Can the SUNW.gds and SUNW.apache resource types be used together?
Howdy,
We are looking at setting up a number of Solaris zones on highly available shared storage, and would like to configure these zones to failover to other nodes in the cluster. Setting this up is pretty straight forward, and is covered in the data service guide for Solaris containers. What I can't seem to locate is a reference to using application data services (e.g., SUNW.apache, etc.) to manage applications that are running inside SUNW.gds managed zones. Does anyone happen to know if this is supported? The "Writing Scripts for the Zone Script Resource" section in the containers guide describes how to stop, start and probe applications inside a SUNW.gds manage zone using the SUNW.gds data service framework, which leads me to wonder if we should be using that to manage the applications that are running inside SUNW.gds manage zones.
Thanks,
- Ryan
[888 byte] By [
matty_91a] at [2007-11-27 1:54:28]
# 1
You did not mention which Sun Cluster version you are using.
With Sun Cluster 3.1 08/05 the Sun Cluster framework is only running in the global zone. So within a non-global zone there is no Sun Cluster framework, especially no DSDL API (like the scha_* calls, PMF, etc). So if you are using SC 3.1, the only option is to use the either the sczsh or sczsmf component coming with the HA Container agent. A standard agent can not run within the non-global zone, unless he is failover zone aware (like tomcat, samba, mysql, postgresql, N1 SPS or IBM Webspehere MQ). Apache is unfortunatly not failover zone aware.
With Sun Cluster 3.2 part of the Sun Cluster framework (like DSDL, scha_* callse, PMF, etc) is now also available from within a non-global zone as well. This means such a non-global zone is acting like a virtual node which can also receive a resource groupe by using the nodename:zonename notation within the Nodelist property of the RG.
With Sun Cluster 3.2 you can also combine the HA Container agent with the new nodename:zonename notation. The question is, why do you want to do that, ie. which advantage are you expecting by doing that . In order to decide between the two, the Sun Cluster concept guide has some decision help:
http://docs.sun.com/app/docs/doc/819-2969/6n57kl13t?a=view
So if you have identified the combination of the two possibilities gives you an advantage, then you can do the following:
- configure RG-A with the sczbt component to failover your non-global zone
- configure RG-B with Nodelist=node-a:failoverzone,node-b:failoverzone, configure the apache resource within RG-B, and setup a RG dependecy from RG-B to RG-A.
Now there is a caveat: the apache resource can only perfrom restarts within the failoverzone. It can not trigger a failover, since for RGM the failover zone is only up at one node at a time (this is the concept of a failover zone) - so on the other node the failover zone (since not up) is not recognized as a valid failover target at the point where RG-B would try trigger a give-over.
Of cousrse, if RG-A should decide a give-over. RG-B would follow that through the dependency.
If the above caveat is not a problem for your configuration, then you can use it.
Otherwise, why not have a non-global zone per node, and just configure a RG with Nodelist=node-a:zone-a,node-b:zone-b and let the RG contain the HASP, LH and apache reosurce.
Hope this helps.
Greets
Thorsten
