PVE 4.1, DRBD9 - multiple vgs/lvs support

[Firm]

Hi,

Using https://forum.proxmox.com/threads/drbd9-how-to-configure-multiple-volumes.26010 I was able to create lv with both ssd and sata disks combined. But still unhappy because pvmove have to be manually executed every time I added vm/ct to assing it to ssd or sata.

Is there any chance to support multiple volume groups / logical volumes in PVE? So I would:
* create vg-ssd, vg-sata volume groups on appropriate disks (instead of drbdpool)
* place .drbdctrl_0/.drbdctrl_1 volumes on selected volume group
* create several logical volumes on selected volume groups (instead of drbdthinpool)
* create CTs/VMs on selected logical volumes (attached via /etc/pve/storage.cfg)

Regards,

 

[fabian]

multiple volume groups are already supported (you can define more than one lvm storage, just like you can define more than one dir, zfs, .. storage) - but you have to create the VGs yourself integrating more storage features into the GUI / pvesm is on our long-term TODO list

 

[Firm]

Yes, adding more LVM storages is already possible. But I thought about adding 2 (or more) DRBD volumes with different replication policies. Right now only one VG/LV is possible: drbdpool/drbdthinpool (by default, one may use other types of DRBD storage plugins). With 2 (or more) DRBD VG/LV groups it's very easy to implement different HA strategies, e.g.:

drbdpool-ssd - for mission/speed critical VMs, using SSD storage and replication factor 3
drbdpool-sata - for non-vital "slow" VMs, using SATA storage and replication factor 2

I wonder if it will be possible to manage all clusters from one UI (implement cluster of clusters strategy).

 

[e100]

I've just started using DRBD9 and Proxmox 4.1 this week.

AFAIK drbdmanage does not support multiple storage pools.
Each node can have a single storage plugin that points to a single pool and currently there are only three storage plugins:
http://drbd.linbit.com/en/users-guide-9.0/s-drbdmanage-storage-plugins.html

Proxmox team can correct me if I am wrong but I think we would need to:
1. Work with drbdmanage developers to get new multi-pool support added
2. Work with Proxmox team to support the new drbdmanage feature.

drbdmanage does support specifying what nodes should be used when creating a new resources.
I would really like to see Proxmox support that.
You could then have DRBD SSD nodes and DRBD SATA nodes and when you create a disk you could select the nodes (or group of nodes) you want to use.
Today, similar to pvmove, you can use drbdmanage to move the backing storage for a resource to another node after its created by proxmox.
So at least there are some work-arounds for both limitations.

Does Proxmox plan to support snapshots with DRBD?

 

[Firm]

I wonder what is "redundancy X" use scenario (for drbd: block in /etc/pve/storage.cfg). If I have 3 nodes and set redundancy value to 2, how will nodes be chosen to kept VM/CT volume? What will happen in case of 5 nodes and redundancy value set to 3?

 

[e100]

I believe that Proxmox simply passes the redundancy value to drbdmanage. So what nodes get used is up to drbdmanage and how its configured.
http://drbd.linbit.com/en/doc/users-guide-90/s-dm-new-volume

Default is to use nodes with the most free space

 

[Firm]

Thank you, I got the idea.