proxmox ha cluster: what if shared storage goes down

[silvered.dragon]

Hi to all,
I'm new with proxmox but I want to learn the much as possible cause I think is a great project, question is:
- supposing that I built a 2 or 3 nodes proxmox Ha cluster based on a storage device, such a server that offers iSCSI or NFS. If this server have an hardware failure (let's say something that need complete hardware replacement) what happens to my cluster? of course it will go down for a while.. so this is not something that I can call High availability if the based storage is not rendundant.. how can I manage this to feel more protected against hardware failure?
many thanks and sorry if this is a pointless question

 

[adamb]

Most central storage enclosures are built around redundancy. Even the entry level enterprise stuff has dual raid cards, dual path's to each card, dual power supplies. It greatly minimizes the chance of failure.

If you are still concerned then you would need to look at some type of replicated storage, most vendors have replication solutions.

You could also look to use DRBD. Lots of options.

 

[Q-wulf]

You could also look to use DRBD. Lots of options.

like e.g. a ceph cluster hosted right on your Proxmox Ha Cluster.

 

[silvered.dragon]

like e.g. a ceph cluster hosted right on your Proxmox Ha Cluster.

ooo ok thanks for your suggestion, I tried to googling about ceph cluster and drbd.. I found that is very interesting, so you think that a 2 node HA proxmox cluster based on a ceph cluster hosted on same 2 nodes will be a good solution? may you can suggest to me some documentation..
many many thanks.

 

[silvered.dragon]

found this https://pve.proxmox.com/wiki/Ceph_Server but seems to work only with 3 nodes.. is there the possibility to run on 2 nodes only?

 

[adamb]

found this https://pve.proxmox.com/wiki/Ceph_Server but seems to work only with 3 nodes.. is there the possibility to run on 2 nodes only?

To achieve quorum you need atleast 3 nodes for HA. If you aren't looking for HA you can use only two nodes but failover will require human intervention.

 

[Q-wulf]

To achieve quorum you need atleast 3 nodes for HA. If you aren't looking for HA you can use only two nodes but failover will require human intervention.

actually you can use a single node for ceph. All you gotta do is make sure that your crush rules do not use the leaf "host" but instead use the leaf "osd".

But it introduces single point of failure as you mentioned.

there is a ceph thread that might be interesting to the OP :
http://forum.proxmox.com/threads/24176-Newbie-need-your-input

Question is always what type of Hardware your using (ie. are the disks suitable for ceph, whats the node interconnectivity like, is CPU/Ram enough to run your vm's + Ceph)