[linux-lvm] [PATCH 10/10] man: document --node option to lvchange

[Vladislav Bogdanov]

19.03.2013 18:32, David Teigland wrote:
> On Tue, Mar 19, 2013 at 01:32:50PM +0000, Vladislav Bogdanov wrote:
>> +.B \-\-node \fINodeID
>> +Perform specified activation command on a remote cluster node (the same to running
>> +corresponding command on that node locally).
>> +Currently tested only for corosync clusters (\fB-I\fP \fIcorosync\fP option to clvmd)
>> +for corosync versions from 2.0.
>> +If corosync configuration has node names in a nodelist (nodelist.node.X.name = name
>> +in CMAP tems) or node names are used for ring0_addr (nodelist.node.X.ring0_addr = name),
>> +then that names may be used as NodeID. Otherwise numeric node IDs should be used.
> 
> lvm tools should move away from doing remote command execution.

For me that is just convenient - why not?.

> As you mentioned before, this is the equivalent of "ssh node lvchange".
> ssh or some other tool outside lvm is the right way to run the remote
> commands.

I think about porting this to LVM API as well, and use it in libvirt.

> Also, lvm should not assume that it's using dlm/corosync, or that this
> kind of remote option will be possible to support.

That all was already "almost" supported by clvmd. I only pass node id
from command line to an existing clvm header (and fix some related
functions).

All cluster/dlm staff is in corosync clvmd module - so both clvmd and
tools are cluster-stack-agnostic. May be the same is possible for both
cman and openais modules - I do not know because I do not use them and
do not plan to. You know they both are almost dead, and only
corosync/dlm actually remains.

And... Every cluster has nodes. Clustered LVM VG assumes you have
cluster. Why not be cluster-aware in tools then?