這是單一進程順序執行的程式碼:
import requests,time,os,random
def img_down(url):
with open("{}".format(str(random.random())+os.path.basename(url)),"wb") as fob:
fob.write(requests.get(url).content)
urllist=[]
with open("urllist.txt","r+") as u:
for a in u.readlines():
urllist.append(a.strip())
s=time.clock()
for i in range(len(urllist)):
img_down(urllist[i])
e=time.clock()
print ("time: %d" % (e-s))
這是多進程的程式碼:
from multiprocessing import Pool
import requests,os,time,random
def img_down(url):
with open("{}".format(str(random.random())+os.path.basename(url)),"wb") as fob:
fob.write(requests.get(url).content)
if __name__=="__main__":
urllist=[]
with open("urllist.txt","r+") as urlfob:
for s in urlfob.readlines():
urllist.append(s.strip())
s=time.clock()
p=Pool()
for i in range(len(urllist)):
p.apply_async(img_down,args=(urllist[i],))
p.close()
p.join()
e=time.clock()
print ("time: {}".format(e-s))
但是單一進程和多進程花費的時間幾乎沒差別,問題大概是requests阻塞IO,請問理解的對不對,程式碼該怎麼修改達到多進程的目的?
謝謝!
寫檔案的瓶頸在磁碟IO,並不在CPU,你並行並沒有多大作用,你可以試試不要寫入檔案再對比時間
Pool 不帶參數的話 是採用
os.cpu_count() or 1
如果是單核心CPU,或是採集不到數量 就只有1個進程而已。
應該是這個原因。