Stabilize worker sync flow

This commit is contained in:
qananasikq
2026-04-17 17:30:28 +03:00
parent 3c3aea8eb3
commit 05d1ffb5ac
9 changed files with 803 additions and 80 deletions

View File

@@ -150,7 +150,11 @@ class TestWorkerTaskLockHelpers(unittest.TestCase):
# Roundtrip: save → load.
redis_client = MagicMock()
storage: dict[str, str] = {}
redis_client.set.side_effect = lambda key, value: storage.__setitem__(key, value)
def _fake_set(key, value, ex=None):
storage[key] = value
return True
redis_client.set.side_effect = _fake_set
redis_client.get.side_effect = lambda key: storage.get(key)
tasks._save_sync_checkpoint(
@@ -165,7 +169,7 @@ class TestWorkerTaskLockHelpers(unittest.TestCase):
with patch.object(tasks, "_get_persistence") as get_persistence, \
patch.object(tasks, "_get_redis") as get_redis, \
patch.object(tasks, "_acquire_lock", return_value=True), \
patch.object(tasks, "_is_full_scan_done", return_value=True), \
patch.object(tasks, "_is_full_scan_done", return_value=False), \
patch.object(tasks, "_start_lock_heartbeat") as start_heartbeat, \
patch.object(tasks, "_release_lock_if_owner") as release_lock, \
patch.object(tasks, "_clear_sync_checkpoint") as clear_checkpoint, \
@@ -204,6 +208,311 @@ class TestWorkerTaskLockHelpers(unittest.TestCase):
clear_checkpoint.assert_called()
release_lock.assert_called_once()
def test_sync_listing_ignores_checkpoint_after_full_scan_completed(self) -> None:
with patch.object(tasks, "_get_persistence") as get_persistence, \
patch.object(tasks, "_get_redis") as get_redis, \
patch.object(tasks, "_acquire_lock", return_value=True), \
patch.object(tasks, "_is_full_scan_done", return_value=True), \
patch.object(tasks, "_start_lock_heartbeat") as start_heartbeat, \
patch.object(tasks, "_release_lock_if_owner") as release_lock, \
patch.object(tasks, "_clear_sync_checkpoint") as clear_checkpoint, \
patch.object(tasks, "_run_browser_job", side_effect=lambda fn: fn()), \
patch.object(tasks.sync_listing_task, "update_state"), \
patch("iaai_scraper.worker.tasks.parse_listing_segments", return_value=[]):
get_persistence.return_value = MagicMock()
redis_client = MagicMock()
redis_client.get.side_effect = lambda key: json.dumps({
"status": "in_progress", "last_successful_page": 25,
"make": None, "model": None, "lane": "iaai_cars",
}) if key == tasks.SYNC_LISTING_CHECKPOINT_KEY else None
get_redis.return_value = redis_client
stop_event = MagicMock()
heartbeat_thread = MagicMock()
start_heartbeat.return_value = (stop_event, heartbeat_thread)
sync_listing_mock = MagicMock(return_value={
"run_id": 14, "status": "success", "full_scan_completed": True,
"cars_upserted": 1, "cars_failed": 0, "images_upserted": 0,
"skipped_existing": 0, "elapsed_seconds": 1.0, "failures": [],
})
scraper_ctx = MagicMock()
scraper_ctx.__enter__.return_value.sync_listing = sync_listing_mock
scraper_ctx.__exit__.return_value = None
with patch.object(tasks, "IAAIScraper", return_value=scraper_ctx):
tasks.sync_listing_task.push_request(id="task-792")
try:
result = tasks.sync_listing_task.run()
finally:
tasks.sync_listing_task.pop_request()
self.assertEqual(result["status"], "success")
self.assertEqual(sync_listing_mock.call_args.kwargs["start_page"], 1)
self.assertIsNone(sync_listing_mock.call_args.kwargs["progress_callback"])
clear_checkpoint.assert_called()
release_lock.assert_called_once()
def test_sync_listing_checkpoint_zero_page_restarts_from_first_page(self) -> None:
with patch.object(tasks, "_get_persistence") as get_persistence, \
patch.object(tasks, "_get_redis") as get_redis, \
patch.object(tasks, "_acquire_lock", return_value=True), \
patch.object(tasks, "_is_full_scan_done", return_value=True), \
patch.object(tasks, "_start_lock_heartbeat") as start_heartbeat, \
patch.object(tasks, "_release_lock_if_owner") as release_lock, \
patch.object(tasks, "_run_browser_job", side_effect=lambda fn: fn()), \
patch.object(tasks.sync_listing_task, "update_state"), \
patch("iaai_scraper.worker.tasks.parse_listing_segments", return_value=[]):
get_persistence.return_value = MagicMock()
redis_client = MagicMock()
redis_client.get.side_effect = lambda key: json.dumps({
"status": "in_progress", "last_successful_page": 0,
"make": None, "model": None, "lane": "iaai_cars",
}) if key == tasks.SYNC_LISTING_CHECKPOINT_KEY else None
get_redis.return_value = redis_client
stop_event = MagicMock()
heartbeat_thread = MagicMock()
start_heartbeat.return_value = (stop_event, heartbeat_thread)
sync_listing_mock = MagicMock(return_value={
"run_id": 12, "status": "success", "full_scan_completed": True,
"cars_upserted": 1, "cars_failed": 0, "images_upserted": 0,
"skipped_existing": 0, "elapsed_seconds": 1.0, "failures": [],
})
scraper_ctx = MagicMock()
scraper_ctx.__enter__.return_value.sync_listing = sync_listing_mock
scraper_ctx.__exit__.return_value = None
with patch.object(tasks, "IAAIScraper", return_value=scraper_ctx):
tasks.sync_listing_task.push_request(id="task-790")
try:
result = tasks.sync_listing_task.run()
finally:
tasks.sync_listing_task.pop_request()
self.assertEqual(result["status"], "success")
self.assertEqual(sync_listing_mock.call_args.kwargs["start_page"], 1)
release_lock.assert_called_once()
def test_save_sync_checkpoint_sets_ttl(self) -> None:
redis_client = MagicMock()
tasks._save_sync_checkpoint(
redis_client,
task_id="task-1",
page_number=3,
make=None,
model=None,
lane="iaai_cars",
)
redis_client.set.assert_called_once()
self.assertEqual(redis_client.set.call_args.kwargs["ex"], tasks.SYNC_LISTING_CHECKPOINT_TTL_SECONDS)
def test_load_sync_checkpoint_clears_invalid_payload(self) -> None:
redis_client = MagicMock()
redis_client.get.return_value = "{broken-json"
checkpoint = tasks._load_sync_checkpoint(redis_client)
self.assertIsNone(checkpoint)
redis_client.delete.assert_called_once_with(tasks.SYNC_LISTING_CHECKPOINT_KEY)
def test_sync_listing_task_clears_checkpoint_on_partial_success_when_full_scan_completed(self) -> None:
with patch.object(tasks, "_get_persistence") as get_persistence, \
patch.object(tasks, "_get_redis") as get_redis, \
patch.object(tasks, "_acquire_lock", return_value=True), \
patch.object(tasks, "_is_full_scan_done", return_value=True), \
patch.object(tasks, "_start_lock_heartbeat") as start_heartbeat, \
patch.object(tasks, "_release_lock_if_owner") as release_lock, \
patch.object(tasks, "_clear_sync_checkpoint") as clear_checkpoint, \
patch.object(tasks, "_run_browser_job", return_value={
"run_id": 13,
"status": "partial_success",
"full_scan_completed": True,
"cars_upserted": 2,
"cars_failed": 1,
"images_upserted": 3,
"skipped_existing": 0,
"elapsed_seconds": 4.0,
"failures": [{"vehicle_url": "v", "error": "e"}],
}), \
patch.object(tasks.sync_listing_task, "update_state"):
get_persistence.return_value = MagicMock()
redis_client = MagicMock()
redis_client.get.return_value = None
get_redis.return_value = redis_client
stop_event = MagicMock()
heartbeat_thread = MagicMock()
start_heartbeat.return_value = (stop_event, heartbeat_thread)
tasks.sync_listing_task.push_request(id="task-791")
try:
result = tasks.sync_listing_task.run(make="Toyota")
finally:
tasks.sync_listing_task.pop_request()
self.assertEqual(result["status"], "partial_success")
clear_checkpoint.assert_called_once_with(redis_client)
release_lock.assert_called_once()
def test_try_set_followup_pending_deduplicates(self) -> None:
redis_client = MagicMock()
redis_client.set.side_effect = [True, False]
self.assertTrue(tasks._try_set_followup_pending(redis_client, ttl_seconds=120))
self.assertFalse(tasks._try_set_followup_pending(redis_client, ttl_seconds=120))
def test_bump_bootstrap_failure_streak_opens_circuit_breaker(self) -> None:
redis_client = MagicMock()
redis_client.incr.return_value = tasks.SYNC_LISTING_BOOTSTRAP_FAILURE_STREAK_LIMIT
streak, should_enqueue = tasks._bump_bootstrap_failure_streak(
redis_client,
reason="max_retries_exceeded",
)
self.assertEqual(streak, tasks.SYNC_LISTING_BOOTSTRAP_FAILURE_STREAK_LIMIT)
self.assertFalse(should_enqueue)
redis_client.expire.assert_called_once_with(
tasks.SYNC_LISTING_BOOTSTRAP_FAILURE_STREAK_KEY,
tasks.SYNC_LISTING_BOOTSTRAP_FAILURE_STREAK_TTL_SECONDS,
)
def test_bump_bootstrap_failure_streak_allows_retry_before_limit(self) -> None:
redis_client = MagicMock()
redis_client.incr.return_value = 1
streak, should_enqueue = tasks._bump_bootstrap_failure_streak(
redis_client,
reason="bootstrap_not_completed",
)
self.assertEqual(streak, 1)
self.assertTrue(should_enqueue)
def test_sync_listing_task_does_not_enqueue_followup_after_bootstrap_error_limit(self) -> None:
with patch.object(tasks, "_get_persistence") as get_persistence, \
patch.object(tasks, "_get_redis") as get_redis, \
patch.object(tasks, "_acquire_lock", return_value=True), \
patch.object(tasks, "_is_full_scan_done", return_value=False), \
patch.object(tasks, "_start_lock_heartbeat") as start_heartbeat, \
patch.object(tasks, "_release_lock_if_owner") as release_lock, \
patch.object(tasks, "_try_set_followup_pending", return_value=True), \
patch.object(tasks, "_bump_bootstrap_failure_streak", return_value=(tasks.SYNC_LISTING_BOOTSTRAP_FAILURE_STREAK_LIMIT, False)) as bump_streak, \
patch.object(tasks, "_clear_followup_pending") as clear_pending, \
patch.object(tasks.sync_listing_task, "update_state"), \
patch.object(tasks, "_run_browser_job", return_value={
"run_id": 99,
"status": "failed",
"full_scan_completed": False,
"cars_upserted": 0,
"cars_failed": 0,
"images_upserted": 0,
"skipped_existing": 0,
"elapsed_seconds": 1.0,
"failures": [{"vehicle_url": "listing", "error": "bad resume"}],
"listing": {"vehicles_collected": 0},
}):
get_persistence.return_value = MagicMock()
redis_client = MagicMock()
redis_client.get.return_value = None
get_redis.return_value = redis_client
stop_event = MagicMock()
heartbeat_thread = MagicMock()
start_heartbeat.return_value = (stop_event, heartbeat_thread)
with patch.object(tasks.sync_listing_task, "app", new=MagicMock()) as task_app:
tasks.sync_listing_task.push_request(id="task-900")
try:
result = tasks.sync_listing_task.run()
finally:
tasks.sync_listing_task.pop_request()
self.assertEqual(result["status"], "failed")
bump_streak.assert_called_once()
clear_pending.assert_called()
task_app.send_task.assert_not_called()
def test_bump_checkpoint_resume_failure_deletes_after_second_failure(self) -> None:
redis_client = MagicMock()
checkpoint = {
"status": "in_progress",
"last_successful_page": 9,
"resume_failures": 1,
"lane": "iaai_cars",
}
failures, deleted = tasks._bump_checkpoint_resume_failure(
redis_client,
checkpoint,
reason="resume failed",
)
self.assertEqual(failures, 2)
self.assertTrue(deleted)
redis_client.delete.assert_called_once_with(tasks.SYNC_LISTING_CHECKPOINT_KEY)
def test_bump_checkpoint_resume_failure_persists_first_failure(self) -> None:
redis_client = MagicMock()
checkpoint = {
"status": "in_progress",
"last_successful_page": 9,
"resume_failures": 0,
"lane": "iaai_cars",
}
failures, deleted = tasks._bump_checkpoint_resume_failure(
redis_client,
checkpoint,
reason="resume failed",
)
self.assertEqual(failures, 1)
self.assertFalse(deleted)
redis_client.set.assert_called_once()
def test_sync_listing_task_deletes_checkpoint_after_second_resume_failure(self) -> None:
with patch.object(tasks, "_get_persistence") as get_persistence, \
patch.object(tasks, "_get_redis") as get_redis, \
patch.object(tasks, "_acquire_lock", return_value=True), \
patch.object(tasks, "_is_full_scan_done", return_value=False), \
patch.object(tasks, "_start_lock_heartbeat") as start_heartbeat, \
patch.object(tasks, "_release_lock_if_owner") as release_lock, \
patch.object(tasks, "_bump_checkpoint_resume_failure", return_value=(2, True)) as bump_failures, \
patch.object(tasks.sync_listing_task, "update_state"), \
patch.object(tasks.sync_listing_task, "retry", side_effect=tasks.sync_listing_task.MaxRetriesExceededError()), \
patch("iaai_scraper.worker.tasks.parse_listing_segments", return_value=[]):
get_persistence.return_value = MagicMock()
redis_client = MagicMock()
redis_client.get.side_effect = lambda key: json.dumps({
"status": "in_progress",
"last_successful_page": 9,
"resume_failures": 1,
"make": None,
"model": None,
"lane": "iaai_cars",
}) if key == tasks.SYNC_LISTING_CHECKPOINT_KEY else None
get_redis.return_value = redis_client
stop_event = MagicMock()
heartbeat_thread = MagicMock()
start_heartbeat.return_value = (stop_event, heartbeat_thread)
with patch.object(tasks, "IAAIScraper") as scraper_cls:
scraper_ctx = MagicMock()
scraper_ctx.__enter__.return_value.sync_listing = MagicMock(side_effect=RuntimeError("Failed to resume listing at page 10"))
scraper_ctx.__exit__.return_value = None
scraper_cls.return_value = scraper_ctx
tasks.sync_listing_task.push_request(id="task-793")
try:
result = tasks.sync_listing_task.run()
finally:
tasks.sync_listing_task.pop_request()
self.assertEqual(result["status"], "failed")
bump_failures.assert_called_once()
release_lock.assert_called_once()
if __name__ == "__main__":
unittest.main()