a way to share internal disks for shared storage?

[manilx]

Hi

I have 2 protectli FW6E with 1TB internal msata configured as 2 nodes. All working well.
I have configured a nfs share on my qnap nas to serve as shared storage but after a short while the nodes lost connection to it and the vm's error-ed out. I was able to copy the disks back to local storage.
Somehow the nfs share is not stable enough (I guess sms won't be better). And the Units only have 1gig ethernets anyway (which would be enough for my usage and mirroring the vm's was ok as only the memory is copied anyway). BUT the unstable connection when serving as an external storage is troublesome.

So I ask: is there a way to share the internal storage of each one as external storage? Probably not.
I did not yet understand the ceph thing....

Thx in advance.

 

[apoc]

AFAIK you need to have multiple disks to use ceph and the usage with 2 nodes is questionable. You need a tie breaker (clust needs to have quorum to run).

So in the end the question is: why shared storage?
HA? You at moving your spof onto the storage potentially. Or to the network.
Because it is cool? What is your rationale behind? Doing training, gaining experience?
DR? Maybe ZFS Send/Receive is an option for you.

My experience is: the less complex the better it is for availability, at least for home usage. Running my single proxmox server since years, never had an outage due to failing server itself. Disk? Yes! Multiple times? Network as well, but the things I was once most concerned (Main board, Memory,) is keeping going. I guess my spare parts are never going to be used...

 

[manilx]

AFAIK you need to have multiple disks to use ceph and the usage with 2 nodes is questionable. You need a tie breaker (clust needs to have quorum to run).

So in the end the question is: why shared storage?
HA? You at moving your spof onto the storage potentially. Or to the network.
Because it is cool? What is your rationale behind? Doing training, gaining experience?
DR? Maybe ZFS Send/Receive is an option for you.

My experience is: the less complex the better it is for availability, at least for home usage. Running my single proxmox server since years, never had an outage due to failing server itself. Disk? Yes! Multiple times? Network as well, but the things I was once most concerned (Main board, Memory,) is keeping going. I guess my spare parts are never going to be used...

Having been running one server for almost a year without issues (I didn't cause and a 15min reinstall of OS and then restore from backups was easy) BUT I've learned to better be safe than sorry as I'm running a few critical VM's (i.e OPNsense) and a hardware break (for whatever reason) would be an issue (as getting spares today can be a problem!).
Anyway got a 2nd mirror hardware and set up the cluster.
Migrating VM's live is fine BUT takes time without shared storage (that's the reason for wanting it). Just looking ahead, no urgent issue at the moment.

 

[apoc]

As said. Shared storage except Ceph or similar concepts introduce just another problem in terms of a Spof. And this is often overseen. Especially at home or semi-professional environments. Hence I am mentioning it.

If you are only pushing your problem from one level to another including the additional complexity involved you are not gaining anything.

Ask yourself: what exactly do you want to achieve? I'd look into each of your VMs/workloads and ask the following questions:
- what is the delta of data I can accept? (Data loss)
- how fast do I need to recover?
For your internet connection you might end with a Opncluster. So it can take over on VM level. For a different VM you might do a ZFS send/receive, rsync whatsoever.

I truely believe that a centralised storage approach won't do you any good. Take a different perspective.

Hope that makes sense

 

[manilx]

I understand what you mean and you certainly have a point!