locking files on Linux

[andrea crotti]

I'm trying to understand how I can lock a file while writing on it,
because I might have multiple processes working on it at the same time.

I found the fcntl.lockf function but if I do this:

In [109]: locked = open('locked.txt', 'w')

In [110]: fcntl.lockf(locked, fcntl.LOCK_EX)

I can happily open the file with vim from somewhere and write on it, so
it doesn't seem to be very useful, or am I missing something?

I can otherwise use the classic ".lock" file when working on it, but I
would be glad to see a smarter solution (should only work on Linux)..

 

[Grant Edwards]

I'm trying to understand how I can lock a file while writing on it,
because I might have multiple processes working on it at the same time.

I found the fcntl.lockf function but if I do this:

In [109]: locked = open('locked.txt', 'w')

In [110]: fcntl.lockf(locked, fcntl.LOCK_EX)

I can happily open the file with vim from somewhere and write on it, so
it doesn't seem to be very useful, or am I missing something?

File locks under Unix have historically been "advisory". That means
that programs have to _choose_ to pay attention to them. Most
programs do not.

Linux does support mandatory locking, but it's rarely used and must be
manually enabled at the filesystem level. It's probably worth noting
that in the Linux kernel docs, the document on mandatory file locking
begins with a section titled "Why you should avoid mandatory locking".

http://en.wikipedia.org/wiki/File_locking#In_Unix-like_systems
http://kernel.org/doc/Documentation/filesystems/locks.txt
http://kernel.org/doc/Documentation/filesystems/mandatory-locking.txt
http://www.thegeekstuff.com/2012/04/linux-file-locking-types/
http://www.hackinglinuxexposed.com/articles/20030623.html

 

[andrea crotti]

File locks under Unix have historically been "advisory". That means
that programs have to _choose_ to pay attention to them. Most
programs do not.

Linux does support mandatory locking, but it's rarely used and must be
manually enabled at the filesystem level. It's probably worth noting
that in the Linux kernel docs, the document on mandatory file locking
begins with a section titled "Why you should avoid mandatory locking".

http://en.wikipedia.org/wiki/File_locking#In_Unix-like_systems
http://kernel.org/doc/Documentation/filesystems/locks.txt
http://kernel.org/doc/Documentation/filesystems/mandatory-locking.txt
http://www.thegeekstuff.com/2012/04/linux-file-locking-types/
http://www.hackinglinuxexposed.com/articles/20030623.html

--
Grant Edwards grant.b.edwards Yow! Your CHEEKS sit like
at twin NECTARINES above
gmail.com a MOUTH that knows no
BOUNDS --

Uhh I see thanks, I guess I'll use the good-old .lock file (even if it
might have some problems too).

Anyway I'm only afraid that my same application could modify the
files, so maybe I can instruct it to check if the file is locked.

Or maybe using sqlite would work even if writing from different
processes?

I would prefer to keep something human readable as INI-format though,
rather then a sqlite file..

Thanks

 

[Grant Edwards]

Uhh I see thanks, I guess I'll use the good-old .lock file (even if
it might have some problems too).

Anyway I'm only afraid that my same application could modify the
files, so maybe I can instruct it to check if the file is locked.

If what you're guarding against is multiple instances of your
application modifying the file, then either of the advisory file
locking schemes or the separate lock file should work fine.

 

[Oscar Benjamin]

Uhh I see thanks, I guess I'll use the good-old .lock file (even if it
might have some problems too).

I think you've misunderstood what Grant meant.

Anyway I'm only afraid that my same application could modify the
files, so maybe I can instruct it to check if the file is locked.

In that case fcntl will work for you. The point is that fcntl only
locks the file if all of the applications accessing the file use
fcntl. Any other application such as vim can simply ignore the fcntl
lock. Have a read of the links that Grant posted.

Did you try writing twice from the same application that uses fcntl?

Or maybe using sqlite would work even if writing from different
processes?

That would also work.


Oscar

 

[andrea crotti]

If what you're guarding against is multiple instances of your
application modifying the file, then either of the advisory file
locking schemes or the separate lock file should work fine.

Ok so I tried a small example to see if I can make it fail, but this
below just works perfectly fine.

Maybe it's too fast and it release the file in time, but I would
expect it to take some time and fail instead..

import fcntl

from multiprocessing import Process

FILENAME = 'file.txt'


def long_text():
return ('some text' * (100 * 100))


class Locked:
def __init__(self, fileobj):
self.fileobj = fileobj

def __enter__(self):
# any problems here?
fcntl.lockf(self.fileobj, fcntl.LOCK_EX)
return self.fileobj

def __exit__(self, type, value, traceback):
fcntl.lockf(self.fileobj, fcntl.LOCK_UN)


def write_to_file():
with open(FILENAME, 'w') as to_lock:
with Locked(to_lock):
to_lock.write(long_text())


if __name__ == '__main__':
Process(target=write_to_file).start()
Process(target=write_to_file).start()

 

[Oscar Benjamin]

Ok so I tried a small example to see if I can make it fail, but this
below just works perfectly fine.

Maybe it's too fast and it release the file in time, but I would
expect it to take some time and fail instead..

Why not come up with a test that actually shows you if it works? Here
are two suggestions:

1) Use time.sleep() so that you know how long the lock is held for.
2) Write different data into the file from each process and see what
you end up with.

import fcntl

from multiprocessing import Process

FILENAME = 'file.txt'


def long_text():
return ('some text' * (100 * 100))


class Locked:
def __init__(self, fileobj):
self.fileobj = fileobj

def __enter__(self):
# any problems here?
fcntl.lockf(self.fileobj, fcntl.LOCK_EX)
return self.fileobj

def __exit__(self, type, value, traceback):
fcntl.lockf(self.fileobj, fcntl.LOCK_UN)


def write_to_file():
with open(FILENAME, 'w') as to_lock:

I don't think it will work if you truncate the file like this. This
will empty the file *before* checking for the lock. Try opening the
file for reading and writing (without truncating).

with Locked(to_lock):
to_lock.write(long_text())


if __name__ == '__main__':
Process(target=write_to_file).start()
Process(target=write_to_file).start()

Oscar

 

[andrea crotti]

Why not come up with a test that actually shows you if it works? Here
are two suggestions:

1) Use time.sleep() so that you know how long the lock is held for.
2) Write different data into the file from each process and see what
you end up with.

Ok thanks I will try, but I thought that what I did was the worst
possible case, because I'm opening and writing on the same file from
two different processes, locking the file with LOCK_EX.

It should not open it at all as far as I understood...

 

[Oscar Benjamin]

Ok thanks I will try, but I thought that what I did was the worst
possible case, because I'm opening and writing on the same file from
two different processes, locking the file with LOCK_EX.

It should not open it at all as far as I understood...

I don't think you have understood. Read the link that Grant posted:
http://en.wikipedia.org/wiki/File_locking#In_Unix-like_systems

And my other comment:

I don't think it will work if you truncate the file like this. This
will empty the file *before* checking for the lock. Try opening the
file for reading and writing (without truncating).

The lock is cooperative. It does not prevent the file from being
opened or overwritten. It only prevents any other process from
obtaining the lock. Here you open the file with mode 'w' which
truncates the file instantly (without checking for the lock).


Oscar

 

[andrea crotti]

The lock is cooperative. It does not prevent the file from being
opened or overwritten. It only prevents any other process from
obtaining the lock. Here you open the file with mode 'w' which
truncates the file instantly (without checking for the lock).


Oscar

Very good thanks now I understood, actually my problem was in the
assumption that it should fail when the lock is already taken, but by
default lockf just blocks until the lock is released.

It seems to work quite nicely so I'm going to use this..

 

[Nobody]

Uhh I see thanks, I guess I'll use the good-old .lock file (even if it
might have some problems too).

In which case, you don't see. A lock file is also advisory, i.e. it only
affects applications which explicitly check for a lock file.

Historically, the advantage of lock files was that they worked on NFS
implementations which didn't implement locking (it's a long-standing Unix
joke that "NFS" stands for "Not a FileSystem", because it failed to
conform to established filesystem semantics).

Nowadays, NFS implementations which don't support locking are sufficiently
rare that they can safely be ignored. So lock files don't offer any
advantages, and one fairly obvious disadvantage (the possibility of a
"stale" lock file if the program terminates unexpectedly without removing
the lock file).

For any form of advisory locking, the one thing which matters is that all
progams which access the file agree on the mechanism used, i.e. whether to
use lockf(), fcntl(), flock() (locks created by one mechanism may or may
not be recognised by the others), or lock files, and in the case of lock
files, the naming convention which is used.

If the file is specific to a particular program, and you just need to
protect against multiple instances of that program, you can use whichever
mechanism you wish, and would be strongly advised to use kernel locks
(fcntl() is the most portable, followed by lockf(); flock() is a BSD-ism).