Advice on storage and HA

[Oldsocks]

Hello!

I've been asked to set up a proxmox environment for testing stuff. I'd like to go HA. I've got a very basic machine (single cpu, some RAM, single drive) to use as a third for quorum. The other two machines have a bit more - 1 drive for the OS, two enterprise drives each for data, meatier compute.

For Ha I know I need shared storage and I've been using networked storage for that until now. However, the test environment needs to be separate from the network storage. Is it possible - and I think it is, but could use guidance - to configure 2 drives in each server so that they're a single shared pool? That'd be 4 x 4Tb drives, RAID 1 in each chassis (the controller is rubbish so I'm happy to not use it if that's the answer), for 8Tb total. Is this a gluster? ceph? zfs? solution?

I'm coming from the other side of the fence so I'm having to learn an awful lot in an awful hurry; sorry for asking what feels very much like a dumb question with an obvious answer. I have been looking at buying a cookbook but the one I saw listed is a few years old - would it still be relevant to the current proxmox release?

Many thanks for helping an old dog trying to do some new tricks!

 

[wolfgang]

Hi,

for testing stuff. I'd like to go HA.

For Testing stuff you need HA? HA sould only used if really needed. Because with HA you have to make every part redundant.

As you descripe your setup I would go with GlusterFS, because for Ceph you have not the resources. ZFS will not work because the Raid controller.

I have been looking at buying a cookbook but the one I saw listed is a few years old
- would it still be relevant to the current proxmox release?

This is the new updated version from Nov 2017.
https://www.packtpub.com/big-data-and-business-intelligence/mastering-proxmox-third-edition

 

[totalimpact]

gluster or drbd+lvm...

rubish raid card will give rubish results, test with low impact vms like linux or win7... a single ssd/nvme, no raid may do you better. you will learn fencing, barbed wired, etc

ZFS is a local file system, it does not have "live clustering" abilities, but if you had a pair of "better" machines it could be made in to a poor mans HA using pve-zsync, you will not have live synchronization, nor automated recovery, but you could have up to the last 10-15 minutes, and manual restore. I would try the other options first if I were you. Proper ZFS will start with 64gb ram, and lots of drives on a non-raid card (ie LSI HBA).

 

[Oldsocks]

Thanks for the pointers!

HA is one of the things to test. Not a production system so performance isn't a super high priority. Both boxes (aside from the one intended for quorum) have 96GB of RAM and 15k drives - not SSD, but they'll do. The drive controller is whatever weak rubbish dell were inflicting at the time and there's no 10Gb nics any more - I was under no illusions regarding performance. But, if it all goes well there's a lot of shiny new hardware which will have SSDs and 10Gb and a decent controller.

 

[totalimpact]

It most likely has a h 200 or h310 RAID controller. These are really bad because they have a queue depth of 25. If you have the full card not the mini mono, it can be flashed to an LSI 9211 firmware which has a queue depth of 600. As raid cards these things are absolutely horrible. But if it is a full-length card reflashing them to pure HBA mode or it mode can yield some good performance on ZFS. I am not sure about gluster I have not tested it.