Merge branch 'master' of https://github.com/monash-merc/ansible_cluster_in_a_box

dynamicInventory-mcc2

 #!/usr/bin/env python
 import sys, os, string, socket, re
 import shlex, multiprocessing, time, shutil, json
-import novaclient.v1_1.client as nvclient
+from novaclient import client as nvclient
 import novaclient.exceptions as nvexceptions
 import keystoneclient.v2_0.client as ksclient
 from joblib import Parallel, delayed
 from multiprocessing import Process, Manager, Pool
+from libnmap.process import NmapProcess
+from libnmap.parser import NmapParser, NmapParserException
 
 def gatherInfo(tenantName, tenantID, userName, passwd, authUrl, inventory):
-	
 	## Fetch the Nova Object
 	projectName = os.path.basename(sys.argv[0])
 	nc = nvclient.Client(	auth_url=authUrl,
@@ -16,24 +17,23 @@ def gatherInfo(tenantName, tenantID, userName, passwd, authUrl, inventory):
 		api_key=passwd,
 		project_id=tenantName,
 		tenant_id=tenantID,
-		service_type="compute"
+		version="2"
 		)
 	for server in nc.servers.list():
 		if server.metadata and \
 		'ansible_host_groups' in server.metadata and \
-		server.metadata['project_name'] == projectName.strip():
+		'project_name' in  server.metadata:
+			if server.metadata['project_name'].strip() != projectName.strip(): continue
 			unwantedChars = """][")("""
 			rgx = re.compile('[%s]' % unwantedChars)
 			ansible_groups = rgx.sub('', server.metadata['ansible_host_groups']).split(',')
 			hostname = socket.gethostbyaddr(server.networks.values()[0][0])[0]
-			closed=True
-			while closed:
-				hostSocket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
-				if not hostSocket.connect_ex((hostname, 22)): 
-					closed = False
-					break
-				time.sleep(5)
-				hostSocket.close()
+			novaVolumes = nc.volumes.get_server_volumes(server.id)
+			# Let's do some port scanning using nmap
+			nmproc = NmapProcess(hostname, "-p 22 -sV -Pn")
+			rc = nmproc.run()
+			if rc != 0: continue
+			parsed = NmapParser.parse(nmproc.stdout)
 			# Set Ansible Host Group
 			for group in ansible_groups:
 				groupName = group.strip()
@@ -43,11 +43,12 @@ def gatherInfo(tenantName, tenantID, userName, passwd, authUrl, inventory):
 			for key, value in server.metadata.iteritems():
 				if key not in ('project_name','ansible_host_groups'):
 					inventory['_meta']['hostvars'][hostname] = { key:value }
+			if novaVolumes:
+				inventory['_meta']['hostvars'][hostname]['volumeList'] = [ volume.id for volume in novaVolumes ]
+			inventory['_meta']['hostvars'][hostname]['status']  = parsed.hosts[0].status
 		else:
 			continue
 	#print inventory
-	#inventoryList.append(inventory)
-	#print json.dumps(inventory)
 
 if __name__ == "__main__":
 	inventory = {}

roles/centos7Base/tasks/installBasePackages.yml

+---
+-
+ name: Removing the RDO repository
+ file: path=/etc/yum.repos.d/rdo-release.repo state=absent
+ sudo: true
+-
+ name: Install epel-release
+ yum: name=epel-release-7-5.noarch state=present
+ sudo: true
+-
+ name: Enable epel
+ command: yum-config-manager --enable epel
+ sudo: true
+-
+ name: Installing Base Packages
+ yum: name={{ item }} state=present
+ with_items:
+  - yum-utils
+  - deltarpm-3.6-3.el7.x86_64
+  - yum-plugin-versionlock
+ sudo: true
+-
+ name: Installing Core packages
+ yum: name="{{ item.software }}-{{ item.version }}.{{ item.arch }}" state=present
+ with_items: package_list
+ sudo: true
+-
+ name: Performing version lock on the packages
+ shell: yum versionlock \*
+ sudo: true
+

roles/centos7Base/tasks/main.yml

+---
+- include: installBasePackages.yml

roles/gpu/tasks/main.yml

@@ -60,8 +60,8 @@
   
 - name: restart_host
   command: shutdown -r now "Reboot triggered by Ansible"
-  async: 0
-  poll: 0
+  async: 900
+  poll: 60
   sudo: true
   ignore_errors: true
   when: has_been_compiled | failed

roles/karaage3.1.17/tasks/karaage.yml

@@ -166,7 +166,6 @@
  sudo: true
  when: karaage_db_init.stdout.find("0") == 0
 
-
 - name: install postfix
   apt: name=postfix state=present
   sudo: true

roles/karaage3.1.17/templates/kg_init.j2

@@ -8,7 +8,8 @@ from karaage.institutes.models import Institute
 from karaage.machines.models import MachineCategory
 from karaage.people.models import Person, Group
 
-DEBUG = False
+CONSOLE_DEBUG = False
+
 class HpcIdInit():
     import django
     django.setup()
@@ -18,6 +19,9 @@ class HpcIdInit():
         self.path = configfile 
         self.password = password
         self.debug = debug
+        if not debug:
+            self.logfile = open("/tmp/kg_init.log", "w") 
+
         if self.path and os.path.exists(self.path):
             with open(self.path) as data:
                 config_data = json.load(data)
@@ -27,11 +31,14 @@ class HpcIdInit():
         else:
             log("Invalid input data")    
 
+    def __del__(self):
+        self.logfile.close()
+
     def log(self, message):
         if self.debug:
             print message
         else:
-            pass
+            self.logfile.write(message + "\n") 
 
     def getGroup(self, name):
         group = None
@@ -72,6 +79,7 @@ class HpcIdInit():
                 self.log("Find insititute %s" %(institute.name))
                 project = Project.objects.create(pid = pid, name = name, institute = institute, group = institute.group, is_active = True, is_approved = True, approved_by = superuser)
                 if project:
+                    project.leaders.add(superuser)
                     self.log("Create project OK")
                 else:
                     self.log("Create project failed")
@@ -156,11 +164,32 @@ class HpcIdInit():
             institute = Institute.objects.get(name = user["institute_name"])
             if institute:
                 person = Person.objects.create(username = user["username"], email = user["email"], password = self.password, short_name = user["short_name"], full_name = user["full_name"], is_admin = True, is_active = True, institute = institute)    
-                person.full_clean()
+                if person:
+                    person.set_password(self.password)
+                    person.save()
+                    result = self.addInstituteDelegate(person, institute)
+                    if result:
+                        log("Add super user %s to institute %s delegate" %(person.username, institute.name))
+                    else:
+                        log("Faired to add super user %s to institute %s delegate" %(person.username, institute.name))
+                    person.full_clean()
         except:
             log("Create super user exception: %s" %(traceback.format_exc()))
         finally:
             return person
+    
+    def addInstituteDelegate(self, su, institute):
+        result = True 
+        try:
+            delegates = institute.delegates.all().filter(username = su.username)
+            if len(delegates) == 0:
+                self.log("Create institution delegate %s"%(su.username))
+                institute.delegates.add(su)
+        except:
+            result = False 
+            self.log("Create institution delegate exception: %s" %(traceback.format_exc()))
+        finally:
+            return result 
 
     def setup(self):
         self.log("Password = %s, debug = %s" %(self.password, self.debug))
@@ -176,7 +205,7 @@ class HpcIdInit():
         if self.mc:
             mc = self.getOrCreateMachineCategory(self.mc)
             if mc:
-                self.log("Get machine category = '%s'" %(mc.name))
+                self.log("Get machine category = '%s'" %(self.mc))
             else:    
                 self.log("Failed to get machine category = '%s'" %(self.mc))
         if su:
@@ -198,7 +227,7 @@ def main(argv):
     if len(sys.argv) > 2:
         config_path = argv[0]
         password = argv[1]
-        debug = DEBUG 
+        debug = CONSOLE_DEBUG 
         if len(sys.argv) > 3:
             debug = argv[2]
         init = HpcIdInit(config_path, password, debug)

roles/karaage3.1.17/templates/settings.py.j2

@@ -31,6 +31,10 @@
 #
 # DEBUG = True
 
+{% if user_id_file is defined %}
+USER_ID_FILES = {{ user_id_file }}
+{% endif %}
+
 # Implemented by Shahaan due to the django-pipeline bug
 PIPELINE_ENABLED = False
 STATICFILES_STORAGE = 'pipeline.storage.PipelineStorage'

roles/modulefiles/tasks/main.yml

@@ -25,3 +25,4 @@
     line: /usr/local/Modules/modulefiles
   ignore_errors: true
   sudo: true
+  when: ansible_os_family == 'Debian'

roles/mysql/handlers/main.yml

 ---
 - name: "Restart MySQL" 
-  service: name=mysql state=restarted
+  service: name={{ sqlServiceName }} state=restarted
   sudo: true
-  when: ansible_os_family == "Debian"
-
-- name: "Restart MySQL" 
-  service: name=mysqld state=restarted
-  sudo: true
-  when: ansible_os_family == "RedHat"
-

roles/slurm-common/tasks/installSlurmFromSource.yml

@@ -32,3 +32,7 @@
   args:
     chdir: /tmp/slurm-{{ slurm_version }}
     creates: "{{ slurm_dir }}/bin/srun"
+
+- name: add slurm log rotate config
+  template: src=slurmlog.j2 dest=/etc/logrotate.d/slurm mode=644
+  sudo: true

roles/slurm-common/templates/nhclog.j2

@@ -4,6 +4,7 @@
    missingok
    notifempty
    weekly
-   size 200k
+   compress 
+   size 800k
 }
 

roles/slurm-common/templates/slurmlog.j2

+{% if slurmctrl == inventory_hostname %}
+{{ slurmctlddebug.log }}
+{{ slurmschedlog.log }}
+{% else %}
+{{ slurmddebug.log }}
+{% endif %}
+{
+ compress
+ missingok 
+ nocopytruncate 
+ nocreate 
+ nodelaycompress 
+ nomail  
+ notifempty 
+ noolddir 
+ rotate 5 
+ sharedscripts 
+ size=5M 
+ create 640 slurm root 
+{% if ansible_os_family == 'RedHat' and ansible_distribution_version >= '7' %}
+ postrotate
+{% if slurmctrl == inventory_hostname %}
+   {{ slurm_dir }}/sbin/slurmctld flushlogs 1>/dev/null || true
+{% else %}
+   {{ slurm_dir }}/sbin/slurmd flushlogs 1>/dev/null || true
+{% endif %}
+{% else %}
+ postrotate /etc/init.d/slurm reconfig 
+{% endif %}
+ endscript
+}
+

roles/slurmdb-config/tasks/main.yml

@@ -32,3 +32,7 @@
   sudo: true
   when: slurm_dir is not defined
 
+- name: add slurm db log rotate config
+  template: src=slurmdblog.j2 dest=/etc/logrotate.d/slurmdb mode=644
+  sudo: true
+

roles/slurmdb-config/templates/slurmdblog.j2

+{{ slurmdbdlog.log }}
+{% endif %}
+{
+ compress
+ missingok 
+ nocopytruncate 
+ nocreate 
+ nodelaycompress 
+ nomail  
+ notifempty 
+ noolddir 
+ rotate 5 
+ sharedscripts 
+ size=5M 
+ create 640 slurm root 
+{% if ansible_os_family == 'RedHat' and ansible_distribution_version >= '7' %}
+  postrotate
+    {{ slurm_dir }}/sbin/slurmdbd flushlogs  1>/dev/null || true
+{% else %}
+ postrotate /etc/init.d/slurmdbd reconfig 
+{% endif %}
+ endscript
+}
+