mirror of
https://github.com/pretix/pretix.git
synced 2026-05-08 15:44:02 +00:00
Move quota cache from database to redis (#2010)
This commit is contained in:
@@ -1,25 +1,22 @@
|
||||
import sys
|
||||
import time
|
||||
from collections import Counter, defaultdict
|
||||
from datetime import timedelta
|
||||
from itertools import zip_longest
|
||||
|
||||
from django.conf import settings
|
||||
from django.db import OperationalError, models
|
||||
from django.db import models
|
||||
from django.db.models import (
|
||||
Case, Count, F, Func, Max, OuterRef, Q, Subquery, Sum, Value, When,
|
||||
)
|
||||
from django.dispatch import receiver
|
||||
from django.utils.timezone import now
|
||||
from django_scopes import scopes_disabled
|
||||
from django_redis import get_redis_connection
|
||||
|
||||
from pretix.base.models import (
|
||||
CartPosition, Checkin, Event, LogEntry, Order, OrderPosition, Quota,
|
||||
Voucher, WaitingListEntry,
|
||||
CartPosition, Checkin, Order, OrderPosition, Quota, Voucher,
|
||||
WaitingListEntry,
|
||||
)
|
||||
from pretix.celery_app import app
|
||||
|
||||
from ...helpers.periodic import minimum_interval
|
||||
from ..signals import periodic_task, quota_availability
|
||||
from ..signals import quota_availability
|
||||
|
||||
|
||||
class QuotaAvailability:
|
||||
@@ -89,7 +86,11 @@ class QuotaAvailability:
|
||||
def queue(self, *quota):
|
||||
self._queue += quota
|
||||
|
||||
def compute(self, now_dt=None):
|
||||
def compute(self, now_dt=None, allow_cache=False, allow_cache_stale=False):
|
||||
"""
|
||||
Compute the queued quotas. If ``allow_cache`` is set, results may also be taken from a cache that might
|
||||
be a few minutes outdated. In this case, you may not rely on the results in the ``count_*`` properties.
|
||||
"""
|
||||
now_dt = now_dt or now()
|
||||
quotas = list(set(self._queue))
|
||||
quotas_original = list(self._queue)
|
||||
@@ -97,6 +98,35 @@ class QuotaAvailability:
|
||||
if not quotas:
|
||||
return
|
||||
|
||||
if allow_cache:
|
||||
if self._full_results:
|
||||
raise ValueError("You cannot combine full_results and allow_cache.")
|
||||
|
||||
elif not self._count_waitinglist:
|
||||
raise ValueError("If you set allow_cache, you need to set count_waitinglist.")
|
||||
|
||||
elif settings.HAS_REDIS:
|
||||
rc = get_redis_connection("redis")
|
||||
quotas_by_event = defaultdict(list)
|
||||
for q in quotas_original:
|
||||
quotas_by_event[q.event_id].append(q)
|
||||
|
||||
for eventid, evquotas in quotas_by_event.items():
|
||||
d = rc.hmget(f'quotas:{eventid}:availabilitycache', [str(q.pk) for q in evquotas])
|
||||
for redisval, q in zip(d, evquotas):
|
||||
if redisval is not None:
|
||||
data = [rv for rv in redisval.decode().split(',')]
|
||||
if time.time() - int(data[2]) < 120 or allow_cache_stale:
|
||||
quotas_original.remove(q)
|
||||
quotas.remove(q)
|
||||
if data[1] == "None":
|
||||
self.results[q] = int(data[0]), None, int(data[2])
|
||||
else:
|
||||
self.results[q] = int(data[0]), int(data[1]), int(data[2])
|
||||
|
||||
if not quotas:
|
||||
return
|
||||
|
||||
self._compute(quotas, now_dt)
|
||||
|
||||
for q in quotas_original:
|
||||
@@ -105,36 +135,48 @@ class QuotaAvailability:
|
||||
self.results[q] = resp
|
||||
|
||||
self._close(quotas)
|
||||
try:
|
||||
self._write_cache(quotas, now_dt)
|
||||
except OperationalError as e:
|
||||
# Ignore deadlocks when multiple threads try to write to the cache
|
||||
if 'deadlock' not in str(e).lower():
|
||||
raise e
|
||||
self._write_cache(quotas, now_dt)
|
||||
|
||||
def _write_cache(self, quotas, now_dt):
|
||||
if not settings.HAS_REDIS or not quotas:
|
||||
return
|
||||
|
||||
rc = get_redis_connection("redis")
|
||||
# We write the computed availability to redis in a per-event hash as
|
||||
#
|
||||
# quota_id -> (availability_state, availability_number, timestamp).
|
||||
#
|
||||
# We store this in a hash instead of inidividual values to avoid making two many redis requests
|
||||
# which would introduce latency.
|
||||
|
||||
# The individual entries in the hash are "valid" for 120 seconds. This means in a typical peak scenario with
|
||||
# high load *to a specific calendar or event*, lots of parallel web requests will receive an "expired" result
|
||||
# around the same time, recompute quotas and write back to the cache. To avoid overloading redis with lots of
|
||||
# simultaneous write queries for the same page, we place a very naive and simple "lock" on the write process for
|
||||
# these quotas.
|
||||
|
||||
lock_name = '_'.join([str(p) for p in sorted([q.pk for q in quotas])])
|
||||
if rc.exists(f'quotas:availabilitycachewrite:{lock_name}'):
|
||||
return
|
||||
rc.setex(f'quotas:availabilitycachewrite:{lock_name}', '1', 10)
|
||||
|
||||
update = defaultdict(list)
|
||||
for q in quotas:
|
||||
update[q.event_id].append(q)
|
||||
|
||||
for eventid, quotas in update.items():
|
||||
rc.hmset(f'quotas:{eventid}:availabilitycache', {
|
||||
str(q.id): ",".join(
|
||||
[str(i) for i in self.results[q]] +
|
||||
[str(int(time.time()))]
|
||||
) for q in quotas
|
||||
})
|
||||
rc.expire(f'quotas:{eventid}:availabilitycache', 3600 * 24 * 7)
|
||||
|
||||
# We used to also delete item_quota_cache:* from the event cache here, but as the cache
|
||||
# gets more complex, this does not seem worth it. The cache is only present for up to
|
||||
# 5 seconds to prevent high peaks, and a 5-second delay in availability is usually
|
||||
# tolerable
|
||||
update = []
|
||||
for q in quotas:
|
||||
rewrite_cache = self._count_waitinglist and (
|
||||
not q.cache_is_hot(now_dt) or self.results[q][0] > q.cached_availability_state
|
||||
or q.cached_availability_paid_orders is None
|
||||
)
|
||||
if rewrite_cache:
|
||||
q.cached_availability_state = self.results[q][0]
|
||||
q.cached_availability_number = self.results[q][1]
|
||||
q.cached_availability_time = now_dt
|
||||
if q in self.count_paid_orders:
|
||||
q.cached_availability_paid_orders = self.count_paid_orders[q]
|
||||
update.append(q)
|
||||
if update:
|
||||
Quota.objects.using('default').bulk_update(update, [
|
||||
'cached_availability_state', 'cached_availability_number', 'cached_availability_time',
|
||||
'cached_availability_paid_orders'
|
||||
], batch_size=50)
|
||||
|
||||
def _close(self, quotas):
|
||||
for q in quotas:
|
||||
@@ -404,44 +446,8 @@ class QuotaAvailability:
|
||||
self.results[q] = Quota.AVAILABILITY_GONE, 0
|
||||
|
||||
|
||||
@receiver(signal=periodic_task)
|
||||
@minimum_interval(minutes_after_success=60)
|
||||
def build_all_quota_caches(sender, **kwargs):
|
||||
refresh_quota_caches.apply()
|
||||
|
||||
|
||||
def grouper(iterable, n, fillvalue=None):
|
||||
"""Collect data into fixed-length chunks or blocks"""
|
||||
# grouper('ABCDEFG', 3, 'x') --> ABC DEF Gxx
|
||||
args = [iter(iterable)] * n
|
||||
return zip_longest(fillvalue=fillvalue, *args)
|
||||
|
||||
|
||||
@app.task
|
||||
@scopes_disabled()
|
||||
def refresh_quota_caches():
|
||||
# Active events
|
||||
active = LogEntry.objects.using(settings.DATABASE_REPLICA).filter(
|
||||
datetime__gt=now() - timedelta(days=7)
|
||||
).order_by().values('event').annotate(
|
||||
last_activity=Max('datetime')
|
||||
)
|
||||
for a in active:
|
||||
try:
|
||||
e = Event.objects.using(settings.DATABASE_REPLICA).get(pk=a['event'])
|
||||
except Event.DoesNotExist:
|
||||
continue
|
||||
quotas = e.quotas.filter(
|
||||
Q(cached_availability_time__isnull=True) |
|
||||
Q(cached_availability_time__lt=a['last_activity']) |
|
||||
Q(cached_availability_time__lt=now() - timedelta(hours=2))
|
||||
).filter(
|
||||
Q(subevent__isnull=True) |
|
||||
Q(subevent__date_to__isnull=False, subevent__date_to__gte=now() - timedelta(days=14)) |
|
||||
Q(subevent__date_from__gte=now() - timedelta(days=14))
|
||||
)
|
||||
|
||||
for qs in grouper(quotas, 100, None):
|
||||
qa = QuotaAvailability(early_out=False)
|
||||
qa.queue(*[q for q in qs if q is not None])
|
||||
qa.compute()
|
||||
|
||||
Reference in New Issue
Block a user