Very fast asynchronous FTP server library
Python FTP server library provides a high-level portable interface to easily write very efficient, scalable and asynchronous FTP servers with Python. It is the most complete RFC-959 FTP server implementation available for Python programming language and it’s used in projects like Google Chromium and Bazaar and included in Debian, Fedora and FreeBSD package repositories.
- Extremely lightweight, fast and scalable (see why and benchmarks).
- Uses sendfile(2) (see pysendfile) system call for uploads.
- Uses epoll() / kqueue() / select() to handle concurrency asynchronously.
- …But can optionally skip to a multiple thread / process model (as in: you’ll be free to block or use slow filesystems).
- Portable: entirely written in pure Python; works with Python from 2.6 to 3.5 by using a single code base.
- Supports FTPS (RFC-4217), IPv6 (RFC-2428), Unicode file names (RFC-2640), MLSD/MLST commands (RFC-3659).
- Support for virtual users and virtual filesystem.
- Extremely flexible system of “authorizers” able to manage both “virtual” and “real” users on on both UNIX and Windows.
- Test coverage close to 100%.
Despite being written in an intepreted language, pyftpdlib has transfer rates superior to most common UNIX FTP servers. It also scales better since whereas vsftpd and proftpd use multiple processes to achieve concurrency, pyftpdlib will only use one process and handle concurrency asynchronously (see the C10K problem). Here are some benchmarks made against my Linux 3.0.0 box, Intel core-duo 3.1 Ghz:
pyftpdlib vs. proftpd 1.3.4
|STOR (client -> server)||585.90 MB/sec||600.49 MB/sec||-0.02x|
|RETR (server -> client)||1652.72 MB/sec||1524.05 MB/sec||+0.08|
|300 concurrent clients (connect, login)||0.19 secs||9.98 secs||+51x|
|STOR (1 file with 300 idle clients)||585.59 MB/sec||518.55 MB/sec||+0.1x|
|RETR (1 file with 300 idle clients)||1497.58 MB/sec||1478.19 MB/sec||0x|
|300 concurrent clients (RETR 10M file)||3.41 secs||3.60 secs||+0.05x|
|300 concurrent clients (STOR 10M file)||8.60 secs||11.56 secs||+0.3x|
|300 concurrent clients (QUIT)||0.03 secs||0.39 secs||+12x|
pyftpdlib vs. vsftpd 2.3.5
|STOR (client -> server)||585.90 MB/sec||611.73 MB/sec||-0.04x|
|RETR (server -> client)||1652.72 MB/sec||1512.92 MB/sec||+0.09|
|300 concurrent clients (connect, login)||0.19 secs||20.39 secs||+106x|
|STOR (1 file with 300 idle clients)||585.59 MB/sec||610.23 MB/sec||-0.04x|
|RETR (1 file with 300 idle clients)||1497.58 MB/sec||1493.01 MB/sec||0x|
|300 concurrent clients (RETR 10M file)||3.41 secs||3.67 secs||+0.07x|
|300 concurrent clients (STOR 10M file)||8.60 secs||9.82 secs||+0.07x|
|300 concurrent clients (QUIT)||0.03 secs||0.01 secs||+0.14x|
For more benchmarks see here.
>>> from pyftpdlib.authorizers import DummyAuthorizer >>> from pyftpdlib.handlers import FTPHandler >>> from pyftpdlib.servers import FTPServer >>> >>> authorizer = DummyAuthorizer() >>> authorizer.add_user("user", "12345", "/home/giampaolo", perm="elradfmwMT") >>> authorizer.add_anonymous("/home/nobody") >>> >>> handler = FTPHandler >>> handler.authorizer = authorizer >>> >>> server = FTPServer(("127.0.0.1", 21), handler) >>> server.serve_forever() [I 13-02-19 10:55:42] >>> starting FTP server on 127.0.0.1:21 <<< [I 13-02-19 10:55:42] poller: <class 'pyftpdlib.ioloop.Epoll'> [I 13-02-19 10:55:42] masquerade (NAT) address: None [I 13-02-19 10:55:42] passive ports: None [I 13-02-19 10:55:42] use sendfile(2): True [I 13-02-19 10:55:45] 127.0.0.1:34178- FTP session opened (connect) [I 13-02-19 10:55:48] 127.0.0.1:34178-[user] USER 'user' logged in. [I 13-02-19 10:56:27] 127.0.0.1:34179-[user] RETR /home/giampaolo/.vimrc completed=1 bytes=1700 seconds=0.001 [I 13-02-19 10:56:39] 127.0.0.1:34179-[user] FTP session closed (disconnect).
A lot of time and effort went into making pyftpdlib as it is right now. If you feel pyftpdlib is useful to you or your business and want to support its future development please consider donating me (Giampaolo Rodola’) some money. I only ask for a small donation, but of course I appreciate any amount.
Don’t want to donate money? Then maybe you could write me a recommendation on Linkedin. In case you’re using pyftpdlib into a software of yours mail me and I’ll add your software to the adoptions list.
Some famous trademarks which adopted pyftpdlib (complete list).