12 Mar
2020
12 Mar
'20
6:09 p.m.
If the ceph roadmap is getting rid of named clusters, how will multiple clusters be
supported? How (for instance) would `/var/lib/ceph/mon/{name}` directories be resolved?
On Mar 11, 2020, at 8:29 PM, Brian Topping
<brian.topping(a)gmail.com> wrote:
On Mar 11, 2020, at 7:59 PM, Anthony D'Atri
<anthony.datri(a)gmail.com> wrote:
It is a converged architecture in that all three boxes are running Kubernetes. There is
one k8s cluster on each side of the link, let’s call them “primary” and “secondary”:
* The primary k8s cluster will only access storage from the primary Ceph cluster,
secondary k8s only accesses storage from secondary Ceph.
* Primary Ceph gets monitors on both sides of the link. Secondary Ceph only has monitors
on the secondary side.
In a netsplit situation, the primary Ceph will maintain quorum with both nodes on the
primary side. The secondary Ceph cluster only exists separately for this netsplit
situation and the secondary k8s cluster can continue unaffected.
With this in place, the primary side can continue operating with either primary node
downed for maintenance via a suboptimal quorum over the WAN link. I cannot do that today.
I am sacrificing the case where there is a netsplit at the same time I am doing
maintenance.
Thanks for your input!
Brian
This is all possible with a single cluster, but
this limited node also needs storage.
Are you saying that the limited node needs to access Ceph-based storage? Is this some
sort of converged architecture?