Lars Marowsky-Bree wrote:
On 2006-01-16T17:48:44, James Pan <[EMAIL PROTECTED]> wrote:
Hi, all.
currently start up heartbeat will lead the machine to join the
cluster. if no nodes defined in ha.cf, heartbeat will fail to start.
Could we make a little change? Start heartbeat even without nodes
defined?
If there are nodes defined, heartbeat work in the current way. If no
nodes defined, heartbeat will not activate the membership components,
the machine is out of the cluster. So heartbeat can be running on
every machine by default, but the machine is not necessary to be a
member.
Users may run cl_addnode or something like that to add a machine to
the cluster, then heartbeat may update the ha.cf on the new add
machine. hence the HA service will be running automatically on this
node.
In this way, heartbeat become the reliable communication
infrastructure and a supervisor for all components. I think this is
good for management at least.
This sounds great, because then we can dynamically add nodes later.
Yes, and it is good for our management tools.
Because the management tools will have the capacity to restart the
high-availability service on Node B
from node A. Therefor the management daemon need communication between
these two nodes.
If heartbeat can startup without nodes defined and starting heartbeat
will not lead a machine to join
the cluster, we can have heartbeat running on every machine.
So the management daemon may utilize heartbeat instead of implementing
another communication
mechanism to communicate with each node, since the heartbeat
communication mechanism is mature
and has plentiful HBcomm plugins.
In this case,
heartbeat is running does not mean the HA service is running, just means
the machine is ready to
start the HA service,
The HA service on a node will start if we add this node to the cluster.
And the HA service on this node
will start automatically even after reboot, because when this node was
added to the cluster, the heartbeat
running on this node has updated the ha.cf, now it has nodes defined.
We can stop the HA service by remove the node from the cluster.
Sincerely,
Lars Marowsky-Brée
--
JMP - James Pan
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
