Use clusterid attribute to filter servers in dynamic inventory

  - At this moment there is only condition that verifies either
    server has clusterid attribute in metadata, if yes then we
    add it to inventory and use in deployment. Such approach is not
    working for deployment of multiple environments in the same
    openstack project. Env variable `OPENSHIFT_CLUSTER` is used
    to indentify to which openshift cluster, servers are related,
    if variable is not defined then old approach is used.

  - Update openstack configuration docs

playbooks/openstack/configuration.md

@@ -18,6 +18,7 @@ Environment variables may also be used.
 * [DNS Configuration](#dns-configuration)
 * [Floating IP Address Configuration](#floating-ip-address-configuration)
 * [All-in-one Deployment Configuration](#all-in-one-deployment-configuration)
+* [Multi-env Deployment Configuration](#multi-env-deployment-configuration)
 * [Building Node Images](#building-node-images)
 * [Kuryr Networking Configuration](#kuryr-networking-configuration)
 * [Provider Network Configuration](#provider-network-configuration)
@@ -546,6 +547,36 @@ added, because there are no dedicated infra nodes, so you will have to add it
 manually. See
 [Custom DNS Records Configuration](#custom-dns-records-configuration).
 
+## Multi-env Deployment Configuration
+
+If you want to deploy multiple OpenShift environments in the same OpenStack
+project, you can do so with a few configuration changes.
+
+First, set the `openshift_openstack_clusterid` option in the
+`inventory/group_vars/all.yml` file with specific unique name for cluster.
+
+```
+vi inventory/group_vars/all.yml
+
+openshift_openstack_clusterid: foobar
+openshift_openstack_public_dns_domain: example.com
+```
+
+Second, set `OPENSHIFT_CLUSTER` environment variables. The `OPENSHIFT_CLUSTER`
+environment variable has to consist of `openshift_openstack_clusterid` and
+`openshift_openstack_public_dns_domain`, that's required because cluster_id
+variable stored in the instance metadata is concatanated in the same way.
+If value will be different then instances won't be accessible in ansible inventory.
+
+```
+export OPENSHIFT_CLUSTER='foobar.example.com'
+```
+
+Then run the deployment playbooks as usual. When you finish deployment of first
+environment, please update above options that correspond to a new environment
+and run the deployment playbooks.
+
+
 ## Building Node Images
 
 It is possible to build the OpenShift images in advance (instead of installing

playbooks/openstack/resources.py

@@ -19,6 +19,8 @@ except ImportError:
 from keystoneauth1.exceptions.catalog import EndpointNotFound
 import shade
 
+OPENSHIFT_CLUSTER = os.getenv('OPENSHIFT_CLUSTER')
+
 
 def base_openshift_inventory(cluster_hosts):
     '''Set the base openshift inventory.'''
@@ -124,11 +126,12 @@ def build_inventory():
     # Use an environment variable to optionally skip returning the app nodes.
     show_compute_nodes = os.environ.get('OPENSTACK_SHOW_COMPUTE_NODES', 'true').lower() == "true"
 
-    # TODO(shadower): filter the servers based on the `OPENSHIFT_CLUSTER`
-    # environment variable.
+    # If `OPENSHIFT_CLUSTER` env variable is defined then it's used to
+    # filter servers by metadata.clusterid attribute value.
     cluster_hosts = [
         server for server in cloud.list_servers()
-        if 'metadata' in server and 'clusterid' in server.metadata and
+        if 'clusterid' in server.get('metadata', []) and
+        (OPENSHIFT_CLUSTER is None or server.metadata.clusterid == OPENSHIFT_CLUSTER) and
         (show_compute_nodes or server.metadata.get('sub-host-type') != 'app')]
 
     inventory = base_openshift_inventory(cluster_hosts)
@@ -183,7 +186,7 @@ def build_inventory():
 
 def _get_stack_outputs(cloud_client):
     """Returns a dictionary with the stack outputs"""
-    cluster_name = os.getenv('OPENSHIFT_CLUSTER', 'openshift-cluster')
+    cluster_name = OPENSHIFT_CLUSTER or 'openshift-cluster'
 
     stack = cloud_client.get_stack(cluster_name)
     if stack is None or stack['stack_status'] not in (