fix: pace failed calendar refresh retries

backend/calendar_refresh.py

@@ -10,7 +10,7 @@ from sqlalchemy import or_, select
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from . import calendar as calendar_service
-from .models import CalendarSubscription, utcnow
+from .models import CalendarSubscription
 
 logger = logging.getLogger(__name__)
 DEFAULT_REFRESH_INTERVAL = timedelta(minutes=15)
@@ -22,7 +22,9 @@ async def refresh_subscription_cache(
     row: CalendarSubscription,
     *,
     fail_without_cache: bool = True,
+    attempted_at: datetime | None = None,
 ) -> bool:
+    attempt_time = attempted_at or datetime.now(UTC)
     requested_url = row.url
     requested_version = getattr(row, "updated_at", None)
     try:
@@ -51,7 +53,8 @@ async def refresh_subscription_cache(
             row.ics_cache = result.content.decode("utf-8-sig")
             row.etag = result.etag
             row.last_modified = result.last_modified
-        row.refreshed_at = utcnow()
+        row.refreshed_at = attempt_time
+        row.updated_at = attempt_time
         row.last_error = None
         await db.commit()
         return True
@@ -61,6 +64,7 @@ async def refresh_subscription_cache(
             await db.rollback()
             raise HTTPException(502, error) from exc
         row.last_error = error
+        row.updated_at = attempt_time
         await db.commit()
         return False
 
@@ -76,13 +80,15 @@ async def refresh_due_subscriptions(
     rows = (await db.scalars(select(CalendarSubscription).where(
         CalendarSubscription.enabled.is_(True),
         or_(
-            CalendarSubscription.refreshed_at.is_(None),
+            CalendarSubscription.updated_at.is_(None),
-            CalendarSubscription.refreshed_at <= cutoff,
+            CalendarSubscription.updated_at <= cutoff,
         ),
     ).order_by(CalendarSubscription.refreshed_at, CalendarSubscription.created_at))).all()
     for row in rows:
         try:
-            await refresh_subscription_cache(db, row, fail_without_cache=False)
+            await refresh_subscription_cache(
+                db, row, fail_without_cache=False, attempted_at=current
+            )
         except Exception:
             await db.rollback()
             logger.exception("Unexpected calendar refresh failure", extra={"subscription_id": str(row.id)})

tests/test_calendar_subscriptions.py

@@ -351,3 +351,58 @@ def test_refresh_discards_response_when_url_changes_in_flight(monkeypatch):
     assert row.url == "https://example.com/new.ics"
     assert row.ics_cache is None
     assert db.committed is False
+
+
+def test_failed_source_waits_until_next_interval_and_does_not_abort_batch(client, monkeypatch):
+    client = initialized(client)
+    monkeypatch.setattr(
+        "backend.calendar.validate_calendar_url",
+        lambda url: (url, "93.184.216.34", 443),
+    )
+    monkeypatch.setattr(
+        "backend.calendar.fetch_calendar",
+        lambda *args, **kwargs: FetchResult(ICS, None, None, False),
+    )
+    first = client.post(
+        "/api/v1/calendar-subscriptions",
+        json={"name": "first", "url": "https://example.com/first.ics"},
+    ).json()
+    client.post(
+        "/api/v1/calendar-subscriptions",
+        json={"name": "second", "url": "https://example.com/second.ics"},
+    )
+    attempts = []
+
+    def fetch(url, **kwargs):
+        attempts.append(url)
+        if url.endswith("first.ics"):
+            raise HTTPException(502, "upstream down")
+        return FetchResult(ICS, None, None, False)
+
+    monkeypatch.setattr("backend.calendar.fetch_calendar", fetch)
+
+    async def refresh_twice():
+        from sqlalchemy.ext.asyncio import async_sessionmaker
+
+        from backend.db import get_engine
+
+        factory = async_sessionmaker(get_engine(), expire_on_commit=False)
+        now = datetime.now(UTC) + timedelta(minutes=16)
+        async with factory() as db:
+            first_count = await refresh_due_subscriptions(
+                db, now=now, refresh_interval=timedelta(minutes=15)
+            )
+        async with factory() as db:
+            second_count = await refresh_due_subscriptions(
+                db, now=now + timedelta(minutes=1), refresh_interval=timedelta(minutes=15)
+            )
+        return first_count, second_count
+
+    assert asyncio.run(refresh_twice()) == (2, 0)
+    assert attempts == [
+        "https://example.com/first.ics",
+        "https://example.com/second.ics",
+    ]
+    listed = client.get("/api/v1/calendar-subscriptions").json()
+    failed = next(item for item in listed if item["id"] == first["id"])
+    assert failed["stale"] is True