Keep the job thread alive after an error of the queue

An error outside a job, such as a full disk, stopped the thread, and no
job ran until a restart. The health check now answers 503 when a job
thread stopped.

README.md

@@ -426,6 +426,9 @@ After a restart, the waiting jobs run. A job that was running starts again at th
 queue, and skips the chapters that it stored. If Bindery stops during the same job three times, the
 job fails.
 
+If a job thread stops, `/api/health` answers 503, so the container shows as unhealthy. Restart
+Bindery to start the thread again.
+
 ### Backups
 
 Once a day, Bindery copies its database and `personal.ini` to `/config/backups`, and keeps the

bindery/orchestrator.py

@@ -1290,25 +1290,38 @@ def _run_job(job: Job) -> str:
     return _run_site_job(job)
 
 
+def _run_next_job() -> Job | None:
+    """Run the job at the front of the queue. Returns the job, or None if no job waits."""
+    job = job_queue.get_next()
+    if not job:
+        return None
+    job_queue.update(job, status=QueueStatus.RUNNING, message="Starting")
+    try:
+        message = _run_job(job)
+        job_queue.update(job, status=QueueStatus.DONE, message=message, progress=1.0)
+        logger.info(f"Job {job.id} ({job.mode.value}) for {job.title}: {message}")
+    except (StoryError, CwaError, EpubError, RuleError, KeyError, ValueError) as e:
+        _fail(job, str(e))
+    except Exception as e:
+        logger.exception(f"Job {job.id} failed: {e}")
+        _fail(job, f"{type(e).__name__}: {e}")
+    return job
+
+
 def download_loop() -> None:
     """Background thread for processing the job queue."""
     logger.info("Starting download loop")
     while True:
-        job = job_queue.get_next()
-        if not job:
-            time.sleep(2)
-            continue
-        job_queue.update(job, status=QueueStatus.RUNNING, message="Starting")
         try:
-            message = _run_job(job)
-            job_queue.update(job, status=QueueStatus.DONE, message=message, progress=1.0)
-            logger.info(f"Job {job.id} ({job.mode.value}) for {job.title}: {message}")
-        except (StoryError, CwaError, EpubError, RuleError, KeyError, ValueError) as e:
-            _fail(job, str(e))
+            job = _run_next_job()
         except Exception as e:
-            logger.exception(f"Job {job.id} failed: {e}")
-            _fail(job, f"{type(e).__name__}: {e}")
-        if job.mode in SITE_MODES:
+            # The queue or the store failed, for example on a full disk. The loop goes on, because no job runs
+            # after it stops.
+            logger.exception(f"Job loop error: {e}")
+            job = None
+        if job is None:
+            time.sleep(2)
+        elif job.mode in SITE_MODES:
             time.sleep(config.SITE_DELAY)
 
 
@@ -1354,10 +1367,20 @@ def scheduler_loop() -> None:
         time.sleep(60)
 
 
+_threads: list[threading.Thread] = []
+
+
 def start_threads() -> None:
     try:
         adopt_build_keys()
     except Exception as e:
         logger.exception(f"Cannot check the built books: {e}")
     for target in (download_loop, scheduler_loop):
-        threading.Thread(target=target, daemon=True, name=target.__name__).start()
+        thread = threading.Thread(target=target, daemon=True, name=target.__name__)
+        thread.start()
+        _threads.append(thread)
+
+
+def stopped_threads() -> list[str]:
+    """The names of the job threads that stopped. Bindery must restart to start them again."""
+    return [t.name for t in _threads if not t.is_alive()]

bindery/web.py

@@ -201,6 +201,9 @@ def server_error(e: Exception) -> Any:
 
 @app.get("/api/health")
 def health() -> Any:
+    stopped = orchestrator.stopped_threads()
+    if stopped:
+        return {"status": "error", "error": f"Stopped: {', '.join(stopped)}. Restart Bindery."}, 503
     return {"status": "ok"}
 
 

tests/test_worker_flows.py

@@ -6,7 +6,7 @@ from unittest.mock import Mock
 
 import pytest
 
-from bindery import config, fff
+from bindery import config, fff, web
 from bindery import orchestrator as o
 from bindery.diffing import text_hash
 from bindery.jobs import DownloadMode, Job, JobQueue, QueueStatus
@@ -87,6 +87,42 @@ def test_worker_persists_outcomes_and_continues_after_failure(state, monkeypatch
     assert [c.args[0] for c in sleeps.call_args_list] == [2, 7]
 
 
+def test_worker_goes_on_after_an_error_of_the_queue(state, monkeypatch, caplog):
+    store, queue = state
+    job = queue.add(Job(mode=DownloadMode.REBUILD))
+    monkeypatch.setattr(queue, "get_next", Mock(side_effect=[OSError("disk full"), job, StopLoop()]))
+    monkeypatch.setattr(queue, "update", Mock(side_effect=[OSError("disk full"), None, None]))
+    sleeps = Mock()
+    monkeypatch.setattr(o.time, "sleep", sleeps)
+    with pytest.raises(StopLoop):
+        o.download_loop()
+    assert caplog.text.count("Job loop error: disk full") == 2
+    assert [c.args[0] for c in sleeps.call_args_list] == [2, 2]
+
+
+def test_health_reports_a_stopped_job_thread(state, monkeypatch):
+    client = web.app.test_client()
+    assert o.stopped_threads() == []
+    alive, dead = Mock(), Mock()
+    alive.is_alive.return_value, dead.is_alive.return_value, dead.name = True, False, "download_loop"
+    monkeypatch.setattr(o, "_threads", [alive, dead])
+    response = client.get("/api/health")
+    assert response.status_code == 503
+    assert response.json == {"status": "error", "error": "Stopped: download_loop. Restart Bindery."}
+    monkeypatch.setattr(o, "_threads", [alive])
+    assert client.get("/api/health").json == {"status": "ok"}
+
+
+def test_start_threads_keeps_the_threads_for_the_health_check(state, monkeypatch):
+    monkeypatch.setattr(o, "_threads", [])
+    monkeypatch.setattr(o, "download_loop", lambda: None)
+    monkeypatch.setattr(o, "scheduler_loop", lambda: None)
+    o.start_threads()
+    for thread in o._threads:
+        thread.join()
+    assert o.stopped_threads() == ["<lambda>", "<lambda>"]
+
+
 def test_scheduler_skips_disabled_busy_and_recent_stories(state, monkeypatch):
     store, queue = state
     now = datetime(2026, 6, 1, 12)