Yes, but you can't have IP failover anymore. If the load balancer goes down, your website is offline as well...
[edit] oups - forgot english is language of choice here: [/edit] Did someone get a reply from Hetzner with good news yet? I read all the related Hetzner forums mentioned here (and 2 or 3 linked there), but apart from some "maybe this, maybe that" there is not a single final reply. I mailed to Hetzner myself just today, but it would be good to now from other users. I scan the market for dedicated servers a lot, at least once every 6 months. For the last few years I always ended up with Hetzner winning with the unique combination of features, pricing and powerful hardware. Disabling the flexible IP-Routing is a killer-argument to leave the house - but where to go? From all my scanning I did not find a provider nearly as good. Same service with at least double the pricing or vice-versa. Does someone know an alternative? greets Markus
Have a recommendation for everyone. Don't know if you have ever checked out http://www.gplhost.com. I worked with these guys over the last few months and are extremely helpful and supportive. Their prices are also resonable and have multitude of solutions from dedicated servers to xen machines. They prefer to work with the debian distro, but also work with CENTOS, UBUNTU, Gentoo and a few others.
I phoned with Arno Pirner @Hetzner since there was no reply to my question via mail. The answer was simple and clear: custom IP-Routing (shared IP) will be available in the next Datacenter which is planned to be ready April 2009. Most possibly it will then be available only with an additional (paid) option. I'm in contact right now with ovh.de, the upper-class servers are definitly able to share IPs (they have 16 or more ips each which call all be shared). Even the smallest "istgenug" (eng: is enough) servers for 25 Euros/month have a failover-IP assigned. It can be shared, managed by their web interface. Until now, they did not give a definitive answer if it can be shared via script (heartbeat etc.), Ill test it as soon as I can get two servers from there. Their cheap servers sometimes have issues, which would be no problem in a small failover-cluster of 2 isgenug (available with raid 1 also for 36 Euros). The bigger systems seem to be have better overall stability for a still decent price. greets Markus
I was thinking that ovh.de gives you a web interface where you can assign the virtual IP to a server manually, which means it's not automatic and therefore useless for HA. So this is not true for their bigger servers?
The interface is just used for the special "failover-IP", which even the smallest server has - the bigger ones have 2 or 4 failover-IPs. I found some hints, but no clear answer, that the swich can also be done automatically. The bigger servers not only have the special failover-IPs, but also an additional 16+ IP-Range assigned, which can be used for clustering. ovh itself does not guarantee anything with clustering, because they had ready-to-use clusters themselves and want to bring that service back soon. Only option I see so far is testing - I just ordered 2 isgenug. If you're interested in testing for yourself - we could share the month runtime, I just need the servers for one weekend to fully test all functions I need. greets Markus
First result: IP-switching is not possible. I have an open ticket on how to remotely control the switch port / routing, no answer yet. Second problem was: ovh uses selfmade kernels without module support. There's no "quick install" because the kernel has to be recompiled for a lot of funktionality apart from module support (which is needed for drbd, nfs and other stuff). Positive: the Webinterface is quite nice, you can even partition your drives via webgui for reinstalls. greets Markus
Don't they offer distribution images (Debian, Ubuntu, SUSE, etc.) with the distributions' default kernels?
Nope, just their own kernels (with added support for their special hardware combinations, and missing support for modules). They have a broad palette of Linux distributions, even 2 Windows Versions and some ready-to-use virtualization packages (xen, vmware, proxmox), but nothing with the standard kernels. regards Markus
Does anybody have any experiences with Amazon EC2 ( http://www.amazon.com/EC2-AWS-Servi...node=201590011&no=689343011&me=A36L942TSJ2AJA ) and EBS ( http://www.amazon.com/b/ref=sc_fe_c...node=689343011&no=201590011&me=A36L942TSJ2AJA )? This could be a replacement for our HA setups, but I haven't used it so far, and the pricing is quite confusing. If you have lots of traffic, you probably pay a few hundred dollars per month...
Yes it does work, sometimes a technician has to help, which happens fast and automatically. And if you have special hardware in your box (some have a new Intel Gigabit Ethernet that's not yet in the kernel), there are plenty of patches around the forums. After really intensly testing OVH, I recommend using it, as I myself will do with my first real servers beginning early september.
Any update on this? I know from Feb 2011 added the possibility from Robot to Request separate MAC address for an additional IP.
Thank you for the very good tutorial! Unfortunatelly I read your warning about Hetzner Online to late and now I have that setup installed in their data center RZ21. Did somebody get around this problem? They now seem to have a failover IP you can order for 5 Euro per month and traffic will be accounted for the whole cluster with that one server you have ordered the IP. There is aparently a sript which is supposed to handle the automatic switching. Unfortunatelly I can not manage to get it running and hope to find some help here. Thank you for any input and help on this.
Is there a reason why you did not implement the script they provided? According to Hetztner support this script would reside on the second server and monitor the first server. In case of failure they claim it would take 20s to switch automatically.
I haven't tested their script. We used automatic IP switching for several years in the past and all solutions that we tested caused more outages in the end than we would have had without them, so we use now a hot standby solution but we don't let it switch automatically.
I understand. But how do you handle internal shared IPs? I have a 3 node MySQL Galera cluster running which I configured like explained in the tutorial over keepalived HAProxy servers on 10.0.0.10. If Heztner does not allow shared ips, I would need to go over external Failover IPs or introduce a single point of failure if adressing one HAProxy. Did you solve that question already?
We don't use an automatic IP failover at the moment, the only option at hetzner would be an external failover IP in my opinion.
