[DRBD-user] linstor-proxmox-5.0.0rc1

[Andreas Pflug]

Am 16.03.20 um 12:36 schrieb Roland Kammerer:
> Dear PVE users,
> 
> This is a feature release that will bring 1 very useful feature, but
> also breaks (very legacy) compatibility with older plugins.
> 
> - The feature:
> What this does is to try to create the DRBD resource diskfully if it is
> possible on the node where the VM will be started.
> 
> A common setup might consist of 3 nodes, where all have a given storage pool,
> with a replica count of 2. By just auto-placing the resource, it might happened
> that the storage is placed on nodes A and C, but the VM is actually executed on
> B. This is fine, the plugin creates a diskless DRBD resource on B, and all data
> is read and written via the network. Still, it is obviously preferable to have
> local storage.
> 
> What happens is that the plugin tries to create a diskful resource on the node
> that will run the VM (with the same storage pool as defined in the RG (resource
> group)).
> 
> If this fails: Ignore it, run an auto-place (as we always did). It might find
> another placement.
> If this succeeds: Cool, still run auto-place, we might need to create other
> replicas on other nodes to fulfill the overall replica count.
> 
> As this is a new behavior (i.e., storage is no longer equally distributed, but
> the local node is preferred), this behavior has to be enabled as follows:
> 
> drbd: drbdboth
>    content images,rootdir
>    controller 192.168.122.94
>    resourcegroup onboth
>    preferlocal yes
> 
> You will need the latest stable LINSTOR for that to work!
> 
> - What we remove:
> This new feature is very much coupled to LINSTOR RGs. As this is the preferred
> method to configure DRBD/LINSTOR storage for some time, this commit also
> removes the old legacy options (setting "redundancy" and "storagepool" via
> /etc/pve/storage.cfg). 

I see redundancy still used in the current documentation at
https://www.linbit.com/linstor-setup-proxmox-ve-volumes/ , so it appears
to me as being switched from "documented/supported" to "not supported"
directly, without a deprecation phase.

I wonder what is going to happen when I dist-upgrade the pxe-hosts, with
storage.cfg containing several storages with storagepool and redundancy
definitions.

Regards,
Andreas