[DRBD-user] Hardware specs & upgrades

[Roland Kammerer]

On Thu, Aug 08, 2019 at 10:41:37PM +0300, Vito Botta wrote:
> Hi,
> 
> at the moment I am testing with 3 nodes that have 4 cores and 16 GB of
> ram, and all works great. However since DRBD/Linstor have very little
> overhead compared to other storage products I've tried with
> Kubernetes, I was wondering whether 2 cores and 8 GB of ram would be
> sufficient? 

LINSTOR/DRBD are not that resource hungry. Java is a but, but Rene got
some pretty nice results with some minimal configurations. Rene?

> Also, if I have three servers with volumes being replicated, how can I
> avoid downtime when upgrading whenever a new release is out?

That depends a bit what you want to update:

The really nice thing about DRBD and LINSTOR are that the control plane
is separated from the data plane. Which means you shut down LINSTOR
completely and your data is still up, because DRBD continues to
replicate data. So updating LINSTOR is easy, just do it :)

If you need to update DRBD (i.e., the kernel modules), then you
basically do this:
- move all resources away from the node you want to update (i.e, no DRBD
  resource is Primary anymore)
- drbdadm down all
- rmmod
- upgrade

And this then you do the same for the next node. We are very careful
that you can have mixed DRBD version until you finally upgraded your
last storage node. Obviously you need to have at least 2 storage nodes,
if you only have one diskful and one diskless node, then yeah, you will
have downtime. And maybe you want to first provide "over redundancy" via
an additional diskful replica, or you are fine that your N times
redundant resource is only N-1 during you upgrade your storage. But that
all depends on your use case.

The important thing is that data and control plane.

Best, rck