#!/usr/bin/env python3 # -*-coding:UTF-8 -* import datetime import os import sys sys.path.append(os.environ['AIL_BIN']) ################################## # Import Project packages ################################## from lib import ConfigLoader config_loader = ConfigLoader.ConfigLoader() r_statistics = config_loader.get_redis_conn("ARDB_Statistics") # r_serv_trend = ConfigLoader().get_redis_conn("ARDB_Trending") config_loader = None PIE_CHART_MAX_CARDINALITY = 8 def incr_module_timeout_statistic(module_name): curr_date = datetime.date.today() r_statistics.hincrby(curr_date.strftime("%Y%m%d"), 'paste_by_modules_timeout:{}'.format(module_name), 1) def create_item_statistics(item_id, source, size): pass def get_item_sources(): return r_statistics.smembers('all_provider_set') def get_nb_items_processed_by_day_and_source(date, source): nb_items = r_statistics.hget(f'{source}_num', date) if not nb_items: nb_items = 0 return int(nb_items) def get_items_total_size_by_day_and_source(date, source): total_size = r_statistics.hget(f'{source}_size', date) if not total_size: total_size = 0 return float(total_size) def get_items_av_size_by_day_and_source(date, source): av_size = r_statistics.hget(f'{source}_avg', date) if not av_size: av_size = 0 return float(av_size) def _create_item_stats_size_nb(date, source, num, size, avg): r_statistics.hset(f'{source}_num', date, num) r_statistics.hset(f'{source}_size', date, size) r_statistics.hset(f'{source}_avg', date, avg) def get_item_stats_size_avg_by_date(): return r_statistics.zrange(f'top_avg_size_set_{date}', 0, -1, withscores=True) def get_item_stats_nb_by_date(): return r_statistics.zrange(f'providers_set_{date}', 0, -1, withscores=True) def _set_item_stats_nb_by_date(date, source): return r_statistics.zrange(f'providers_set_{date}', ) # # TODO: load ZSET IN CACHE => FAST UPDATE def update_item_stats_size_nb(item_id, source, size, date): # Add/Update in Redis r_statistics.sadd('all_provider_set', source) nb_items = int(r_statistics.hincrby(f'{source}_num', date, 1)) sum_size = float(r_statistics.hincrbyfloat(f'{source}_size', date, size)) new_avg = sum_size / nb_items r_statistics.hset(f'{source}_avg', date, new_avg) # TOP Items Size if r_statistics.zcard(f'top_size_set_{date}') < PIE_CHART_MAX_CARDINALITY: r_statistics.zadd(f'top_avg_size_set_{date}', new_avg, source) else: member_set = r_statistics.zrangebyscore(f'top_avg_size_set_{date}', '-inf', '+inf', withscores=True, start=0, num=1) # Member set is a list of (value, score) pairs if float(member_set[0][1]) < new_avg: # remove min from set and add the new one r_statistics.zrem(f'top_avg_size_set_{date}', member_set[0][0]) r_statistics.zadd(f'top_avg_size_set_{date}', new_avg, source) # TOP Nb Items if r_statistics.zcard(f'providers_set_{date}') < PIE_CHART_MAX_CARDINALITY or r_statistics.zscore(f'providers_set_{date}', source) != None: r_statistics.zadd(f'providers_set_{date}', float(nb_items), source) else: # zset at full capacity member_set = r_statistics.zrangebyscore(f'providers_set_{date}', '-inf', '+inf', withscores=True, start=0, num=1) # Member set is a list of (value, score) pairs if int(member_set[0][1]) < nb_items: # remove min from set and add the new one r_statistics.zrem(member_set[0][0]) r_statistics.zadd(f'providers_set_{date}', float(nb_items), source) # keyword num def _add_module_stats(module_name, total_sum, keyword, date): r_statistics.zadd(f'top_{module_name}_set_{date}', float(total_sum), keyword) # # TODO: ONE HSET BY MODULE / CUSTOM STATS def update_module_stats(module_name, num, keyword, date): # Add/Update in Redis r_statistics.hincrby(date, f'{module_name}-{keyword}', int(num)) # # TODO: RENAME ME !!!!!!!!!!!!!!!!!!!!!!!!! # Compute Most Posted # check if this keyword is eligible for progression keyword_total_sum = 0 curr_value = r_statistics.hget(date, f'{module_name}-{keyword}') keyword_total_sum += int(curr_value) if curr_value is not None else 0 if r_statistics.zcard(f'top_{module_name}_set_{date}') < PIE_CHART_MAX_CARDINALITY: r_statistics.zadd(f'top_{module_name}_set_{date}', float(keyword_total_sum), keyword) else: # zset at full capacity member_set = r_statistics.zrangebyscore(f'top_{module_name}_set_{date}', '-inf', '+inf', withscores=True, start=0, num=1) # Member set is a list of (value, score) pairs if int(member_set[0][1]) < keyword_total_sum: #remove min from set and add the new one r_statistics.zrem(f'top_{module_name}_set_{date}', member_set[0][0]) r_statistics.zadd(f'top_{module_name}_set_{date}', float(keyword_total_sum), keyword) def get_module_tld_stats_by_tld_date(date, tld): nb_tld = r_statistics.hget(f'credential_by_tld:{date}', tld) if not nb_tld: nb_tld = 0 return int(nb_tld) def get_module_tld_stats_by_date(module, date): return r_statistics.hgetall(f'{module}_by_tld:{date}') def add_module_tld_stats_by_date(module, date, tld, nb): r_statistics.hincrby(f'{module}_by_tld:{date}', tld, int(nb)) # r_stats.zincrby('module:Global:incomplete_file', datetime.datetime.now().strftime('%Y%m%d'), 1) # r_stats.zincrby('module:Global:invalid_file', datetime.datetime.now().strftime('%Y%m%d'), 1)