Private
Public Access
Push due-feed scheduling into an indexed DB query (C1)
poll_due_feeds loaded every active feed's timing fields and recomputed each one's error-backoff due-ness in Python on every beat tick (every 1-2 minutes) — scheduler CPU and DB reads grew linearly with total feed count regardless of how many were actually due. Feed.next_fetch_at (indexed together with is_active) is now set by fetch_feed itself right after every fetch attempt, success or error, using the same backoff formula that used to run per-feed in poll_due_feeds. The scheduler's query is now a direct indexed range scan (next_fetch_at <= now), ordered so the most overdue feeds claim first, and bounded to POLL_BATCH_SIZE so a large backlog spreads across several ticks instead of dispatching everything at once. A migration backfills next_fetch_at for existing feeds using the same formula. reset_feed_backoff also clears next_fetch_at so a manually-reset feed is immediately due again, matching its prior behavior. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
1 parent
b84cfe6474
commit
b33fe905d2
6 files changed
+186
-50
No files matched your search
+34
-23
@@ -22,27 +22,31 @@ logger = logging.getLogger(__name__)
|
||||
# while a worker that died mid-fetch doesn't wedge the feed forever.
|
||||
CLAIM_STALE_MINUTES = 10
|
||||
|
||||
# Caps how many feeds one poll_due_feeds tick will claim/dispatch, so a
|
||||
# large backlog (first run after a long outage, a huge import) spreads
|
||||
# across several beat ticks instead of enqueuing everything — and every
|
||||
# bounded Feed query below stays index-only work, not a full scan (C1).
|
||||
POLL_BATCH_SIZE = 500
|
||||
|
||||
|
||||
@shared_task
|
||||
def poll_due_feeds():
|
||||
"""Find every active Feed whose polling interval has elapsed, atomically
|
||||
claim it (so overlapping beat ticks or a slow-running previous fetch
|
||||
can't dispatch the same feed twice), and dispatch a fetch_feed task for
|
||||
each claimed feed — fan-out happens here, not via one Celery Beat entry
|
||||
per feed.
|
||||
"""Find a bounded batch of active Feeds whose next_fetch_at has
|
||||
elapsed (set by fetch_feed itself after every attempt — see Feed.
|
||||
next_fetch_at), atomically claim them (so overlapping beat ticks or a
|
||||
slow-running previous fetch can't dispatch the same feed twice), and
|
||||
dispatch a fetch_feed task for each claimed feed — fan-out happens
|
||||
here, not via one Celery Beat entry per feed.
|
||||
"""
|
||||
now = timezone.now()
|
||||
claim_stale_before = now - timedelta(minutes=CLAIM_STALE_MINUTES)
|
||||
# Loaded once per call, not per feed — the interval is site-wide
|
||||
# (apps.core.models.SiteSettings), so every feed below shares it.
|
||||
configured_interval = SiteSettings.load().fetch_interval_minutes
|
||||
due_feed_ids = [
|
||||
feed.id
|
||||
for feed in Feed.objects.filter(is_active=True).only(
|
||||
"id", "last_fetched_at", "fetch_error_count"
|
||||
)
|
||||
if _is_due(feed, configured_interval)
|
||||
]
|
||||
due_feed_ids = list(
|
||||
Feed.objects.filter(is_active=True)
|
||||
.filter(Q(next_fetch_at__isnull=True) | Q(next_fetch_at__lte=now))
|
||||
.filter(Q(claimed_at__isnull=True) | Q(claimed_at__lt=claim_stale_before))
|
||||
.order_by("next_fetch_at")
|
||||
.values_list("id", flat=True)[:POLL_BATCH_SIZE]
|
||||
)
|
||||
claimed_ids = []
|
||||
for feed_id in due_feed_ids:
|
||||
# The UPDATE's WHERE clause re-checks staleness at the database
|
||||
@@ -61,11 +65,14 @@ def poll_due_feeds():
|
||||
return len(claimed_ids)
|
||||
|
||||
|
||||
def _is_due(feed: Feed, configured_interval_minutes: int) -> bool:
|
||||
if feed.last_fetched_at is None:
|
||||
return True
|
||||
interval = effective_interval_minutes(feed, configured_interval_minutes)
|
||||
return feed.last_fetched_at <= timezone.now() - timedelta(minutes=interval)
|
||||
def _schedule_next_fetch(feed: Feed) -> None:
|
||||
"""Sets feed.next_fetch_at from its (just-updated) fetch_error_count —
|
||||
called right before every feed.save() in fetch_feed, so the next
|
||||
poll_due_feeds tick can find it with a plain indexed range query
|
||||
instead of recomputing every active feed's backoff in Python (C1)."""
|
||||
configured_interval = SiteSettings.load().fetch_interval_minutes
|
||||
interval = effective_interval_minutes(feed, configured_interval)
|
||||
feed.next_fetch_at = timezone.now() + timedelta(minutes=interval)
|
||||
|
||||
|
||||
# Bounds a single feed's fetch+parse+ingest, independent of
|
||||
@@ -102,10 +109,12 @@ def fetch_feed(feed_id: int):
|
||||
record_fetch_error(feed, "Unexpected error — see worker logs")
|
||||
feed.last_fetched_at = timezone.now()
|
||||
feed.claimed_at = None
|
||||
_schedule_next_fetch(feed)
|
||||
feed.save()
|
||||
raise
|
||||
|
||||
feed.claimed_at = None
|
||||
_schedule_next_fetch(feed)
|
||||
feed.save()
|
||||
|
||||
changed_article_ids = result.new_article_ids + result.updated_article_ids
|
||||
@@ -127,9 +136,11 @@ def fetch_feed(feed_id: int):
|
||||
|
||||
@shared_task
|
||||
def reset_feed_backoff(feed_id: int):
|
||||
"""Admin-triggered: clear a feed's error count so it's retried at its
|
||||
normal configured interval instead of the backed-off one."""
|
||||
Feed.objects.filter(pk=feed_id).update(fetch_error_count=0)
|
||||
"""Admin-triggered: clear a feed's error count and due-date backoff so
|
||||
it's retried at its normal configured interval instead of the backed-
|
||||
off one (next_fetch_at=None also makes it immediately due for the
|
||||
next poll_due_feeds tick, same as a feed that's never been fetched)."""
|
||||
Feed.objects.filter(pk=feed_id).update(fetch_error_count=0, next_fetch_at=None)
|
||||
|
||||
|
||||
@shared_task
|
||||
|
||||
Reference in new issue
Block a user