[Cluster-devel] "Remote" Corosync

[Cluster-devel] "Remote" Corosync

[Christine Caulfield]

Hi all,

This isn't my idea, it's Steven Dake's, but I've chosen to "run with it" 
as the saying goes.

The idea is to provide a TCP/IP implementation of the corosync IPC 
system. This will allow remote clients to connect to corosync and do all 
the things that a local user can do - though we might decide to restrict 
this in some way.

This has a number of uses that I can see, and probably quite a few that 
I can't. One is use of corosync as a remote lock manager for nodes that 
are not part of a cluster but need to share some resources, eg clvmd. 
Another is remote management using the CFG interface. Or perhaps remote 
monitoring of services for stretch clusters.

Whatever it gets used for its important to realise that such clients are 
NOT part of the cluster. They are simply being allowed access to some 
selected cluster resources. As such they will not be subject to fencing 
and this will restrict the sort of resources that they should be allowed 
to access. This restriction should be in the client code rather than as 
a policy in corosync itself though.

One issue that needs to be addressed carefully is that of security. 
Remote access must be carefully secured otherwise it simply become a 
hole through which the cluster can be compromised. I propose that the 
security key that corosync already uses for inter-system network 
communications could be used; the file holding this key can be suitable 
protected on remote systems. I also think that it should be possible to 
disable the feature on the server.

Selection of TCP instead of local communications would be selectable at 
run-time using an environment variable. At least one variable would be 
needed to determine the remote node and (perhaps) port so the presence 
of that would be the trigger to use remote communications. Maybe we 
should register a URL type: corosync://myhost anyone ??

I think there are a huge number of potential uses for this. There is 
also the potential for abuse, I realise. But if carefully done I think 
it will be a great addition to the clustering software.

Comments?

Chrissie