Showing results for 
Search instead for 
Did you mean: 

Service Group Failover Order ?

Level 2

I have a 4 node cluster using a load failover policy.

On one node I have four service groups with respective weightings of 50, 10, 10, 10

If this node fails I want to ensure that the service group with a weighting of 50 is the first to be assessed for failover and the groups with a weighting of 10 are only assessed once the 50 has been removed from the available capacity on the node it comes up on.

Is there a mechanism to configure this ?

Level 3
Employee Certified

Allgroups go go online in parallel.

To have certain groups go online first, put dependencies between groups.

Say Group4 (with 10 weight), will depend on Group1 (with 50 weight)......

Dependencies can go 4 groups deep ... and there are a number of different dependencies (which will cause different  behaviours).
All of this in the Group Dependencies part of the User's Guide (search for your version doco)

Level 2

"All groups go online in parallel." - I do not think this statement is totally correct. The online may appear to work in parallel in that you do not need to wait for one group to online before another starts onlining but from the VCS Server Guide -

"During a failover involving multiple service groups, failover decisions are made serially to facilitate a proper load-based choice."

I would like to know the criteria for what order these failover decisions are made. The documentation does not appear to make any reference to this

Level 6

The reason the documentation doesn't make reference to the "criteria" you're asking about is because you don't need to know this detail to achieve the result you describe here.

As mightknowsomething mentioned, service group dependencies can be used to control the group behaviour - ie: dependencies are the mechanism to control group behaviour.

If we call the groups A (load 50), B (10), C (10), D (10) - from what you described, you basically want A to be online before B, C, D are evaluated to go online.

Look at the descriptions of online global soft or online global firm dependencies (depending on whether you want B, C, D to be restarted/remain online if A faults) - would basically be making A the child and B, C, D would be parent.

Level 2

Thanks to g_lee and mightknowsomething. It looks like the soft dependencies should do what I require.

However I still think there must be some sort of rule to the order that the service groups are dealt with serially when they come online (even if it is just alphabetical). If I could configure them using that and not use the soft dependencies then I think that would reduce complexity within the solution.