def test_one_second_buckets(): r = fakeredis.FakeStrictRedis() bucket_size = 1 distribution = RedisPopularityDistribution( r, "DECKS", namespace="test", ttl=3600, bucket_size=bucket_size ) current_ts = datetime.utcnow() td = timedelta(microseconds=current_ts.microsecond) t_0 = current_ts - td t_0_token = int(t_0.timestamp()) start_token = distribution._to_start_token(t_0) assert start_token == t_0_token end_token = distribution._to_end_token(t_0) assert end_token == t_0_token end_token_from_start_token = distribution._convert_to_end_token(start_token) assert end_token_from_start_token == end_token next_token = distribution._next_token(start_token) assert next_token == (start_token + bucket_size) t_3 = t_0 + timedelta(seconds=3) buckets = distribution._generate_bucket_tokens_between( start_token, distribution._to_end_token(t_3) ) for index, bucket in enumerate(buckets): assert bucket[0] == t_0_token + index assert bucket[1] == bucket[0] + (bucket_size - 1)
def test_bucket_sizes(): r = fakeredis.FakeStrictRedis() td = timedelta(days=1) current_ts = datetime.utcnow() yesterday_ts = current_ts - td # Test bucket sizes between 15 minutes and 6 hours in 15 minute increments for bucket_size in range(900, 21600, 900): dist = RedisPopularityDistribution( r, "DECKS", namespace="test", bucket_size=bucket_size ) start_token = dist._to_start_token(current_ts) end_token = dist._to_end_token(current_ts) next_start_token = dist._next_token(start_token) assert end_token + 1 == next_start_token assert end_token - start_token == bucket_size - 1 assert dist._convert_to_end_token(start_token) == end_token yesterday_start_token = dist._to_start_token(yesterday_ts) buckets = dist._generate_bucket_tokens_between(yesterday_start_token, end_token) # Assert the first bucket contains the start token first_bucket = buckets[0] assert first_bucket[0] == yesterday_start_token # Assert the last bucket matches the end token last_bucket = buckets[-1] assert last_bucket[1] == end_token # Assert the total number of buckets matches the expected number expected_num_buckets = ceil((end_token - yesterday_start_token) / bucket_size) assert len(buckets) == expected_num_buckets
def __init__(self, redis, name, max_items=9, bucket_size=5, ttl=600):
self.name = name
self.max_items = max_items
self.bucket_size = bucket_size
self.observations = RedisPopularityDistribution(
redis,
name="%s_OBSERVATIONS" % self.name,
namespace="POPULARITY",
ttl=ttl,
max_items=self.max_items,
bucket_size=self.bucket_size)
self.wins = RedisPopularityDistribution(redis,
name="%s_WINS" % self.name,
namespace="POPULARITY",
ttl=ttl,
max_items=self.max_items,
bucket_size=self.bucket_size)
def _popularity_distribution(self, node):
dist = RedisPopularityDistribution(
self.redis_primary,
name=node.key,
namespace="POPULARITY",
ttl=self.popularity_ttl,
max_items=self._max_collection_size_for_depth(node.depth),
bucket_size=self.bucket_size)
return dist
def get_played_cards_distribution(game_type, redis_client=None, ttl=600):
if redis_client:
redis = redis_client
else:
redis = get_live_stats_redis()
name = "PLAYED_CARDS_%s" % game_type
return RedisPopularityDistribution(redis,
name=name,
namespace="POPULARITY",
ttl=ttl,
max_items=5000,
bucket_size=5)
def test_bucket_sizes_and_ttls(_mock_lock):
r = fakeredis.FakeStrictRedis()
distribution = RedisPopularityDistribution(
r,
"DECKS",
ttl=5,
namespace="test",
bucket_size=1
)
# Create t_0 as 5 seconds in the past
current_ts = datetime.utcnow()
td = timedelta(seconds=5, microseconds=current_ts.microsecond)
t_0 = current_ts - td
def t_N(N):
return t_0 + timedelta(seconds=N)
distribution.increment("A", as_of=t_N(1))
distribution.increment("B", as_of=t_N(2))
distribution.increment("A", as_of=t_N(3))
distribution.increment("A", as_of=t_N(4))
# First assert the full distribution exists
expected_distribution = {"A": 3.0, "B": 1.0}
actual_distribution = distribution.distribution()
assert expected_distribution == actual_distribution
# Then assert accessing a partial distribution (t_2, t_3) within the full time range
expected_distribution = {"A": 1.0, "B": 1.0}
actual_distribution = distribution.distribution(start_ts=t_N(2), end_ts=t_N(3))
assert expected_distribution == actual_distribution
# Finally, assert that the first observation of "A" has aged out due to the TTL
time.sleep(1)
expected_distribution = {"A": 2.0, "B": 1}
actual_distribution = distribution.distribution()
assert expected_distribution == actual_distribution
def test_redis_popularity_distribution(_mock_lock): r = fakeredis.FakeStrictRedis() distribution = RedisPopularityDistribution(r, "DECKS", namespace="test") actuals = defaultdict(int) for deck in DECKS: actuals[deck] += 1 distribution.increment(deck) assert distribution.size() == 18 assert distribution.observations() == len(DECKS) actual_most_popular = list(sorted(actuals.items(), key=lambda t: t[1], reverse=True)) dist_most_popular = list( sorted(distribution.distribution().items(), key=lambda t: t[1], reverse=True) ) actual_result = actual_most_popular[0][0] expected_result = int(dist_most_popular[0][0]) assert actual_result == expected_result assert distribution.popularity(expected_result) == 32.0
class PopularityWinrateDistribution:
def __init__(self, redis, name, max_items=9, bucket_size=5, ttl=600, use_lua=None):
self.name = name
self.max_items = max_items
self.bucket_size = bucket_size
self.observations = RedisPopularityDistribution(
redis,
name="%s_OBSERVATIONS" % self.name,
namespace="POPULARITY",
ttl=ttl,
max_items=self.max_items,
bucket_size=self.bucket_size,
use_lua=use_lua
)
self.wins = RedisPopularityDistribution(
redis,
name="%s_WINS" % self.name,
namespace="POPULARITY",
ttl=ttl,
max_items=self.max_items,
bucket_size=self.bucket_size,
use_lua=use_lua
)
def increment(self, key, win=False, as_of=None):
self.observations.increment(key, as_of=as_of)
if win:
self.wins.increment(key, as_of=as_of)
def distribution(self, start_ts, end_ts):
games = self.observations.distribution(
start_ts=start_ts,
end_ts=end_ts,
)
wins = self.wins.distribution(
start_ts=start_ts,
end_ts=end_ts,
)
result = {}
for key, val in games.items():
result[key] = {
"games": val,
"wins": wins.get(key, 0)
}
return result