Gorka Eguileor geguileo at redhat.com
Tue May 25 16:23:39 UTC 2021

On 21/05, Franck VEDEL wrote:
> Hello.
> First, sorry…poor english… so it’s a google translation. I hope you could understand my problem.
> Following the installation (with kolla ansiible) of an openstack (wallaby) on a physical server (under centos8), we were able to see all the possibilities offered in our teaching by Openstack. (i’m working in a french university).
> For various reasons, we need to extend this manipulation. We will have 3 nodes (dell R740) and a Dell bay for storage.(dell compellent)
> After having mounted a first test with 3 servers (T340), and put the storage (LVM) on node 3 (without the bay therefore) to test whether we had understood certain things correctly (in particular the parameter of the « multinode" file and that of « globals.xml"), we want to test with the Compellent bay.


I have never used kolla ansible, so I can only offer general Cinder

> My question is as follows: knowing that the 3 nodes are three identical servers, in the multinode file, how to configure [storage]… should one of the 3 servers be put, add the iscsid docker… or else put the IP of the bay . I admit that this aspect is problematic for me.

If your storage is iSCSI, then you will need iscsid on ALL your nodes.
Compute nodes need it to attach volumes to instances, and controller
nodes where Cinder is running will need it for create volume from image
and some migration/retype operations.

> what about « enable_iscsii » and « enable_backends_lvm » parameters ?

It's important to realize that your 2 scenarios (LVM & Dell) are quite
different.  LVM storage is local to the host where cinder is working (no
HA option is possible), but in the Dell case storage is external, so you
can do HA.

According to the docs [1] for the Dell storage you'll need to use
"enable_cinder_backend_iscsi", but for LVM you should use
"enable_cinder_backend_lvm" instead.

> In a configuration like mine, would you put a "controller" or 2 or 3?

It depends on what you want to do, if those 3 nodes are only going to be
used for the controllers, then I would recommend Cinder to be deployed
on the 3 nodes for the Dell case, and only on 1 for the LVM case.

If the 3 nodes are also going to be used for computing, then you have to
decide what is best, having more resources available for your VMs (only
1 cinder node) or have better resiliency (> 1 cinder node).

> Should we instead put an LVM on one of the servers and with iscsi make this LVM point the bay?

That's more limiting, so I would only go that way if the storage bay was
not supported by Cinder.

> Maybe these are silly questions, I'm not sure. This use of the bay is new and I don't know how best to do it. Between cinder, lvm, iscsi, the bay, the multinode file and the options of globals.xml, it is still not easy at first
> Franck VEDEL

I don't know if your storage array has multiple interfaces, but if it
does you'll want to enable multipathing, with "enable_multipathd".


[1]: https://github.com/openstack/kolla-ansible/blob/stable/wallaby/doc/source/reference/storage/cinder-guide.rst#cinder-backend-with-external-iscsi-storage

More information about the openstack-discuss mailing list