))
 
     def _cleanup(self) -> None:
-        shutil.rmtree(self._get_tmp_dir(), ignore_errors=True)
+        try:
+            os.remove(self._get_tmp_file())
+        except OSError:
+            pass
 
     def _convert(self) -> None:
         pass

docker/dev.docker-compose.yml

@@ -29,13 +29,10 @@ services:
   redis:
     container_name: whisperbox_redis
     image: redis:7-alpine
-    command: ["redis-server", "--save", "60 1"]
     ports:
       - 6379:6379
     networks:
       - app
-    volumes:
-      - redis-data:/data
 
   app:
     container_name: whisperbox_app
@@ -50,6 +47,8 @@ services:
     volumes:
       - ../:/code
     depends_on:
+      postgres:
+        condition: service_healthy
       postgres:
         condition: service_healthy
       redis:
@@ -64,10 +63,14 @@ services:
       - ../:/code
     environment: *app-variables
     depends_on:
-      - app
       - redis
     networks:
       - app
+    healthcheck:
+      test: ["CMD-SHELL", "celery inspect ping -A app.worker.main.transcribe -d celery@$$HOSTNAME"]
+      interval: 5s
+      timeout: 5s
+      retries: 5
 
   flower:
     container_name: whisperbox_flower
@@ -82,7 +85,6 @@ services:
 
 volumes:
   postgres-data:
-  redis-data:
 
 networks:
   app: