k2 not k1 and ubuntu fixes

Former-commit-id: 81dc6d67

CICD/heat/gc_HOT.yaml

@@ -69,12 +69,12 @@ resources:
      list_join: [ '-', [ { get_param: "OS::stack_name" }, 'sql0' ]]
     availability_zone: { get_param: avz }
     flavor: m3.xsmall
-    image: { get_param: centos_7_image_id }
+    image: { get_param: ubuntu_1804_image_id }
     key_name: { get_param: ssh_key }
     security_groups: [ { get_param: SSHMonashSecGroupID }, { get_param: SlurmSecGroupID }, { get_param: MySQLSecGroupID }, { get_param: NFSSecGroupID } ]
     metadata:
      ansible_host_groups: [ SQLNodes, NFSNodes ]
-     ansible_ssh_user: ec2-user
+     ansible_ssh_user: ubuntu
      project_name: { get_param: project_name }
     networks:
       - network: { get_param: NetID }
@@ -267,7 +267,7 @@ resources:
        list_join: [ '-', [ { get_param: "OS::stack_name" }, 'gpudesktopu%index%' ]]
       security_groups: [ default, { get_param: SSHMonashSecGroupID }, { get_param: SlurmSecGroupID }, { get_param: NFSSecGroupID } ]
       metadata:
-       ansible_host_groups: [ DesktopNodes, GPU, ComputeNodes, K1Nodes, VisNodes ]
+       ansible_host_groups: [ DesktopNodes, GPU, ComputeNodes, VisNodes ]
        ansible_ssh_user: ubuntu
        project_name: { get_param: project_name }
       networks:

CICD/heat/gc_secgroups.hot

@@ -10,13 +10,16 @@ resources:
      name: "heatslurmsecgroup"
      rules: [ { protocol: tcp,
                port_range_min: 12000,
-               port_range_max: 12999},
+               port_range_max: 12999,
+               remote_mode: "remote_group_id"},
               { protocol: tcp,
                port_range_min: 6817,
-               port_range_max: 6819},
+               port_range_max: 6819,
+               remote_mode: "remote_group_id"},
               { protocol: tcp,
                port_range_min: 1019,
-               port_range_max: 1019}]
+               port_range_max: 1019,
+               remote_mode: "remote_group_id"}]
   NFSSecGroup:
    type: "OS::Neutron::SecurityGroup"
    properties:

CICD/plays/computenodes.yml

@@ -96,7 +96,7 @@
 
 - hosts: 'VisNodes'
   roles:
-  - { role: systemd-nvidia-uvm, tags: [ uvm ] }
+  - { role: systemd-nvidia-uvm, tags: [ uvm,SiteSpecific ] }
 
 - hosts: 'VisNodes'
   roles:

roles/deploy-xorg/files/scripts/nvidia-xconf-gen.py

@@ -11,14 +11,18 @@ from subprocess import call
 import re
 import json
 
-def grab_card_ids():
-	# This method runs nvidia-smi to grab the card ids, then returns a list
-    
-    if not os.path.isfile("/bin/nvidia-smi"):
+def getNvidia_smi_path():
+    if os.path.isfile("/bin/nvidia-smi"):
+    	return "/bin/nvidia-smi"
+    elif os.path.isfile("/usr/bin/nvidia-smi"):
+    	return "/usr/bin/nvidia-smi"
+    else:
     	print("nvidia-smi binary not found!")
-    	exit(1)
+    	exit(1)	
 
-    cmd = ["/bin/nvidia-smi", "--query-gpu=pci.bus_id","--format=csv,noheader"]
+def grab_card_ids():
+	# This method runs nvidia-smi to grab the card ids, then returns a list
+    cmd = [getNvidia_smi_path(), "--query-gpu=pci.bus_id","--format=csv,noheader"]
     p = subprocess.Popen(cmd, shell=False, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
     
     cards = []
@@ -27,15 +31,11 @@ def grab_card_ids():
         line = line.rstrip().split(":")[2]
         pcibus_num = int(re.sub('[.:]', '', line).rstrip("0"),16)
         card = "PCI:0:{}:0".format(str(pcibus_num))
-    	cards.append(card)
+        cards.append(card)
     return cards
 
 def grab_card_boardname():
-	if not os.path.isfile("/bin/nvidia-smi"):
-	 	print("nvidia-smi binary not found!")
-	 	exit(1)
-
-	cmd = ["/bin/nvidia-smi", "--query-gpu=name","--format=csv,noheader"]
+	cmd = [getNvidia_smi_path(), "--query-gpu=name","--format=csv,noheader"]
 	cards = []
 	p = subprocess.Popen(cmd, shell=False, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
 	for line in p.stdout.readlines():