Python Multiprocessing map_async

问题:

I’d like to skip results that are returned from map_async. They are growing in memory but I don’t need them.

Here is some code:

def processLine(line):
    #process something
    print "result"
pool = Pool(processes = 8)
for line in sys.stdin:
    lines.append(line)
    if len(lines) >= 100000:
        pool.map_async(processLine, lines, 2000)
pool.close()
pool.join()

When I have to process file with hundreds of millions of rows, the python process grows in memory to a few gigabytes. How can I resolve that?

Thanks for your help :)

回答1:

Your code has a bug:

for line in sys.stdin:
    lines.append(line)
    if len(lines) >= 100000:
        pool.map_async(processLine, lines, 2000)

This is going to wait until lines accumulates more than 100000 lines. After that, pool.map_async is being called on the entire list of 100000+ lines for each additional line.

It is not clear exactly what you are really trying to do, but if you don't want the return value, use pool.apply_async, not pool.map_async. Maybe something like this:

import multiprocessing as mp

def processLine(line):
    #process something
    print "result"

if __name__ == '__main__':
    pool = mp.Pool(processes = 8)
    for line in sys.stdin:
        pool.apply_async(processLine, args = (line, ))
    pool.close()
    pool.join()

回答2:

Yes you're right. There is some bug

I mean:

def processLine(line):
  #process something
  print "result"
  pool = Pool(processes = 8)

if __name__ == '__main__':
  for line in sys.stdin:
    lines.append(line)
    if len(lines) >= 100000:
      pool.map_async(processLine, lines, 2000)
      lines = [] #to clear buffer
  pool.map_async(processLine, lines, 2000)
  pool.close()
  pool.join()

I used map_async because it has configurable chunk_size so it is more efficient if there are lots of lines which processing time is quite short.