ForkJoin/MapReduce Demo Code

channelonline_sync.py

# -*- coding: utf-8

'''
从数据源同步获取Channel在线的数据，保存到mongodb中

这个脚本一般以crontab方式运行，每30秒运行一次

@author: [email protected]
'''
from app_logger import Logger, set_logging_path
from db import yy_mysql
import threading
import time
import pymongo
from forkjoin import ForkJoin
from map_reduce import map_reduce
from timeit import itertools

#todo change to Configuration
db_session_list = [
    {"host":"1.2.3.4:3306", "db":"appDaemon",
     "user":"user", "password":"password" },
    {"host":"2.3.4.5:3306", "db":"appDaemon",
     "user":"user", "password":"password" }
    ]

def _query(name, sql, host, database, user, password):
    
    def totuple(row):
        ts = int(time.mktime( row["sync_date"].timetuple()))
        ts = ts - (ts % 60)
        users = int(row["users"])
        sessions = int(row["sessions"])
        return (ts, users, sessions)
            
    for i in range(0,3):    # retry 3 times
        try:
            conn = yy_mysql.Connection(host, database, user, password, connect_timeout=10)
            rows = conn.query(sql)
            tuples = map(totuple, rows)
            Logger().debug("%s read data ok: %s", name, tuples)
            return tuples
            break
        except:
            Logger().exception("query database failed")

def __query_datasource():
    sql = """
    select sync_date, sum(totalUser) as users, count(totalUser) as sessions
    from tbl_session_his
    where sync_date >= (
        select date_add( (select max(sync_date) from tbl_session_his), interval -5 minute)
    )
    group by sync_date
    order by sync_date
    """
        
    forkjoin = ForkJoin()
    
    for db in db_session_list:        
        forkjoin.fork(target=_query, kwargs=dict(
            name=db["host"], sql=sql,
            host=db["host"], database=db["db"], user=db["user"], password=db["password"]
            )
        )
    
    results = forkjoin.join(30)
    
    Logger().info("query databases, result: %d", len(results) )
    
    return results;
    
def __save(mongodb, records):
            
    records = sorted(records, cmp = lambda r1,r2: r1[0] - r2[0])
        
    for (ts, users, sessions) in records:
        Logger().info("save channel: %d, %d, %d", ts, users, sessions)
        try:
            mongodb.online2["sum_channel_user"].update(
                        {'_id':ts, "val": {"$lt": users } },
                        {"$set": {"val": users, "sessions": sessions} }, 
                        upsert=True, safe=True)
        except pymongo.errors.DuplicateKeyError:    # 已经存在，且无需更新
            pass
        except:
            Logger().exception("save channel: %d, %d, %d failed", ts, users, sessions)

"""
    删除最近1分钟的数据, 频道数据比较准时，YY一般需要等待超过1分钟
"""
def __removeLatest(records):
    now = time.time()
    return [ rec for rec in records if now - rec[0] >= 60 ]

def channelonline_sync():

    Logger().info("ChannelOnline Data Sync Begin...")    
    mongodb = pymongo.Connection("127.0.0.1", 27017)
    
    results = __query_datasource()
    """[ (time, sessionCount, totalUsers) .. ]
    """
    merged = map_reduce(itertools.chain(*results),
        index = lambda row:(row[0],row),
        mapper = lambda key, row:[(key,row)],
        reducer = lambda key, rows: (key, max([row[1] for row in rows]), max([row[2] for row in rows]))
        )
    result = __removeLatest(merged.values())
    __save(mongodb, result)
    Logger().info("ChannelOnline Data Completed\n")    
    
if __name__ == "__main__":
    set_logging_path("logs/channelonline_sync.log")
    channelonline_sync()

forkjoin.py

# -*- coding: utf-8

'''
提供模拟的fork join编程模式

@author: [email protected]
'''
import threading
import time
import itertools

class ForkJoin:
    
    def __init__(self):
        self.condition = threading.Condition()
        self.condition.acquire();
        self.threads = []
        self.results = []
        pass
    
    def fork(self, target, args=(), kwargs=None):
        
        def run():
            
            self.threads.append(threading.currentThread())
            
            ret = target(*args, **kwargs)
            if ret != None:
                self.results.append(ret)
            
            self.threads.remove(threading.currentThread())
            self.condition.acquire()
            self.condition.notify()
            self.condition.release()
            
        if kwargs == None: kwargs = {}    
        thread = threading.Thread(target = run)
        thread.start()
    
    def join(self, seconds):
        begin = time.time()
        end = begin + seconds
        
        while len(self.threads) > 0:
            remain = end - time.time()
            if remain <= 0:
                break
            self.condition.wait(remain)
        
        self.condition.release()
        return self.results
    

if __name__ == "__main__":
    
    # Pie = sum( (-1)^n/(2*n+1) )
    def task(begin, end):        
        return sum([ float(pow(-1, n)) / (2*n+1) for n in range(begin, end) ])
        
    begin = time.time()
    fork_join = ForkJoin()
    
    for i in range(0, 5):
        loop = 1000000
        fork_join.fork(task, args=(i*loop,i*loop+loop))
    
    results = fork_join.join(30)
    end = time.time()
    print "results = %s, pie=%f time=%d" % (results, 4*sum(itertools.chain(results)), end-begin)
    
    pass

map_reduce.py

# -*- coding: utf-8

'''
简单的map reduce框架

@author: [email protected]
'''

def map_reduce(rows, index, mapper, reducer):
    
    """
        rows: [ row0, row1, .. ]
        index: lambda row: (k1,v1)
        mapper: lambda k,v: [(k1, v2), (k2, v2)]
        reducer: lambda k, vals: val
    """
    
    group = {}
    # group by key
    for row in rows:
        (_key, _value) = index(row)
        for (k,v) in mapper(_key,_value):
            if k in group:
                group[k].append(v)
            else:
                group[k] = [v]
            
    reduced = {}
    for (key,vals) in group.items():
        reduced[key] = reducer(key, vals)
    
    return reduced

if __name__ == "__main__":
    
    array = [   {"_id":1, "value":100},
                {"_id":2, "value":200},
                {"_id":3, "value":300},
                {"_id":1, "value":103},
                {"_id":2, "value":195},
                {"_id":3, "value":320},
            ]
    
    results = map_reduce(
        array,
        index = lambda r: (r["_id"],r),
        mapper = lambda k,v: [(k,v)],
        reducer = lambda k,vals: max(vals, key=lambda val: val["value"])
               )
    print results