Xen project Mailing List

End of the day my number one goal has just been to find the best way to have a stable HA setup for database server VMs.

Here's all the ways I've looked into, and my notes

Please note, it's all just my own opinion here...

XCP HA

Does not support native HA from version 1.6 (I seem to recall it was working back in 1.1 or so, I think? - might've been disabled at some point by Citrix, which would be perfectly fair really since you should pay for the feature)
Does support pooling, but you need "identical" servers

Some support for differing CPU types, but still needs to be close
http://support.citrix.com/article/CTX127059

VMs using local disk storage require a home server (where it starts from)
Pools support live migration even with local storage
Can move running VM from one machine to another fairly seamlessly
Note, again, as above, does not natively support HA

XenServer HA

$1500/server license for Advanced required for HA
Pretty much everything is the same as for XCP but also supports HA
You need shared storage for this (can be DRBD or an expensive SAN type device)
Also requires a shared heartbeat SR for HA (e.g. on a NAS or something)
If you can afford it, then certainly I do recommend this option, Citrix offers a generous trial period as well to give it a shot

XenServer with DRBD and Pacemaker

2-node HA setup
It's possible to use DRBD with Primary/Primary to support live migration, but I don't trust it since I've had too many finicky issues, there's more split brain problems, etc, so I use Primary/Secondary
Linbit provides some docs about DRBD + XenServer if you google it
HA in this case means that Pacemaker automatically moves DRBD to Primary on the secondary node in case of failure, attaches the SR, and restarts the VMs.. this does not save the running machine state / memory state, and takes, say, maybe 45 seconds for the switch-over
This is what I actually have running in some production environments, and to me is the easy "that'll work and be reliable" answer
I wrote some Pacemaker plugins to help manage the VMs and the PBD/SR for automatic fail-over: https://github.com/locatrix/xs-pacemaker
The admin for this type of setup should be fairly DRBD and XenServer savvy in case of problems, but overall it does work, and DRBD is a rock

Xen with Remus

Works by using Xen's existing live migration system, so you have to do everything to get live migration working first, and then you can use Remus to automate fail-over
Has a number of guest and performance limitations; I wrote up some more information here on the updated page: http://wiki.xen.org/wiki/Remus
Once it's running it's really cool since it does literally allow a VM to fail-over to a secondary machine with practically no interruption, migrates memory, network connections, etc
However, to be honest, I've found from a practical perspective, that Remus isn't production ready, and is more interesting as an academic project
I had too many weird problems, crashes, guest incompatibilities, and theoretical performance problems, etc, to count on it
e.g. immediate killer for me is that Ubuntu guests basically don't work without some serious kernel hacking

CloudStack

Citrix CloudStack (now Apache CloudStack) is a mature java based solution for cloud management. It supports KVM, XenServer, and VMWare hosts in a unified cloud platform.
It's free
CS has a really nice web UI interface that is impressive at a first glance.
Respects HA built-in to XenServer, but basically doesn't add anything new there, so for simple HA purposes you don't really gain much by using CloudStack, just additional complexity with a nice UI
In general it's very much oriented towards bigger installations, and there are those who are successfully using it for large scale projects
My opinion for our smaller setup (we have 1 rack in a colo) is that it's overkill and I've had a lot of trouble getting it to work in a basic lab environment... I've often had the experience I've heard from many others about "getting stuck on a spinner and having to dig down through a bunch of obscure log files with java exception errors to try to figure out what went wrong"

OpenStack

TBD; I've haven't done much with it yet; it's on my list
Also a big installation oriented project, and obviously interesting in terms of the volume of industry support, etc

Proxmox

Not Xen based, but a really cool and interesting project for smaller installations, worth mentioning to check it out
I haven't done quite enough experimenting with it yet, but it does have some native clustering and HA support, although it requires a DRBD primary/primary setup to work with it's built-in HA

Google Ganeti, which uses KVM and Xen

Next up on my list of things to play with when I get a chance

Cheers,

Andrew

On Tue, Mar 12, 2013 at 10:29 AM, R. Jeremy <sgrunt91@xxxxxxxxxxx> wrote:

Hi,

I'm new in Xensource World. I have to work on Xen 3.1.2 hypervisor, I've done some searches on how to make Xen HA (not FT for the moment).
The way that is come again and again is to make a cluster with Corosync, pacemaker, DRDB. I just want to now if there are others way to do Xen HA, and which is the best way in your opinion?

I need something very stable and sustainable.

I've seen many others mails of this subject, but we never know what is the best way. And this is often for Xen 4.

I'm french, so I hope you'll understand my english...

Thanks!

BR

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users

Re: [Xen-users] Xen 3 High Availability ways