Minimizing chance of race conditions

[Duke of Hazard]

It is imperative that a section of my code is only executed by only
one user at a time. So I have my program write a lock file when it
reaches that section , then remove the lock file after it completes
that section.

This seemed to work ok, until I knew about race conditions. So I
changed the code to write the lock file twice. First at the beginning
of the script and second when it reaches the critical section of code
that must be executed by only 1 user at a time.

I figure this will significantly lower my probability of a race
condition since two race conditions will need to occur simultaneously
for it to break (which I am fine with, as long as it happens less than
once per 1000 times).

Is that right, or am I missing something?

Thanks!

 

[Jürgen Exner]

Duke of Hazard wrote:
[...]

I figure this will significantly lower my probability of a race
condition since two race conditions will need to occur simultaneously
for it to break (which I am fine with, as long as it happens less than
once per 1000 times).

Is that right, or am I missing something?

Why don't you use code that does not have a race condition?
"perldoc -q lock"

jue

 

[Tassilo v. Parseval]

Also sprach Duke of Hazard:

It is imperative that a section of my code is only executed by only
one user at a time. So I have my program write a lock file when it
reaches that section , then remove the lock file after it completes
that section.

This seemed to work ok, until I knew about race conditions. So I
changed the code to write the lock file twice. First at the beginning
of the script and second when it reaches the critical section of code
that must be executed by only 1 user at a time.

I figure this will significantly lower my probability of a race
condition since two race conditions will need to occur simultaneously
for it to break (which I am fine with, as long as it happens less than
once per 1000 times).

Is that right, or am I missing something?

It's still wrong. Avoiding race conditions means that the chance of one
occuring is zero. You merely lowered its probability. But if there is
still a chance of a race condition, your code remains broken.

You should only create one lockfile, but do it properly. It has to be an
atomic operation:

select undef, undef, undef, 0.5
while ! sysopen LOCK, "file.lock", O_WRONLY|O_EXCL|O_CREAT;

# once the program reaches this line, it succesfully acquired a
# lock-file

...

close LOCK;
unlink "file.lock" or die "Something happened to the lock file: $!";

The above, by the way, is reasonably portable.

There's a number of FAQ entries dealing with locking. See

perldoc -q lock

Tassilo