HA with two simple laptops

[den]

Hey Guys

I want to create a HA environment using two physical laptops.
(Purpose is for a simple app that i want to test)

• I have two physical laptops with the latest pve 4.x installed.
• I have two nodes as a cluster in my datacenter.
• I was hoping to create a simple directory under in /var/lib/vz/shared_dir
• I was hoping anything on this "shared_dir" folder will be synced between the two nodes.
• And if syncing is setup, then install the container's disk on this "shared_dir"
• so if one node is offline, then the second node will continue to run the container..

please help.

thanks.

 

[LnxBil]

You can achieve this by using DRBD and enable HA for the container. After a crash of one node, the container will get started on the other.

Please refer to http://pve.proxmox.com/wiki/DRBD

 

[udo]

You can achieve this by using DRBD and enable HA for the container. After a crash of one node, the container will get started on the other.

Please refer to http://pve.proxmox.com/wiki/DRBD

Hi,
but not with two nodes - you need three because quorum.

Udo

 

[LnxBil]

Hi udo,

You're right. Officially you need 3 nodes for PVE 4.x. In general, you only need STONITH when using DRBD.

Best,
LnxBil

 

[den]

does this only replicate an entire volume / drive?

what if I only wanted to sync a particular folder only? ie. /var/lib/vz/shared_dir ?

I was also looking at a win7 box that is visualized, running an application with a MS SQL database. Will DRBD be a suitable use for such a scenario?

Thanks..

 

[LnxBil]

Hi den,

You cannot share a directory as a filesystem over network without drawbacks (e.g. a third party storage server sharing the disk or asynchronous replication), you have to share a volume at block level. DRBD does exactly that. To have a "real" cluster with HA, you need a replicated storage backend with synchronous write, such that if one node fails, the other will be crash-consistent as the VM would be. In case of a database, you need that the redo logs will be on disk on ALL disks.

I strongly do NOT recommend this:
If you do not need a consistent replica, you can use PVE-zsync (https://pve.proxmox.com/wiki/PVE-zsync), glusterfs or else.

LnxBil

 

[den]

Thanks LnxBil for the recommendations. Very helpful.

Just wondering, when you create a cluster, the folder /etc/pve is synced between clusters (in real time?). What app/function does this?

Thanks..

I strongly do NOT recommend this:
If you do not need a consistent replica, you can use PVE-zsync (https://pve.proxmox.com/wiki/PVE-zsync), glusterfs or else.

LnxBil

 

[LnxBil]

Yes, they have something they call Proxmox Cluster Filesystem, which is described here:

https://pve.proxmox.com/wiki/Proxmox_Cluster_file_system_(pmxcfs)

 

[den]

LnxBil -

Thanks for the info. I was looking at PVE-zsync. So if I wanted to use PVE-zsync to backup a VM to the second node2, then if node1 ever goes down, I need to restore the VM on node1 and get it back up running correct?

And I assume PVE-zsync syncs only in one direction?

I guess its good to keep backups in sync between cluster nodes...

Hi den,

You cannot share a directory as a filesystem over network without drawbacks (e.g. a third party storage server sharing the disk or asynchronous replication), you have to share a volume at block level. DRBD does exactly that. To have a "real" cluster with HA, you need a replicated storage backend with synchronous write, such that if one node fails, the other will be crash-consistent as the VM would be. In case of a database, you need that the redo logs will be on disk on ALL disks.

I strongly do NOT recommend this:
If you do not need a consistent replica, you can use PVE-zsync (https://pve.proxmox.com/wiki/PVE-zsync), glusterfs or else.

LnxBil

 

[LnxBil]

ZFS snaphots in general are syncable in any direction if they do not diverge. It is not possible to sync if both are changed, you need to throw away one side.