Concept with two servers and shared storage

[henning]

Hi all,

I haven't used PVE so far but it looks like the perfekt solution for me.

I have two servers and a storage with 5 TB and fiber channel connection. I want to run 3 KVM machines on one of the servers. The second server should be the backup for the primary server in case there is a hardware failure or it needs maintenance. Two of the VMs are small web servers with little load and only a small amount of data (~5 GB each). The thired VM is a mail server with 3 TB of data.


Here are the questions I have:

1. Does it make sense to configure the servers as cluster? If the master goes down, is it possible to start the machines to the second server?

2. How do I configure the storage? According to this http://pve.proxmox.com/wiki/Storage_Model LVM seems to be the best way to use the same storage on several servers. What I don't understand: Is it possible to access the same Volume on different servers at the same time? Do I need a cluster fs for this?


Thank you in advance
Henning

 

[bohansen]

1. Yes, provided you use shared storage you only need to make sure you rsync configuration files to a backup directory on the other server. So a little manual work is needed in case of hw breakdown on the primary node. However normal maintenance you should be able to live migrate most machines to the other host.

2. If you want shared storage with only two servers you should look into DRBD (http://pve.proxmox.com/wiki/DRBD). It's fairly easy to get started, but I found some time needed to play with the details, split brain recovery, performance and so on before putting it into production. It can be a cheap "synchronous SAN" solution if you need it.

 

[henning]

bohansen, thank you for answering my first question!

I think using local storage with DRBD instead of a central SAN storage is not a good idea for me. I would like to be able to scale. Maybe I will need more servers very quickly.

So question number two still exists .

 

[udo]

bohansen, thank you for answering my first question!

I think using local storage with DRBD instead of a central SAN storage is not a good idea for me. I would like to be able to scale. Maybe I will need more servers very quickly.

So question number two still exists .

Hi,
i have a similiar setup which work very stable.
Both nodes see the same SAN-Disks (with FC normaly you will get an good performance). On the SAN disks (or one big) you have lvm - the volume group is on both nodes active, but the logical volumes (the disks of the VMs) are only active( used) on the node which are run the VM. So you don't need an clusterfilesystem only lvm. The logical volumes are used as raw-devices for the VM.
Simple and fast...

Udo

 

[henning]

Udo, thanks a lot. Your answer is very helpful for me. Without a cluster filesystem my scenario gets much less complex. Good for me!

One last question got in my mind concerning the mail server. I wonder if I should create a VM with > 3 TB of storage, or if I should create a small VM with only the programms in it and mount storage for mail data inside the VM, if possible.

Any suggestion on that?

 

[udo]

Udo, thanks a lot. Your answer is very helpful for me. Without a cluster filesystem my scenario gets much less complex. Good for me!

One last question got in my mind concerning the mail server. I wonder if I should create a VM with > 3 TB of storage, or if I should create a small VM with only the programms in it and mount storage for mail data inside the VM, if possible.

Any suggestion on that?

Hi Henning,
I would recommend following for a mailserver: a small disk for the system and a big one for the mail-storage. If you have different kind of storage you can also split this: one vm-disk on fast sas-raid for mail-spooling (and database? or mailboxes?), one vm-disk on sata-raid for mail-archive... depends on your usage of the mailserver and which kind of storage do you have inside your SAN.

With this configuration you can backup the system-disk also with proxmox (the big disk with "backup=none") and the whole mailserver with an inside-backup solution (like bacula).
This has the improvement, that you don't need a baremetal-recovery for the mailhost - simply use the proxmox-backup for the system and then recover inside the VM.

NFS-Storage for an mailserver is perhaps possible (dovecot should work with nfs-storage), but due to performance reasons i prefer direct storage (in sight of the VM).

Udo

 

[henning]

Let's see if I understand correctly:
On the host system I mount, let's say, 20 GB of memory from the SAN, activate LVM and install my VMs (OS + Software) there. Then I go to the Mail-Server-VM and mount 3 TB of memory from the SAN on i. e. /mnt/mailstore. There I put my mail-data. Correct?

 

[udo]

Let's see if I understand correctly:
On the host system I mount, let's say, 20 GB of memory from the SAN, activate LVM and install my VMs (OS + Software) there. Then I go to the Mail-Server-VM and mount 3 TB of memory from the SAN on i. e. /mnt/mailstore. There I put my mail-data. Correct?

Hi,
no not realy.

I mean following: You define a Partition/Raid-Volume/whatever on your SAN, which you provide to both Proxmox-Nodes (e.g. 5 TB). On the hosts you see this as disk (like sdb).
Then you create an volume group e.g.:

pvcreate /dev/sdb
vgcreate -s 4M san_sata /dev/sdb

Inside the pve-gui you define a new shared storage (lvm) which use the volume-group san_sata.
If you have different sorage you can also use them, like "san_sas".

When you now create an vm, you select as storage san_sata and use e.g. 20GB (system).
After creation, you add an second disk (also on storage san_sata) of 3 TB.

Inside the guest, you see the disk as two harddisk - sda and sdb, or if you use virtio vda and vdb.

Udo

 

[henning]

Ah, I see!

I think now I have every information needed to get my hands on the system. I will set it up just as discussed here.

One last thing to say: I'm very impressed of the PVE project. I want to thank the developers for the great software and the community their nice help.

I believe PVE will be a greate tool for me which makes my day much easier.

Rgs,
Henning