Proxmox VE + DRBD HA Storage

[turboloader]

Cancelled, will be back soon

 

[gregg]

Yes. I very badly want to see a working howto on proper HA utilizing DRBD with Proxmox.

I had been working on some ideas and this is one of them I came up with to replace our current firewall and our current web server setup:

This is all speculative since I havent done any of this yet and dont have much HA experience either. So if anything I have mapped out there looks wrong, please tell me. But its the idea I have right now.

 

[tog]

Out of Topic: The only negative thing I've found was the slow network bandwitch. I only get 2 Mbyte's from my VM to my local Gateway. Wether what network card or driver(also the virtio Paralized dirver)
May be it's normal but it's far away from a real network card. So on.

I just did some tests, with KVM on my 8-core Xeon 2.33GHz, the KVM process tops out at 100% CPU (seen on host system with top) doing a little over 5MByte/sec using FreeBSD 6.3 inside KVM using the Intel/e1000/em virtual NIC. It's not exactly high-performance but it's more than adequate for my purposes as a low-traffic Internet server.

Debian inside an OpenVZ container has exactly the same performance as the host system (40-50MByte/sec via http off the server I am testing from), as expected.

 

[turboloader]

Cancelled, will be back soon

 

[turboloader]

Cancelled, for questions PM

 

[dietmar]

Why do you want a shared storage filesystem? Sharing a volume group is all you need.

 

[turboloader]

Cancelled, for questions PM

 

[turboloader]

Cancelled, for questions PM

 

[dietmar]

For snycing I've got a seperate network. Does somebody know where to change the path/network for live migration between the hosts. It's still going the longest way over the client network instead the short and fast way directly between the hosts.

Migration commands looks like:

# qmigrate [--online] [--verbose] destination_address VMID
# vzmigrate [--online] [-v] destination_address CTID

The web interface uses the primary IP of the cluster nodes (IP assigned to eth0 or vmbr0) for destination_address, you cant change that.

 

[turboloader]

Ok,

Thanks

 

[turboloader]

I got it now, I just play a bit with my network config.
Now I have the cluster over my eth0 interface and the public network over eth1.

 

[turboloader]

For questions PM

 

[turboloader]

For questions PM

 

[dietmar]

My setup for now. Now I will look for a automatic failover (heartbeat) if one node dies. But for now the system works fine. I'm running win2003 vm's and linux vm's. Also the live migration is going very fast over my seperate line for drbd. I will post a complete how to after my projekt finished in end of june this year.

great!

The quality of the picture is quite bad - i can't read the text inside.

- Dietmar

 

[turboloader]

Questions please pm

 

[dietmar]

I hope the picture is better now.

Does somebody know, where I can find the configuration files of my kvm vm's ?

/etc/qemu-server/

So if one node dies I can copy these configuration files from a shared store (DRBD device) and just add them to my second node. So I don't have to add my vm's manually and it could be done by a script. And after this I only need to mount the drbd device and copy the vm's into the right folder.

That looks similar to what thomas-krenn is doing with Xen:

http://www.thomas-krenn.com/shopx/g...lerytype=sitex&pic_key=210&table=gallery_pics

 

[turboloader]

PM for questions

 

[dietmar]

So I could run 2 storages in the same time, without unmounting my first one.

It is not possible currently, but I am working on that. But I guess I need a few months to get that working.

 

[turboloader]

PM for questions

 

[turboloader]

pm for questions