Random numbers in forked processes

perl -E 'say rand'
0.933151337857456

$ perl -E 'say rand(6)'
2.20041864735506

Quick solution

Call srand(); in the child process, immediately after forking.

Fork and rand

If we create a small script that will fork n=3 times and call rand() in each child process we can see it nicely prints out 3 random numbers: (tried in 5.18.2)

examples/rand_in_fork.pl

use strict;
use warnings;
use 5.010;

#say "Parent $$ generated " . rand();
my $n = 3;
for (1 .. $n) {
    my $pid = fork();
    if (not $pid) {
       srand();
       say "Child  $$ generated " . rand();
       exit;
    }
}

for (1 .. $n) {
   wait();
}

Child 12943 generated  0.0551759396485103
Child 12944 generated  0.0385383724506063
Child 12945 generated  0.804002655510889

What happens if we change the code slightly enabling that line that was commented out earlier?

Parent 12956 generated 0.873379828947694
Child  12957 generated 0.962009316167922
Child  12958 generated 0.962009316167922
Child  12959 generated 0.962009316167922

Now the parent generates a random number before the forks happen, but then the 3 child process return the same random number.

Calling srand

The problem is that actually rand() returns a pre-defined series of numbers. The reason they look random to us is because every time we run our script, it will start from a different location in the series.

The thing that controls the starting point is the srand() function. Try the following:

$ perl -E 'srand(42); say rand()'
0.744525000061007

$ perl -E 'srand(42); say rand()'
0.744525000061007

$ perl -E 'srand(42); say rand()'
0.744525000061007

$ perl -E 'srand(42); say rand()'
0.744525000061007

Every run of the script gives you the same "random" number.

If on every call we call srand with a different number, rand() will also return us a different "random" number.

$ perl -E 'srand(1); say rand()'
0.0416303447718782

$ perl -E 'srand(2); say rand()'
0.912432653437467

$ perl -E 'srand(3); say rand()'
0.783234962103055

$ perl -E 'srand(4); say rand()'
0.654037270768644

Calling srand with a random number

That's good, but how can we generate a random input to srand(). We can't really, but we can get something much better. We can pass it the current time, maybe combined with the current process id (provided in $$).

$ perl -E 'srand(time+$$); say rand()'
0.642138507058011

$ perl -E 'srand(time+$$); say rand()'
0.512940815723599

$ perl -E 'srand(time+$$); say rand()'
0.383743124389188

$ perl -E 'srand(time+$$); say rand()'
0.866952359051542

The problem explained

Perl calls srand() on your behalf when you first call rand() that's how in a normal script the numbers seem to be random, and they are different in every run of the script. In our first example we called rand() first time in the child process and thus each process called srand() for us. The numbers were random.

In the second example we called rand() in the parent process already and the child processes did not call it again. So they all followed the random series from the same point. That's why they returned the same "random" numbers.

The solution is to call srand() or srand(time+$$) in the child process immediately after forking.

use strict;
use warnings;
use 5.010;

say "Parent $$ generated " . rand();
my $n = 3;
for (1 .. $n) {
    my $pid = fork();
    if (not $pid) {
       srand();
       say "Child  $$ generated " . rand();
       exit;
    }
}

for (1 .. $n) {
   wait();
}

The result is:

Parent 13050 generated 0.873155701746331
Child  13052 generated 0.73201028381391
Child  13053 generated 0.61498113216749
Child  13051 generated 0.877628553033905

Warning

Let me just repeat here what the documentation of rand also says:

"rand()" is not cryptographically secure. You should not rely on it in security-sensitive situations.