from multiprocessing.pool import ThreadPool
#from multiprocessing.dummy import Pool as ThreadPool
#這兩個ThreadPool好像區別不大,方法基本一樣
import time
def test1(x):
print("x1:", x)
try:
time.sleep(0.1)
if x == 8:
time.sleep(10)
elif x == 12:
time.sleep(5)
elif x == 49:
time.sleep(20)
print('這裏等待時間最長!')
else:
return
except Exception as e:
print('error1:', e)
raise Exception('timeout error1 ', x) # 只要沒走return都會觸發此異常
def test2(x):
print("x2:", x)
try:
time.sleep(0.1)
if x == 9:
time.sleep(1)
elif x == 11:
time.sleep(5)
elif x == 40:
time.sleep(10)
else:
return
except Exception as e:
print('error2:', e)
raise Exception('timeout error2 ', x)
def task(x):
ret1 = test1(x)
ret2 = test2(x)
print('ret1 and ret2: ', ret1, ret2)
#下面這些條件判斷其實是不必要的,因爲異常發生前get的結果已經是None了,而異常發生時這裏的else語句沒機會執行
if ret1 is None and ret2 is None:
return
else:
return 1
#raise Exception('test1 or test2 is timeout!')
#apply_async---error
def dome0():
time1 = time.time()
result1 = []
pool = ThreadPool(20)
for i in range(50):
result = pool.apply_async(func=task, args=(i,))
result1.append(result)
pool.close()
pool.join()
time2 = time.time()
print("time:", time2 - time1)
# 結果分析
#無法接收子線程的異常,主線程中不會報錯,這樣的實現是不合理的
#apply_async
def dome1():
time1 = time.time()
result1 = []
result2 = []
pool = ThreadPool(20)
for i in range(50):
result = pool.apply_async(func=task, args=(i,))
# 懷疑會在 x==9 處報異常,因爲sleep時間最短且可能在同一線程池中被處理,然而實際情況是在x==8處報異常
# 在 if 後面最小的x 處中斷 error is ('timeout error1 ', 8),而與sleep時間無關
result1.append(result)
# 下面兩句會影響總時間,只有wait時爲10s,加上join時20s,可以看出join會運行所有的子線程
# pool.close()
# pool.join()
# print('=====if====1') #這種方式能檢測出子線程的超時異常
# try:
# for i in result1:
# i.wait() # 等待線程函數執行完畢
# print('success:',i.successful())
# print('ready:',i.ready())
# print('get:',i.get()) # 線程函數返回值
# if i.ready(): # 線程函數是否已經啓動了
# if i.successful(): # 線程函數是否執行成功
# result2.append(i.get())
# except Exception as e:
# print("error is %s" % str(e))
# print('=====if====1')
print('=====while====2')
# 與1基本一樣,放到tempest中再比較一下,線程的等待情況,基本不要用while,目前來看和if真的沒區別
try:
for i in result1:
i.wait(timeout=60)
while i.ready():
print("ready %s" % i.ready())
print("successful %s" % i.successful())
print("i.get %s" % i.get())
if i.successful():
result2.append(i.get())
break
except Exception as e:
print("error is %s" % str(e))
print('=====while====2')
print("result2 and len: ", (result2, len(result2)))
time2 = time.time()
print("time:", time2 - time1)
# 結果分析
# error is ('timeout error1 ', 8)
# result2 and len: ([None, None, None, None, None, None, None, None], 8)
# time: 20.839099645614624
# 線程按 x 的進入順序觸發異常,而與sleep時間無關
# 可以i.get()得到異常發生前的返回,這一點是map類方法無法做到的
# 異常發生後其他線程並沒有中斷,還是執行最長的sleep(20),導致最後time 爲20s
#map_async
def dome2():
result2 = []
time1 = time.time()
pool = ThreadPool(20)
my_iter = range(50)
try:
result = pool.map_async(task, my_iter)
# 懷疑會在 x==9 處報異常,因爲sleep時間最短且可能在同一線程池中被處理
# 在 if 後面時間最短處中斷 error is ('timeout error2 ', 9),而與sleep時間有關
# map和map_async 不按 x 的進入順序處理
#下面兩句加不加好像不影響,wait已經等待了
pool.close()
pool.join()
result.wait() # 等待所有線程函數執行完畢
#放在這裏打印可以提前預警,知道錯誤產生的原因
print("ready %s" % result.ready())
print("successful %s" % result.successful())
print("i.get %s" % result.get())
print('=====if====1')
if result.ready(): # 線程函數是否已經啓動了
if result.successful(): # 線程函數是否執行成功
result2.append(result.get())
# 以下函數在全部線程都執行成功時可以重複執行,返回值相同,但異常時只能執行一次successful/get
print("ready %s" % result.ready())
print("successful %s" % result.successful())
print("i.get %s" % result.get())
print("ready %s" % result.ready())
print("successful %s" % result.successful())
print("i.get %s" % result.get())
# error is ('timeout error2 ', 9)
# result2 and len: ([], 0)
print('=====if====1')
# print('=====while====2')
# while result.ready():
# if result.successful():
# result2.append(result.get())
# break
# #while 並不合適在map類的方法中,因爲map類是完成全部線程的運行後才返回數據
# #如果非要使用,while外面需要有 result.successful()、result.get()的提前調用
# #來結束線程結果不能返回的異常。(因爲線程一直是ready==True狀態)
# print('=====while====2')
except Exception as e:
print("error is %s" % str(e))
print("result2 and len: ", (result2, len(result2)))
time2 = time.time()
print("time:", time2 - time1)
# 結果分析
# result2 and len: ([], 0)
# time: 20.790918588638306
# map會跑全部的線程,x==9時觸發異常的等待時間最短
# 異常發生後其他線程並沒有中斷,還是執行最長的sleep(20),導致最後time 爲20s
# 一旦有線程異常,result.get()就得不到數據了
#map
def dome3():
time1 = time.time()
pool = ThreadPool(20)
my_iter = range(50)
try:
result = pool.map(task, my_iter)
pool.close()
pool.join()
print('result : ', result)
except Exception as e:
print("error is %s" % str(e))
time2 = time.time()
print("time:", time2 - time1)
# 結果分析
#error is ('timeout error2 ', 9)
#time: 20.84999394416809
# 也是併發執行,x==9時觸發異常的等待時間最短
# 異常發生後其他線程並沒有中斷,還是執行最長的sleep(20),導致最後time 爲20s
# 一旦有線程異常,result = pool.map(task, my_iter)就會Exception,得不到返回數據
# apply
def dome4():
time1 = time.time()
pool = ThreadPool(20)
results = []
try:
for i in range(50):
result = pool.apply(task, (i,))
results.append(result)
print(results)
except Exception as e:
print("error is %s" % str(e))
time2 = time.time()
print("time:", time2 - time1)
#結果分析
#error is ('timeout error1 ', 8)
#time: 11.943454027175903
#一步一步執行,並沒有併發,基本和單線程一樣,異常發生後就不再往下執行
#map_async---上面的map_async其實不需要if判斷的
def dome5():
result2 = []
time1 = time.time()
pool = ThreadPool(20)
my_iter = range(50)
try:
result = pool.map_async(task, my_iter)
# 懷疑會在 x==9 處報異常,因爲sleep時間最短且可能在同一線程池中被處理
# 在 if 後面時間最短處中斷 error is ('timeout error2 ', 9),而與sleep時間有關
# map和map_async 不按 x 的進入順序處理
#下面兩句加不加好像不影響,wait已經等待了
pool.close()
pool.join()
result.wait() # 等待所有線程函數執行完畢
#放在這裏打印可以觸發異常,不然就算子線程異常了,主線程也不會知道
print("ready %s" % result.ready())
print("successful %s" % result.successful())
print("i.get %s" % result.get())
except Exception as e:
print("error is %s" % str(e))
print("result2 and len: ", (result2, len(result2)))
time2 = time.time()
print("time:", time2 - time1)
# 結果分析
#對比dome2,這裏的更簡潔一些,不需要多餘的if判斷,子線程有異常直接接收就好
if __name__ == "__main__":
dome0()
總結:
- 不建議使用while,因爲和if基本一樣的功能,主線程會等待子線程,不需要輪詢
- apply_async中加上join會增加異常發生時總的時間(會把全部的線程都運行了,異常發生時會中斷異常線程,其他線程不影響,但get的結果只收集異常發生前的線程返回)。異常會發生在最快進入sleep的線程中,get返回結果是x最先進入sleep觸發異常前的線程(x=8,sleep(10))的結果,總消耗時間在不加join情況下也以此x爲準,但當加join後總消耗時間會以最長sleep的爲參考,也就是說會運行所有的線程,但get的結果還是第一個異常觸發前的x<8的其他線程的結果
- map_async中加不加join一樣。map總是把全部線程都執行,異常會發生在sleep最短(x=9,sleep(1))的線程中,返回結果時以sleep最短的異常爲準,但總消耗時間以sleep最長的爲準,發生異常就get不到結果
- 無論是map還是apply,若只是多線程發送請求不關注結果,可以不用加result.wait()、result.successful()、result.get(),如果要關注就必須加上,這樣才能接收子線程的異常,這就是異步。