[Linux-ha-dev] lha-md model.
Andrew Beekhof (GMail)
beekhof at gmail.com
Thu Feb 17 12:13:05 MST 2005
On Feb 17, 2005, at 7:39 PM, Kashif Shaikh wrote:
>> From the lha-md.txt:
>> The lha-md system is comprised by two main elements:
>> o lha-md resource
>> o lha-md program
>> The lha-md resource is an OCF resource which activates or deactivates
>> daemon on the cluster following the cluster behavior (specifically,
>> following the Designated Coordinator behaviour).
>> The lha-md daemon is a TCP server which attends to the requests made
>> by the
>> clients. It is activated or deactivated by the lha-md resource.
>> To be sure that the lha-md will be always available for the clients,
>> linked to an IPaddr resource which will be activated or deactivated
>> by the
>> This way, clients can always connect to the cluster management daemon
>> and manage the cluster from anywhere.
> Hmmm...there shouldn't be a requirement that lha-md should always be
> available from an IPaddr resource. I mean the end-user should be given
> option of creating an IPaddr resource to manage the lha-md resource.
I believe thats what he has in mind. That the user defines an IPaddr
resource and specifies that lha-md starts afterwards and always on the
Personally I would go a step further and make this implicit in the
lha-md resource agent but thats something thats easily changed later.
> Another thing: is a single-daemon good enough? Maybe running an lha-md
> daemon on each node is better, where they communicate vi HBcomm API.
> instance some admin operations can be done on one node(i.e. adding a
> where the underlying CIB handles the cluster-wide distribution and
> notification of the node) but some operations must be cluster-wide
> they is no underlying layer to handle such operations.
> A simple command such
> as "shutdown cluster" will need to run 'shutdown -t now' on each
Well I dont think stopping all resources and nodes in the cluster is a
I would also argue that most of the time people dont want to actually
shut down the nodes on the cluster, they want to bring down the cluster
layer so they can work on the nodes (which can be co-ordinated from one
node). But feel free to differ.
> So unless management operations are delegated to a cluster-wide
> CRM), I think it will be better to design lha-md as a multi-node
> that is possible).
I'm still yet to see the need for this.
>> -----Original Message-----
>> From: linux-ha-dev-bounces at lists.linux-ha.org
>> [mailto:linux-ha-dev-bounces at lists.linux-ha.org]On Behalf Of Rodrigo
>> Henriquez M. - Corporacion Linux S.A.
>> Sent: Tuesday, February 15, 2005 12:12 PM
>> To: High-Availability Linux Development List
>> Subject: [Linux-ha-dev] lha-md model.
>> Hi again.
>> Here is the lha-md model with the modifications suggested for some of
>> Feel free to do all your suggestions or critiques.
>> -- Ro
>> No virus found in this incoming message.
>> Checked by AVG Anti-Virus.
>> Version: 7.0.300 / Virus Database: 265.8.8 - Release Date: 2/14/2005
> Linux-HA-Dev: Linux-HA-Dev at lists.linux-ha.org
> Home Page: http://linux-ha.org/
"No means no, and no means yes, and everything in between and all the
rest" - TISM
More information about the Linux-HA-Dev