The process of deciding where in our virtual infrastructure to position a new virtual machine - the hypervisor or cluster, and datastore - is another step that can be automated as part of the VM provisioning workflow. We might wish to locate VMs on the cluster with the lightest current load for example, or restrict visibility of selected datastores to some of our users.
ManageIQ refers to this process as placement, and there is a corresponding Placement stage in the VMProvision_VM state machine.
In this chapter we’ll look at the various options for automating placement, and how we can create customised placement algorithms to suit our own requirements.
There are several alternative placement methods that we can use out-of-the-box to determine where to place our new virtual machines, for example there are three in the ManageIQ domain (see Placement methods in the ManageIQ domain).
The default value for the Placement stage in the VMProvision_VM/template state machine is as follows:
/Infra.../VM/Provisioning/Placement/default#${/#miq_provision.source.vendor}
We can see that this URI includes a message component at the end, which corresponds to the runtime value of the ${/#miq_provision.source.vendor} attribute. This is the string value for the provider type that we are provisioning into.
If we look at the schema fields in the /Infrastructure/VM/Provisioning/Placement/default placement instance, we see that the messages correspond to the same provider type string: 'redhat', 'vmware' or 'microsoft' (see The schema fields of the default placement instance).
This is a neat way of handling provider-specific placement requirements in an automated manner. The message is dynamically translated for each provisioning operation, and this selects the correct placement method.
The redhat_best_fit_cluster method just places the new VM into the same cluster as the source template. The other two methods select the host with the least running VMs, and most available datastore space.
We can see that we have a lot of per-provider control of the placement options available to us when we provision a virtual machine. We can also add our own placement methods to take into account our own specific requirements if we wish.
When we start working with custom placement methods, we also need to take into account the infrastructure components that a user can see from their role-based access control filters. When we configure ManageIQ access control groups, we can set optional assigned filters to selected hosts and clusters. We can also restrict a group’s visibility of infrastructure components to those tagged with specific tags. If we use assigned filters in this way, we need to ensure that our placement logic doesn’t select a host, cluster or datastore that the user doesn’t have RBAC permission to see, otherwise the provisioning operation will fail.