How do I run os.walk in parallel in Python?

Question

I wrote a simple app in Java that takes a list of paths and generates a file with all the file paths under that original list.

If I have paths.txt that has:

c:\folder1\
c:\folder2\
...
...
c:\folder1000\

My app runs the recursive function on each path multithreaded, and returns a file with all the file paths under these folders.

Now I want to write this app in Python.

I've written a simple app that uses os.walk() to run through a given folder and print the filepaths to output.

Now I want to run it in parallel, and I've seen that Python has some modules for this: multithreaded and multiprocessing.

What is the best what to do this? And within that way, how is it performed?

Answer 1

Here is a multiprocessing solution:

from multiprocessing.pool import Pool
from multiprocessing import JoinableQueue as Queue
import os

def explore_path(path):
    directories = []
    nondirectories = []
    for filename in os.listdir(path):
        fullname = os.path.join(path, filename)
        if os.path.isdir(fullname):
            directories.append(fullname)
        else:
            nondirectories.append(filename)
    outputfile = path.replace(os.sep, '_') + '.txt'
    with open(outputfile, 'w') as f:
        for filename in nondirectories:
            print >> f, filename
    return directories

def parallel_worker():
    while True:
        path = unsearched.get()
        dirs = explore_path(path)
        for newdir in dirs:
            unsearched.put(newdir)
        unsearched.task_done()

# acquire the list of paths
with open('paths.txt') as f:
    paths = f.split()

unsearched = Queue()
for path in paths:
    unsearched.put(path)

pool = Pool(5)
for i in range(5):
    pool.apply_async(parallel_worker)

unsearched.join()
print 'Done'

Answer 2

This is a pattern for threads in python which has been useful to me. I'm not sure if threading is going to increase your performance due to the way threads work in CPython though.

import threading
import Queue
import os

class PathThread (threading.Thread):
    def __init__(self, queue):
        threading.Thread.__init__(self)
        self.queue = queue

    def printfiles(self, p):
        for path, dirs, files in os.walk(p):
            for f in files:
                print path + "/" + f

    def run(self):
        while True:
            path = self.queue.get()
            self.printfiles(path)
            self.queue.task_done()

# threadsafe queue
pathqueue = Queue.Queue()
paths = ["foo", "bar", "baz"]

# spawn threads
for i in range(0, 5):
    t = PathThread(pathqueue)
    t.setDaemon(True)
    t.start()

# add paths to queue
for path in paths:
    pathqueue.put(path)

# wait for queue to get empty
pathqueue.join()

Answer 3

Even threading can be pretty helpful for directory traversal. I use the following code for traversing a SharePoint tree, getting a pretty significant speedup for about 50 threads. This particular program returns (path, data) pairs for all xml files in a directory structure, and can be simply expanded for your use. (This is cut and pasted from my program; some additional editing is needed.)

#unique string for error passing error messages
ERROR = '\xffERROR\xff'

class ScanWorker(threading.Thread):
    """Worker class for scanning directory structures.
    pathQueue: queue for pathnames of directories
    resultQueue: results of processFile, pairs of (path, data) to be updated
    """
    lock = threading.Lock()
    dirCount = 0
    def __init__(self, pathQueue, resultQueue):
        self.pathQueue = pathQueue
        self.resultQueue = resultQueue
        super().__init__()

    def run(self):
        """Worker thread.
        Get a directory, process it, and put new directories on the
        queue."""
        try:
            while True:
                self.processDir(self.pathQueue.get())
                self.pathQueue.task_done()
        except Exception as e:
            #pass on exception to main thread
            description = traceback.format_exception(*sys.exc_info())
            description.insert(0,
                "Error in thread {}:\n".format(
                    threading.current_thread().name))
            self.resultQueue.put((ERROR, description))
            self.pathQueue.task_done()

    def processDir(self, top):
        """Visit a directory
        Call self.processFile on every file, and queue the directories.
        """
        #Wait and retry a few times in case of network errors.
        #SharePoint is not reliable, gives errors for no reason
        for retryCount in range(30):
            try:
                names = listdir(top)
                break
            except OSError as e:
                if e.errno in (2,22):
                    lastError = e
                    print(end="L", flush=True)
                    time.sleep(1)
                else:
                    raise
        else:
            print("List: too many retries")
            raise lastError
        #it is not important to worry about race conditions here
        self.__class__.dirCount += 1
        #process contents
        for name in names:
            if isdir(join(top, name)): self.pathQueue.put(join(top, name))
            else: self.processFile(join(top, name))

    def processFile(self, path):
        """Get XML file.
        """
        #only xml files
        if not path.lower().endswith('.xml'): return
        filemtime = datetime.fromtimestamp(getmtime(path))
        #SharePoint is not reliable, gives errors for no reason; just retry
        for retryCount in range(30):
            try:
                data = open(path,'rb').read()
                break
            except OSError as e:
                if e.errno in (2,22):
                    lastError = e
                    print(end="R", flush=True)
                    time.sleep(1)
                else:
                    raise
        else:
            print("Read: too many retries")
            raise lastError
        self.resultQueue.put((path, data))

class Scanner:
    """Interface to the ScanWorkers
    Sharepoint is pretty fast compared to its delay and handles 50 workers well
    Make sure you only create one instance of Scanner!
    """
    def __init__(self, workers):
        #don't restrict the path queue length; this causes deadlock
        #we use a LIFO queue to get more depth-first like search
        #reducing average queue length and hopefully improving server caching
        self.pathQueue = queue.LifoQueue()
        #this is the output queue to the main thread
        self.resultQueue = queue.Queue(5)
        self.workers = workers
        #start workers
        for i in range(workers):
            t = ScanWorker(self.pathQueue, self.resultQueue)
            t.setDaemon(True)
            t.start()

    def startWorkers(self, path):
        #add counter
        self.added = 0
        #and go
        self.pathQueue.put(path)

    def processResult(self, wait=True):
        """Get an element from the result queue, and add to the zip file."""
        path, data = self.resultQueue.get(block=wait)
        if path==ERROR:
            #process gave alarm; stop scanning
            #pass on description
            raise ScanError(data)
        <do whatever you want to do with the file>
        self.resultQueue.task_done()
        self.added += 1

#main
try:
    #set up
    scanner = Scanner(threads)
    scanner.startWorkers(rootpath)
    pathQueue, resultQueue = scanner.pathQueue, scanner.resultQueue
    #scanner is rolling; wait for it to finish
    with pathQueue.all_tasks_done:
        while pathQueue.unfinished_tasks:
            #tasks are still running
            #process results
            while True:
                try: scanner.processResult(wait=False)
                except queue.Empty: break
            #no new files found; check if scanner is ready
            done = pathQueue.all_tasks_done.wait(timeout=1)
            if not done:
                #Not yet; print something while we wait
                print(
                    "\rProcessed {} files from {} directories [{} {}]  "
                    .format(
                        scanner.added,
                        ScanWorker.dirCount,
                        pathQueue.unfinished_tasks,
                        resultQueue.unfinished_tasks,
                    ), end='\r')
    #just to make sure everybody is ready: join the path queue
    pathQueue.join()
    #process remaining of result queue
    while resultQueue.unfinished_tasks: scanner.processResult(wait=True)
    #go to new line to prevent overwriting progress messages
    print()
except ScanError as e:
    print()
    print(*e.args[0], end='')
    print("Process interrupted.")
except KeyboardInterrupt:
    print("\nProcess interrupted.")
print()