[06:26] *** lyarick2 has joined #arpnetworks
[08:36] <lyarick2> anyone have advice for a load-balanced dns config?
[08:36] <lyarick2> Specifically, I have an elasticsearch cluster, instead of selecting nodes at random on the clients, would like to have then connect to head and get redirected
[08:38] <lyarick2> i've done similar with HAProxy — just curious if there are better ways
[08:45] <lyarick2> although I think ES might do it automagically, so I guess I should look that up…
[13:29] <mercutio> haproxy is probably not a terrible way to go
[14:42] <mjp_> redirected based on what criteria?
[14:44] *** nclee has quit IRC (Ping timeout: 256 seconds)
[14:48] *** ncleex2 has joined #arpnetworks
[14:52] <mercutio> he said load, so i suspect based on load average or average time to complete requests or such
[14:52] <mercutio> but i suppose it could just be round-robin or such
[14:52] <mercutio> dns based load balancers aren't very good at balancing normally.
[14:52] <lyarick2> mostly failover
[14:53] <mercutio> relayd on openbsd works easy for failover.
[14:53] <lyarick2> I read up on it, ES does all the internal routing for the request to optimize the queries
[14:53] <mercutio> but haproxy is likely to have higher performance.
[14:54] <lyarick2> higher performance how?
[14:55] <lyarick2> one thing is the cluster/clients are on the same network, so there's some advantage to spreading the network traffic too
[14:55] <mercutio> well relayd isn't really optimised for performance.
[14:55] <mercutio> depending on level of complexity you want there are multiple ways to go for configuration.
[14:55] <mercutio> like the most complicated load balancing way is to just route traffic to the right server who then bypasses the load balancer to send traffic out to the net.
[14:56] <mercutio> but that is also the most complicated
[14:56] <mercutio> i've never had performance issues load balancing at layer 7 though
[14:57] <lyarick2> Yeah this wouldn't be internet facing
[14:58] <mercutio> http://blog.haproxy.com/2011/07/29/layer-4-load-balancing-direct-server-return-mode/
[14:58] <lyarick2> originally we are co-locating with a mapreduce type setup
[14:58] <mercutio> that's the complicated high performance way.
[14:59] <lyarick2> but ES doesn't like that at all, it wants the machines to itself
[14:59] <lyarick2> ok great, thanks for the link
[14:59] <lyarick2> so anyway, connecting to $(/bin/hostname) isn't going to work anymore ;-)
[16:54] <mercutio> it's curious that whatsapp is going free
[19:38] *** up_the_irons has quit IRC (Ping timeout: 250 seconds)
[20:05] *** up_the_irons has joined #arpnetworks
[20:05] *** ChanServ sets mode: +o up_the_irons
[20:09] <up_the_irons> mercutio: wasn't whatsapp always free?  I never had to pay
[20:09] <mercutio> nah it was $1 or something
[20:09] <mercutio> not a lot
[20:10] <mercutio> but it was free for a year, i got a message saying that it was extended.
[20:10] <up_the_irons> i
[20:10] <up_the_irons> i c
[20:11] <mercutio> it maybe because you were early user?
[20:12] <mercutio> https://www.whatsapp.com/faq/en/iphone/30060258
[20:12] <mercutio> they haven't updated their web page it looks like, but it seems that some users had to pay $0.99 to download it and got it for free after, and others paid $0.99/year after first year free.
[20:24] <up_the_irons> ah ok
[21:13] *** lyarick2 has quit IRC (Quit: Leaving.)