Gary,
You can achieve this behavior using separate resource groups for each data
service. One RG would contain the LogicalHostname and the HAStoragePlus
resource for the zpool. Each data service would be in its own RG which would
have a +++ affinity on the network/storage RG. Each data service resource would
have Resource_dependencies upon the network and storage resources. Each data
service resource should have Failover_mode set to SOFT.
You would adjust the Retry_count and Retry_interval properties of the data
service resources to acheive the desired restart/failover behavior upon failure.
This part of the configuration is data-service specific, so you might have to
check the documentation for the particular agent to get the correct settings.
- Marty
On 07/15/11 09:14, Gary Mills wrote:
May be it would help to define what you expected
behaviour you want and
the constraints you have. Marty and Detlef have
suggested different
options for you to try - may be those options will
meet your expectations.
Certainly. I have a two-node cluster with both nodes
connected to a storage device that contains two disks.
These are mirrored into a single zpool. I want to use
this cluster to provide a number of small services in
an HA manner. This is mainly so that we can do maintenance
on one node while the other provides the service, but also
to handle failures automatically. Examples of the services
are: DNS, Mysql, LDAP, and Apache. None of them are
used heavily, so it makes sense to supply them from one
system. As well, I'd like all of those services to be available
from one IP address.
So, the zpool would be imported on one node, the same one
that has the logical hostname presumably. The services would
also all run on that same node. In case of a hardware failure,
I'd expect all of the services to reappear on the other node.
In case of a software failure that only affected one service,
I'd expect the cluster manager to attempt to restart that
service, and if that failed, to migrate them all to the other
node and try again. If that too failed, I'd want all of the other
services to continue to run, except for the one that was broken.
We'd have to fix it ourselves.
I do appreciate all of your suggestions, by the way. Perhaps
this explanation will clarify things. At the moment, I'm still
undecided on my course of actions.
_______________________________________________
ha-clusters-discuss mailing list
ha-clusters-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/ha-clusters-discuss
