I have a program with 1 process that starts a lot of threads.
Each thread might use subprocess.Popen to run some command.
I see that the time to run the command increases with the number of threads.
Example:
>>> def foo():
... s = time.time()
... subprocess.Popen('ip link show'.split(), stdout=subprocess.PIPE, stderr=subprocess.PIPE, universal_newlines=True).communicate()
... print(time.time() - s)
...
>>> foo()
0.028950929641723633
>>> [threading.Thread(target=foo).start() for _ in range(10)]
0.058995723724365234
0.07323050498962402
0.09158825874328613
0.11541390419006348 # !!!
0.08147192001342773
0.05238771438598633
0.0950784683227539
0.10175108909606934 # !!!
0.09703755378723145
0.06497764587402344
Is there another way of executing a lot of commands from single process in parallel that doesn't decrease the performance?
Python's threads are, of course, concurrent, but they do not really run in parallel because of the GIL. Therefore, they are not suitable for CPU-bound applications. If you need to truly parallelize something and allow it to run on all CPU cores, you will need to use multiple processes. Here is a nice answer discussing this in more detail: What are the differences between the threading and multiprocessing modules?.
For the above example, multiprocessing.pool may be a good choice (note that there is also a ThreadPool available in this module).
from multiprocessing.pool import Pool
import subprocess
import time
def foo(*args):
s = time.time()
subprocess.Popen('ip link show'.split(), stdout=subprocess.PIPE, stderr=subprocess.PIPE, universal_newlines=True).communicate()
return time.time() - s
if __name__ == "__main__":
with Pool(10) as p:
result = p.map(foo, range(10))
print(result)
# [0.018695592880249023, 0.009021520614624023, 0.01150059700012207, 0.02113938331604004, 0.014114856719970703, 0.01342153549194336, 0.011168956756591797, 0.014746427536010742, 0.013572454452514648, 0.008752584457397461]
result = p.map_async(foo, range(10))
print(result.get())
# [0.00636744499206543, 0.011589527130126953, 0.010645389556884766, 0.0070612430572509766, 0.013571739196777344, 0.009610414505004883, 0.007040739059448242, 0.010993719100952148, 0.012415409088134766, 0.0070383548736572266]
However, if your function is similar to the example in that it mostly just launches other processes and doesn't do a lot of calculations - I doubt parallelizing it will make much of a difference because the subprocesses can already run in parallel. Perhaps the slowdown occurs because your whole system gets overwhelmed for a moment because of all those processes (could be CPU usage is high or too many disk reads/writes are attempted within a short time). I would suggest taking a close look at system resources (Task Manager etc.) while running the program.
maybe it has nothing to do with python: Opening a new shell = opening a new file since basically everything is a file on linux
take a look at your limit for open files with this command (default is 1024):
ulimit
and try to raise it with this command to see if your code gets faster :
ulimit -n 2048
Related
I am measuring the metrics of an encryption algorithm that I designed. I have declared 2 functions and a brief sample is as follows:
import sys, random, timeit, psutil, os, time
from multiprocessing import Process
from subprocess import check_output
pid=0
def cpuUsage():
global running
while pid == 0:
time.sleep(1)
running=true
p = psutil.Process(pid)
while running:
print(f'PID: {pid}\t|\tCPU Usage: {p.memory_info().rss/(1024*1024)} MB')
time.sleep(1)
def Encryption()
global pid, running
pid = os.getpid()
myList=[]
for i in range(1000):
myList.append(random.randint(-sys.maxsize,sys.maxsize)+random.random())
print('Now running timeit function for speed metrics.')
p1 = Process(target=metric_collector())
p1.start()
p1.join()
number=1000
unit='msec'
setup = '''
import homomorphic,random,sys,time,os,timeit
myList={myList}
'''
enc_code='''
for x in range(len(myList)):
myList[x] = encryptMethod(a, b, myList[x], d)
'''
dec_code='''
\nfor x in range(len(myList)):
myList[x] = decryptMethod(myList[x])
'''
time=timeit.timeit(setup=setup,
stmt=(enc_code+dec_code),
number=number)
running=False
print(f'''Average Time:\t\t\t {time/number*.0001} seconds
Total time for {number} Iters:\t\t\t {time} {unit}s
Total Encrypted/Decrypted Values:\t {number*len(myList)}''')
sys.exit()
if __name__ == '__main__':
print('Beginning Metric Evaluation\n...\n')
p2 = Process(target=Encryption())
p2.start()
p2.join()
I am sure there's an implementation error in my code, I'm just having trouble grabbing the PID for the encryption method and I am trying to make the overhead from other calls as minimal as possible so I can get an accurate reading of just the functionality of the methods being called by timeit. If you know a simpler implementation, please let me know. Trying to figure out how to measure all of the metrics has been killing me softly.
I've tried acquiring the pid a few different ways, but I only want to measure performance when timeit is run. Good chance I'll have to break this out separately and run it that way (instead of multiprocessing) to evaluate the function properly, I'm guessing.
There are at least three major problems with your code. The net result is that you are not actually doing any multiprocessing.
The first problem is here, and in a couple of other similar places:
p2 = Process(target=Encryption())
What this code passes to Process is not the function Encryption but the returned value from Encryption(). It is exactly the same as if you had written:
x = Encryption()
p2 = Process(target=x)
What you want is this:
p2 = Process(target=Encryption)
This code tells Python to create a new Process and execute the function Encryption() in that Process.
The second problem has to do with the way Python handles memory for Processes. Each Process lives in its own memory space. Each Process has its own local copy of global variables, so you cannot set a global variable in one Process and have another Process be aware of this change. There are mechanisms to handle this important situation, documented in the multiprocessing module. See the section titled "Sharing state between processes." The bottom line here is that you cannot simply set a global variable inside a Process and expect other Processes to see the change, as you are trying to do with pid. You have to use one of the approaches described in the documentation.
The third problem is this code pattern, which occurs for both p1 and p2.
p2 = Process(target=Encryption)
p2.start()
p2.join()
This tells Python to create a Process and to start it. Then you immediately wait for it to finish, which means that your current Process must stop at that point until the new Process is finished. You never allow two Processes to run at once, so there is no performance benefit. The only reason to use multiprocessing is to run two things at the same time, which you never do. You might as well not bother with multiprocessing at all since it is only making your life more difficult.
Finally I am not sure why you have decided to try to use multiprocessing in the first place. The functions that measure memory usage and execution time are almost certainly very fast, and I would expect them to be much faster than any method of synchronizing one Process to another. If you're worried about errors due to the time used by the diagnostic functions themselves, I doubt that you can make things better by multiprocessing. Why not just start with a simple program and see what results you get?
When using multiprocessing.Pool in python 3.6 or 3.7 with maxtasksperchild=1, I noticed that some processes spawned by the pool are hanging and do not quit, even though the callback to their tasks was already executed. As a result, Pool.join() will block forever, even though all tasks are finished. In the process tree, running but idle child processes can be seen. The problem does not occur if maxtasksperchild=None.
The problem seems to be related to what the callback precisely does. The docs point out that the callback "should return immediately", as it will block other threads managing the pool.
A minimal example to reproduce this behavior on my machine is as follows: (Give it a few tries or increase the number of tasks when it does not block forever.)
from multiprocessing import Pool
from os import getpid
from random import random
from time import sleep
def do_stuff():
pass
def cb(arg):
sleep(random()) # can be replaced with print('foo')
p = Pool(maxtasksperchild=1)
number_of_tasks = 100 # a value may depend on your machine -- for mine 20 is sufficient to trigger the behavior
for i in range(number_of_tasks):
p.apply_async(do_stuff, callback=cb)
p.close()
print("joining ... (this should take just seconds)")
print("use the following command to watch the process tree:")
print(" watch -n .2 pstree -at -p %i" % getpid())
p.join()
Contrary to what I expected, p.join() in the last line will block forever even though do_stuff and cb were both called 100 times.
I am aware that sleep(random()) is in violation of the docs, but is print() also taking 'too long'? The way the docs are written suggest that a non-blocking callback function is required for performance and efficiency and make not clear that a 'slow' callback function will break the pool entirely.
Is print() forbidden in any multiprocessing.Pool callback function? (How to replace that functionality? What is "returning immediately", what is not?)
If yes, should the python documentation be updated to make this clear?
If yes, is it good python practice to rely on "fast" execution of python threads? Does this violate the rule that one should not make assumptions on execution order of threads?
Should I report this to the python bug tracker?
I have written a simple Python script to test overlapping between I/O-bound and cpu-bound threads. The code is here:
from datetime import datetime
import threading
import shutil
import os
def cpuJob(start,end):
counter=start
sum=0
while counter<=end:
sum+=counter
counter+=1
return sum
def ioJob(from_path, to_path):
if os.path.exists(to_path):
shutil.rmtree(to_path)
shutil.copytree(from_path, to_path)
startTime=datetime.now()
Max=120000000
threadCount=2
if threadCount==1:
t1 = threading.Thread(target=cpuJob, args=(1,Max))
# t1 = threading.Thread(target=ioJob, args=(1,Max))
t1.start()
t1.join()
else:
t1 = threading.Thread(target=ioJob, args=("d:\\1","d:\\2"))
t2 = threading.Thread(target=cpuJob, args=(1,Max))
t1.start()
t2.start()
t1.join()
t2.join()
endTime=datetime.now()
diffTime = endTime - startTime
print("Execution time for " , threadCount , " threads is: " , diffTime)
If I run the threads separately (threadCount==1), each thread takes around 12-13 seconds to be finished on my Windows laptop. But when I run them toghegher (threadCount==2), it takes around 20-22 seconds. As far as I know, Python releases the GIL before doing any blocking I/O operations. If GIL is released before working with I/O, why I get such a performance in the code?
Edit 1: As suggested in the commnets, I checked the code of shutils. It seems that in the implementation of this package, the GIL is not released. Why that is the case? The code of shell utility package should fall outside of the Python runtime implementation, No?
... why I get such performance ?
See https://docs.python.org/3/library/threading.html:
CPython implementation detail: In CPython, due to the Global Interpreter Lock, only one thread can execute Python code at once (even though certain performance-oriented libraries might overcome this limitation). If you want your application to make better use of the computational resources of multi-core machines, you are advised to use multiprocessing or concurrent.futures.ProcessPoolExecutor. However, threading is still an appropriate model if you want to run multiple I/O-bound tasks simultaneously.
Your code runs in a non-preemptive framework, yet it never yields control, not until it exits. So another thread won't be scheduled until then. You used some thread machinery, but you may as well have written a two-line sequential function that calls io_job() followed by cpu_job().
What you're looking for is multiprocessing.
Also, if you literally want to copy file trees around with tools like rsync, consider using gmake -jN or GNU parallel (sudo apt install parallel). Here's an example command:
$ find . -name '*.txt' -type f | parallel gzip -v9
Both make and /usr/bin/parallel let you specify number of simultaneous workers, and will continue to draw a new task from the queue each time a worker completes a task.
According to /usr/lib/python3.6/shutil.py on my machine, it seems these functions rmtree, copytree etc. are implemented as Python code like _rmtree_unsafe. The underlying API behind rmtree etc. are like os.listdir and os.unlink.
Due to the restriction of Python GIL, only one thread can run Python code at a time. Therefore your cpuJob and ioJob cannot run simultaneously (parallel) because both are pure Python code, so you don't observe any performance improvement when you try to run them as "threads".
I have an array of data to handle and handler that executing long (1-2 minutes) and takes a lot of memory for its calculations.
raw = ['a', 'b', 'c']
def handler():
# do something long
Since handler requires a lot of memory, I want to execute it in separate subprocess and kill it after execution to release memory. Something like the following snippet:
from multiprocessing import Process
for r in raw:
process = Process(target=handler, args=(r))
process.start()
The problem is that such approach leads to immediate running len(raw) processes. And it's not good.
Also, it's not needed to interchange any kind of data between subprocesses. Just run them consequently.
Therefore it would be great to run a few processes at the same time and add a new one once existing finishes.
How could it be implemented (if it's even possible)?
to run your processes sequentially, just join each process within the loop:
from multiprocessing import Process
for r in raw:
process = Process(target=handler, args=(r))
process.start()
process.join()
that way you're sure that only one process is running at the same time (no concurrency)
That's the simplest way. To run more than one process but limit the number of processes running at the same time, you can use a multiprocessing.Pool object and apply_async
I've built a simple example which computes the square of the argument, and simulates an heavy processing:
from multiprocessing import Pool
import time
def target(r):
time.sleep(5)
return(r*r)
raw = [1,2,3,4,5]
if __name__ == '__main__':
with Pool(3) as p: # 3 processes at a time
reslist = [p.apply_async(target, (r,)) for r in raw]
for result in reslist:
print(result.get())
Running this I get:
<5 seconds wait, time to compute the results>
1
4
9
<5 seconds wait, 3 processes max can run at the same time>
16
25
I'm trying to understand Python's multiprocessing, and have devised the following code to test it:
import multiprocessing
def F(n):
if n == 0: return 0
elif n == 1: return 1
else: return F(n-1)+F(n-2)
def G(n):
print(f'Fibbonacci of {n}: {F(n)}')
processes = []
for i in range(25, 35):
processes.append(multiprocessing.Process(target=G, args=(i, )))
for pro in processes:
pro.start()
When I run it, I tells me that the computing time was roughly of 6.65s.
I then wrote the following code, which I thought to be functionally equivalent to the latter:
from multiprocessing.dummy import Pool as ThreadPool
def F(n):
if n == 0: return 0
elif n == 1: return 1
else: return F(n-1)+F(n-2)
def G(n):
print(f'Fibbonacci of {n}: {F(n)}')
in_data = [i for i in range(25, 35)]
pool = ThreadPool(10)
results = pool.map(G, in_data)
pool.close()
pool.join()
and its running time was almost 12s.
Why is it that the second takes almost twice as the first one? Aren't they supposed to be equivalent?
(NB. I'm running Python 3.6, but also tested a similar code on 3.52 with same results.)
The reason the second takes twice as long as the first is likely due to the CPython Global Interpreter Lock.
From http://python-notes.curiousefficiency.org/en/latest/python3/multicore_python.html:
[...] the GIL effectively restricts bytecode execution to a single core, thus rendering pure Python threads an ineffective tool for distributing CPU bound work across multiple cores.
As you know, multiprocessing.dummy is a wrapper around the threading module, so you're creating threads, not processes. The Global Interpreter Lock, with a CPU-bound task as here, is not much different than simply executing your Fibonacci calculations sequentially in a single thread (except that you've added some thread-management/context-switching overhead).
With the "true multiprocessing" version, you only have a single thread in each process, each of which is using its own GIL. Hence, you can actually make use of multiple processors to improve the speed.
For this particular processing task, there is no significant advantage to using multiple threads over multiple processes. If you only have a single processor, there is no advantage to using either multiple processes or multiple threads over a single thread/process (in fact, both merely add context-switching overhead to your task).
(FWIW: A join in the true multiprocessing version is apparently being done automatically by the python runtime so adding an explicit join doesn't seem to make any difference in my tests using time(1). And, by the way, if you did want to add join, you should add a second loop for the join processing. Adding join to the existing loop will simply serialize your processes.)