Actions
Bug #11443
closedDNS name resolution takes twice as long as it should when primary name server is unavailable
Status:
Closed
Assignee:
-
Target version:
-
ruby -v:
ruby 2.3.0dev (2015-08-13) [x86_64-linux]
Description
On linux /etc/resolv.conf can be configured to set the timeout of a nameserver before trying the next one (defaults to 5 seconds). So in the event of a failed name server we expect the time taken to resolve a hostname will be timeout + query time. However testing shows it takes (2 * timeout) + query time. Example resolv.conf to replicate the problem of a failed name server:
nameserver 1.1.1.1
nameserver 8.8.8.8
nameserver 8.8.4.4
Example ruby code to test:
require 'net/protocol'
require 'uri'
require 'benchmark'
uri = URI.parse("http://google.com")
Benchmark.measure { TCPSocket.new(uri.host, uri.port) }.real
Example benchmark with a above resolv.conf:
irb(main):005:0> Benchmark.measure { TCPSocket.new(uri.host, uri.port) }.real
=> 10.593100978
Example benchmark with timeout set to 3 in resolv.conf (options timeout:3):
irb(main):005:0> Benchmark.measure { TCPSocket.new(uri.host, uri.port) }.real
=> 6.262138267
and without the invalid nameserver:
irb(main):005:0> Benchmark.measure { TCPSocket.new(uri.host, uri.port) }.real
=> 0.243385728
Example C code that proves the resolver is behaving as expected:
#include <stdio.h>
#include <netdb.h>
int main(int argc, char *argv[])
{
struct hostent *hstnm;
if (argc != 2) {
fprintf(stderr, "usage: %s hostname\n", argv[0]);
return 1;
}
hstnm = gethostbyname (argv[1]);
if (!hstnm)
return 1;
printf ("Name: %s\n", hstnm->h_name);
return 0;
}
Benchmark of C code with failed nameserver:
$ time ./gethostbyname google.com
Name: google.com
real 0m5.189s
user 0m0.000s
sys 0m0.000s
Benchmark of C code without failed nameserver:
$ time ./gethostbyname google.com
Name: google.com
real 0m0.040s
user 0m0.000s
sys 0m0.000s
Actions
Like0
Like0Like0Like0Like0Like0