info
Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
is IO:Socket:INET thread safe?
3 views
Skip to first unread message
Jesper Persson
May 29, 2013, 2:16:57 AM5/29/13
to libwww
Hi everybody,
I have a multithreaded program that monitors the content of web pages.
On a newly installed CentOS server I have seen strange behaviour where I get 404 because http request is sent to a wrong ip address not maching the domain that was requested.
I have inserted a print statement in IO:Socket:INET.pm in the procedure _get_addr
to see what ip address the domain is translated to:
sub _get_addr {
my($sock,$addr_str, $multi) = @_;
my @addr;
if ($multi && $addr_str !~ /^\d+(?:\.\d+){3}$/) {
(undef, undef, undef, undef, @addr) = gethostbyname($addr_str);
} else {
my $h = inet_aton($addr_str);
push(@addr, $h) if defined $h;
}
open(FH, ">>/tmp/socket_get_addr.log");
print FH localtime().": addr_str: $addr_str, addr: ".inet_ntoa($addr[0])."\n";
close(FH);
@addr;
}
From the log file:
Wed May 29 07:38:09 2013: addr_str: dk1.siteimprove.com, addr: 93.160.60.57
Wed May 29 07:38:09 2013: addr_str: siteimprove.dk, addr: 93.160.60.57
the ip address of siteimprove.dk is wrong and at the same second just before a lookup of dk1.siteimprove.com gets the same ip address and that is correct.
7 minutes later in the log file the siteimprove.dk domain is looked up correctly:
Wed May 29 07:45:11 2013: addr_str: siteimprove.dk, addr: 93.160.60.87
Manually looking up the ip on then server gets the right ip address.
this entry is in the hosts file:
93.160.60.87 siteimprove.dk
CentOS release 6.3 (Final)
Summary of my perl5 (revision 5 version 10 subversion 1) configuration:
IO:Socket:INET version: $VERSION = "1.31";
Any help would be greatly appreciated.
Regards
Jesper Persson
Francesco Nidito
May 29, 2013, 4:12:56 AM5/29/13
to Jesper Persson, libwww
Unfortunately, gethostbyname is not reentrant.
(maybe a stupid question) Is there any particular reason why you cannot use multiple processes instead?
Typed on a very small keyboard...
(maybe a stupid question) Is there any particular reason why you cannot use multiple processes instead?
Typed on a very small keyboard...
Jesper Persson
May 29, 2013, 5:06:56 AM5/29/13
to Francesco Nidito, libwww
The program has been running for more than 10 years without this beeing an issue - at least I have never been aware of this kind of problem.
Perhaps I only first see it now on the new server because the server is a virtual machine and maybe doesn't have a dedicate cpu and more often will get interrupts on the cpu.
The reason that the program uses processes is that is has a mother process fetching work from the database and communicating the work/result of the work to/from the threads via Thread::Queue.
Am I the first one using threads doing http requests in perl on linux having this problem? Seems odd.
If I would change my program to using processes instead of threads, would gethostbyname then be process safe?
It looks as if gethostbyname has a version that is reentrant: gethostbyname_r.
Any reason that perl doesn't use that version?
Regards
On 29 May 2013 10:12, Francesco Nidito <francesc...@gmail.com> wrote:
Unfortunately, gethostbyname is not reentrant.
(maybe a stupid question) Is there any particular reason why you cannot use multiple processes instead?
Typed on a very small keyboard...
0 new messages