如何在Python中解析大于100GB的文件？

我有大约 100 GB 大小的文本文件，格式如下（具有线路、ip 和域的重复记录）：

domain|ip 
yahoo.com|89.45.3.5
bbc.com|45.67.33.2
yahoo.com|89.45.3.5
myname.com|45.67.33.2
etc.

我尝试使用以下 python 代码解析它们，但仍然出现内存错误。有人知道解析此类文件的更优化方法吗？ （时间对我来说是一个重要因素）

files = glob(path)
for filename in files:
    print(filename)
    with open(filename) as f:
        for line in f:
            try:
                domain = line.split('|')[0]
                ip = line.split('|')[1].strip('\n')
                if ip in d:
                    d[ip].add(domain)
                else:
                    d[ip] = set([domain])
            except:
                print (line)
                pass

    print("this file is finished")

for ip, domains in d.iteritems():
    for domain in domains:
        print("%s|%s" % (ip, domain), file=output)

Python 对象比磁盘上相同的值占用更多的内存；引用计数有一点开销，并且在集合中还需要考虑每个值的缓存哈希值。

不要将所有这些对象读入（Python）内存；使用数据库代替。 Python 附带了一个用于 SQLite 数据库的库，使用它可以将文件转换为数据库。然后您可以从中构建输出文件：

import csv
import sqlite3
from itertools import islice

conn = sqlite3.connect('/tmp/ipaddresses.db')
conn.execute('CREATE TABLE IF NOT EXISTS ipaddress (domain, ip)')
conn.execute('''\
    CREATE UNIQUE INDEX IF NOT EXISTS domain_ip_idx 
    ON ipaddress(domain, ip)''')

for filename in files:
    print(filename)
    with open(filename, 'rb') as f:
        reader = csv.reader(f, delimiter='|')
        cursor = conn.cursor()
        while True:
            with conn:
                batch = list(islice(reader, 10000))
                if not batch:
                    break
                cursor.executemany(
                    'INSERT OR IGNORE INTO ipaddress VALUES(?, ?)',
                    batch)

conn.execute('CREATE INDEX IF NOT EXISTS ip_idx ON ipaddress(ip)')
with open(outputfile, 'wb') as outfh:
    writer = csv.writer(outfh, delimiter='|')
    cursor = conn.cursor()
    cursor.execute('SELECT ip, domain from ipaddress order by ip')
    writer.writerows(cursor)

这会按 10000 个批次处理您的输入数据，并生成一个索引进行排序after插入。生成索引需要一些时间，但它完全适合您的可用内存。

The UNIQUE在开始时创建的索引可确保仅插入唯一的域 - IP 地址对（因此仅跟踪每个 IP 地址的唯一域）；这INSERT OR IGNORE语句会跳过数据库中已存在的任何对。

仅包含您提供的示例输入的简短演示：

>>> import sqlite3
>>> import csv
>>> import sys
>>> from itertools import islice
>>> conn = sqlite3.connect('/tmp/ipaddresses.db')
>>> conn.execute('CREATE TABLE IF NOT EXISTS ipaddress (domain, ip)')
<sqlite3.Cursor object at 0x106c62730>
>>> conn.execute('''\
...     CREATE UNIQUE INDEX IF NOT EXISTS domain_ip_idx 
...     ON ipaddress(domain, ip)''')
<sqlite3.Cursor object at 0x106c62960>
>>> reader = csv.reader('''\
... yahoo.com|89.45.3.5
... bbc.com|45.67.33.2
... yahoo.com|89.45.3.5
... myname.com|45.67.33.2
... '''.splitlines(), delimiter='|')
>>> cursor = conn.cursor()
>>> while True:
...     with conn:
...         batch = list(islice(reader, 10000))
...         if not batch:
...             break
...         cursor.executemany(
...             'INSERT OR IGNORE INTO ipaddress VALUES(?, ?)',
...             batch)
... 
<sqlite3.Cursor object at 0x106c62810>
>>> conn.execute('CREATE INDEX IF NOT EXISTS ip_idx ON ipaddress(ip)')
<sqlite3.Cursor object at 0x106c62960>
>>> writer = csv.writer(sys.stdout, delimiter='|')
>>> cursor = conn.cursor()
>>> cursor.execute('SELECT ip, domain from ipaddress order by ip')
<sqlite3.Cursor object at 0x106c627a0>
>>> writer.writerows(cursor)
45.67.33.2|bbc.com
45.67.33.2|myname.com
89.45.3.5|yahoo.com