Peach: Local-first video metadata scraper and media manager.
fix(feeds): 空跳保留上一轮的失败,冷却跳过不在补全旧页时空转
longmeidaoaadbbf23ee4c
diff --git a/docs/adr/0132-feed-and-follow-checks-stagger-by-source.md b/docs/adr/0132-feed-and-follow-checks-stagger-by-source.mdindex 33efd9b12..3dd768018 100644--- a/docs/adr/0132-feed-and-follow-checks-stagger-by-source.md+++ b/docs/adr/0132-feed-and-follow-checks-stagger-by-source.md@@ -43,6 +43,6 @@## 后果- 订阅源对 JavDB 的请求回到设计频率:11 个源每小时约 1.8 次,均匀分在各跳里,不再一次 11 个;漂移修掉后比改动前实际的约 0.9 次多。订阅 100 个源时约每小时 17 次,仍远低于 3 秒间隔的每分钟 20 次。-- 定时调度状态里的「上次运行」每 30 分钟更新一次;`last_checked` 是这一跳拉了几个源,空跳为 0。+- 定时调度状态里的「上次运行」只记真正拉了源的那一跳。空跳不改上次运行时间、上次失败与连续失败计数:按成功结算的话,一次真实失败半小时后就从界面上消失,失败的源却还要等到下一次到期才重拉。- 设置页「自动检查更新」一行说明了按活跃度放慢的规则。- rule34xxx 每轮会重复补全约 281 条始终补不齐的详情,是追更请求最多的地方,不在本 ADR 范围。diff --git a/src/peach/follow_scheduler.py b/src/peach/follow_scheduler.pyindex 29154ce46..e99299eff 100644--- a/src/peach/follow_scheduler.py+++ b/src/peach/follow_scheduler.py@@ -197,6 +197,7 @@ class FollowUpdateScheduler:def _run(self) -> None:with self._state_lock:started_before = self._last_started_at+ error_before = self._last_errorself._running = Trueself._last_started_at = _stamp()self._last_error = None@@ -221,6 +222,15 @@ class FollowUpdateScheduler:self._last_skipped_at = _stamp()self._last_started_at = started_beforereturn+ if result.get("idle"):+ # 这一跳没有到期的来源(订阅源每半小时一跳,ADR-0132):什么也没做,+ # 上一轮的结论原样留着。按成功结算的话,一次真实失败半小时后就从界面上消失,+ # 而失败的那几个源还要等到下一次到期才重拉。+ skipped = True+ with self._state_lock:+ self._last_started_at = started_before+ self._last_error = error_before+ returnrows = result.get("results") or []failures = [row for row in rows if not row.get("ok")]with self._state_lock:diff --git a/src/peach/web_follow.py b/src/peach/web_follow.pyindex 9f0e0bb6c..a35ee5228 100644--- a/src/peach/web_follow.py+++ b/src/peach/web_follow.py@@ -2395,8 +2395,8 @@ class _FollowCheckSession:connector_factory=build_connector, older=older, progress=progress,initial_days=self.initial_days, cooldown=self.cooldown))more = (older and self.body.get('backfill_all') and result.get('ok')- and not result.get('exhausted') and isinstance(result.get('page'), int)- and round_index < 500)+ and not result.get('exhausted') and not result.get('deferred')+ and isinstance(result.get('page'), int) and round_index < 500)self.remember(row, result, finished=not more)if not more:breakdiff --git a/tests/test_follow_scheduler.py b/tests/test_follow_scheduler.pyindex 8c458d17f..8af1dc352 100644--- a/tests/test_follow_scheduler.py+++ b/tests/test_follow_scheduler.py@@ -211,6 +211,23 @@ class FollowUpdateSchedulerTests(unittest.TestCase):before + timedelta(minutes=31))self.assertEqual(scheduler.backoff_minutes(2), 60)+ def test_an_idle_tick_keeps_the_last_failure_on_record(self):+ """没有到期源的一跳什么也没做:上一轮的失败、连续失败计数和上次运行时间都留着。"""+ outcomes = [{"ok": True, "checked": 1, "results": [{"ok": False, "added": 0}]},+ {"ok": True, "idle": True, "checked": 0, "results": []}]+ with tempfile.TemporaryDirectory() as temporary:+ scheduler = FollowUpdateScheduler(Path(temporary), lambda: outcomes.pop(0),+ available=True, tick_minutes=30)+ scheduler._run()+ failed = scheduler.status()+ scheduler._run()+ status = scheduler.status()+ self.assertEqual(status["last_error"], "1 个来源检查失败")+ self.assertEqual(status["consecutive_failures"], 1)+ self.assertEqual((status["last_started_at"], status["last_finished_at"], status["last_checked"]),+ (failed["last_started_at"], failed["last_finished_at"], 1))+ self.assertFalse(status["running"])+def test_reader_never_starts_or_accepts_schedule_changes(self):with tempfile.TemporaryDirectory() as temporary:scheduler = FollowUpdateScheduler(Path(temporary), lambda: {}, available=False)diff --git a/tests/test_follow_web.py b/tests/test_follow_web.pyindex dd697c1c1..fee9b55bd 100644--- a/tests/test_follow_web.py+++ b/tests/test_follow_web.py@@ -183,8 +183,11 @@ class FollowContractTests(unittest.TestCase):self.contract.follow_media_resolver.pause('rule34video', 60)with mock.patch.object(web_follow, 'build_connector') as factory:result = self._post('/api/follow/check', {})+ # 往回一路抓到底也一样:冷却中的那条只跳过一次,不在原地空转翻页。+ backfill = self._post('/api/follow/check', {'older': True, 'backfill_all': True})self.assertEqual(factory.call_count, 0)self.assertEqual((result['results'], result['deferred']), ([], 2))+ self.assertEqual((backfill['results'], backfill['deferred']), ([], 2))def test_a_scheduled_round_checks_only_the_sources_that_are_due(self):"""定时那一跳跳过刚查过、又久未更新的来源;手动检查照旧全查。"""