* SMP setups (like the one i have here *grin*).
* This is in comparison to the non-threaded dns, which must monitor the thread sockets
* in a nonblocking fashion, consuming more resources to do so.
+ *
+ * NB: Yes this does scale, thank you. Even with large numbers of connecting clients
+ * in any one timeframe, they wont all connect *at the same time* therefore any argument
+ * of "but there will be thousands of threads it'll blow up" is moot, ive tested this and
+ * there will only ever be somewhere around the listen backlog in number of pending
+ * lookups at any one time. This is significant on any modern SMP system.
*/
void* dns_task(void* arg)
{
std::string ip;
int iterations = 0;
- log(DEBUG,"Thread loc 1");
if (dns1.ReverseLookup(inet_ntoa(u->ip4),false))
{
- log(DEBUG,"Thread loc 2");
/* FIX: Dont make these infinite! */
while ((!dns1.HasResult()) && (++iterations < 20))
usleep(100);
- log(DEBUG,"Thread loc 3");
if (iterations < 20)
{
- log(DEBUG,"Thread loc 4");
if (dns1.GetFD() != -1)
{
host = dns1.GetResult();
if (host != "")
{
- log(DEBUG,"Thread loc 5");
if (dns2.ForwardLookup(host, false))
{
iterations = 0;
while ((!dns2.HasResult()) && (++iterations < 20))
usleep(100);
- log(DEBUG,"Thread loc 6");
if (iterations < 20)
{
if (dns2.GetFD() != -1)
{
- log(DEBUG,"Thread loc 7");
ip = dns2.GetResultIP();
if (ip == std::string(inet_ntoa(u->ip4)))
{
- log(DEBUG,"Thread loc 8");
if (host.length() < 65)
{
- log(DEBUG,"Thread loc 9");
if ((fd_ref_table[thisfd] == u) && (fd_ref_table[thisfd]))
{
- log(DEBUG,"Thread loc 10");
if (!u->dns_done)
{
- log(DEBUG,"Thread loc 11");
strcpy(u->host,host.c_str());
if ((fd_ref_table[thisfd] == u) && (fd_ref_table[thisfd]))
{
}
}
}
- log(DEBUG,"Thread loc 12");
if ((fd_ref_table[thisfd] == u) && (fd_ref_table[thisfd]))
u->dns_done = true;
log(DEBUG,"THREAD EXIT");