AMBARI-5888. Nagios Alerts On SLES Toggle In Maintenance Mode (Jonathan Hurley via smohanty)

ambari-server/src/main/resources/stacks/HDP/1.3.2/services/NAGIOS/package/scripts/nagios_service.py

@@ -20,17 +20,50 @@ Ambari Agent
 
 """
 
+import os
+import signal
+
 from resource_management import *
+from os.path import isfile
+
 
 def nagios_service(action='start'): # start or stop
   import params
+  
+  nagios_pid_file = format("{nagios_pid_file}")
 
-  if action == 'start':
-   command = "service nagios start"
+  command_path = "/usr/local/bin/:/bin/:/sbin/"
+
+  if action == 'start': 
+    Execute("service nagios start", path = command_path )   
   elif action == 'stop':
-   command = format("service nagios stop && rm -f {nagios_pid_file}")
+    # attempt to grab the pid in case we need it later
+    nagios_pid = 0  
+    if isfile(nagios_pid_file):   
+      with open(nagios_pid_file, "r") as file:
+        try:
+          nagios_pid = int(file.read())
+          Logger.info("Nagios is running with a PID of {0}".format(nagios_pid))
+        except:
+          Logger.info("Unable to read PID file {0}".format(nagios_pid_file))
+        finally:
+          file.close()
+  
+    Execute("service nagios stop", path = command_path)
+
+    # on SUSE, there is a bug where Nagios doesn't kill the process 
+    # but this could also affect any OS, so don't restrict this to SUSE
+    if nagios_pid > 0:
+      try:
+        os.kill(nagios_pid, 0)
+      except:
+        Logger.info("The Nagios process has successfully terminated")
+      else:
+        Logger.info("The Nagios process with ID {0} failed to terminate; explicitly killing.".format(nagios_pid))
+        os.kill(nagios_pid, signal.SIGKILL)
 
-  Execute( command,
-     path    = "/usr/local/bin/:/bin/:/sbin/"      
-  )
+    # in the event that the Nagios scripts don't remove the pid file
+    if isfile( nagios_pid_file ):   
+      Execute(format("rm -f {nagios_pid_file}"))
+        
   MonitorWebserver("restart")

ambari-server/src/main/resources/stacks/HDP/2.0.6/services/NAGIOS/package/scripts/nagios_service.py

@@ -20,16 +20,50 @@ Ambari Agent
 
 """
 
+import os
+import signal
+
 from resource_management import *
+from os.path import isfile
+
 
 def nagios_service(action='start'): # start or stop
   import params
+  
+  nagios_pid_file = format("{nagios_pid_file}")
 
   if action == 'start':
-   command = format("service {nagios_service_name} start")
+    command = format("service {nagios_service_name} start")
+    Execute(command)   
   elif action == 'stop':
-   command = format("service {nagios_service_name} stop && rm -f {nagios_pid_file}")
+    # attempt to grab the pid in case we need it later
+    nagios_pid = 0    
+    if isfile(nagios_pid_file):   
+      with open(nagios_pid_file, "r") as file:
+        try:
+          nagios_pid = int(file.read())
+          Logger.info("Nagios is running with a PID of {0}".format(nagios_pid))
+        except:
+          Logger.info("Unable to read PID file {0}".format(nagios_pid_file))
+        finally:
+          file.close()
+
+    command = format("service {nagios_service_name} stop")  
+    Execute(command)
+
+    # on SUSE, there is a bug where Nagios doesn't kill the process 
+    # but this could also affect any OS, so don't restrict this to SUSE
+    if nagios_pid > 0:
+      try:
+        os.kill(nagios_pid, 0)
+      except:
+        Logger.info("The Nagios process has successfully terminated")
+      else:
+        Logger.info("The Nagios process with ID {0} failed to terminate; explicitly killing.".format(nagios_pid))
+        os.kill(nagios_pid, signal.SIGKILL)
 
-  Execute( command,      
-  )
+    # in the event that the Nagios scripts don't remove the pid file
+    if isfile( nagios_pid_file ):   
+      Execute(format("rm -f {nagios_pid_file}"))
+        
   MonitorWebserver("restart")

ambari-server/src/test/python/stacks/1.3.2/NAGIOS/test_nagios_server.py

@@ -49,19 +49,22 @@ class TestNagiosServer(RMFTestCase):
     self.assertNoMoreResources()
 
 
-  def test_stop_default(self):
+  @patch('os.path.isfile')
+  def test_stop_default(self, os_path_isfile_mock):
+    src_dir = RMFTestCase._getSrcFolder()    
+    os_path_isfile_mock.side_effect = [False, True]
+       
     self.executeScript(
       "1.3.2/services/NAGIOS/package/scripts/nagios_service.py",
       classname="NagiosServer",
       command="stop",
       config_file="default.json"
     )
-    self.assertResourceCalled('Execute',
-                              'service nagios stop && rm -f /var/run/nagios/nagios.pid',
-                              path=['/usr/local/bin/:/bin/:/sbin/']
-    )
-    self.assertResourceCalled('MonitorWebserver', 'restart',
-    )
+    
+    self.assertResourceCalled('Execute','service nagios stop', path=['/usr/local/bin/:/bin/:/sbin/'])
+    self.assertResourceCalled('Execute','rm -f /var/run/nagios/nagios.pid')
+    self.assertResourceCalled('MonitorWebserver', 'restart')
+    
     self.assertNoMoreResources()
 
 

ambari-server/src/test/python/stacks/2.0.6/NAGIOS/test_nagios_server.py

@@ -18,7 +18,9 @@ See the License for the specific language governing permissions and
 limitations under the License.
 '''
 
-from mock.mock import MagicMock, patch
+import os
+
+from mock.mock import Mock, MagicMock, patch
 from stacks.utils.RMFTestCase import *
 
 
@@ -47,18 +49,22 @@ class TestNagiosServer(RMFTestCase):
     self.assertNoMoreResources()
 
 
-  def test_stop_default(self):
+  @patch('os.path.isfile')
+  def test_stop_default(self, os_path_isfile_mock):
+    src_dir = RMFTestCase._getSrcFolder()    
+    os_path_isfile_mock.side_effect = [False, True]
+       
     self.executeScript(
       "2.0.6/services/NAGIOS/package/scripts/nagios_service.py",
       classname="NagiosServer",
       command="stop",
       config_file="default.json"
     )
-    self.assertResourceCalled('Execute',
-                              'service nagios stop && rm -f /var/run/nagios/nagios.pid'
-    )
-    self.assertResourceCalled('MonitorWebserver', 'restart',
-    )
+    
+    self.assertResourceCalled('Execute','service nagios stop')
+    self.assertResourceCalled('Execute','rm -f /var/run/nagios/nagios.pid')
+    self.assertResourceCalled('MonitorWebserver', 'restart')
+    
     self.assertNoMoreResources()
 
 

contrib/addons/src/addOns/nagios/scripts/nagios_alerts.php

@@ -100,13 +100,12 @@ function hdp_mon_generate_response( $response_data )
   define ("WEBHCAT_SERVICE_CHECK", "WEBHCAT::WebHCat Server status check");
   define ("PUPPET_SERVICE_CHECK", "PUPPET::Puppet agent down");
 
-  /* If SUSE, status file is under /var/lib/nagios */
-  if (file_exists("/etc/SuSE-release")) {
-    $status_file="/var/nagios/status.dat";
-  } else {
-    $status_file="/var/nagios/status.dat";
+  // on SUSE, some versions of Nagios stored data in /var/lib
+  $status_file = "/var/nagios/status.dat";
+  if (!file_exists($status_file) && file_exists("/etc/SuSE-release")) {
+    $status_file = "/var/lib/nagios/status.dat";
   }
-
+  
   $q1="";
   if (array_key_exists('q1', $_GET)) {
     $q1=$_GET["q1"];
@@ -296,11 +295,10 @@ function hdp_mon_generate_response( $response_data )
     #echo $matches[1][0] . ", " . $matches[1][1] . "\n";
     $services_objects = array ();
     $i = 0;
-    foreach ($matches[1] as $object) {
+    foreach ($matches[1] as $object) {      
       $servicestatus = getParameterMap($object, $servicestatus_attributes);
       switch ($alert_type) {
       case "all":
-
         if (empty($host) || $servicestatus['host_name'] == $host) {
           $servicestatus['service_type'] = get_service_type($servicestatus['service_description']);
           $srv_desc = explode ("::",$servicestatus['service_description'],2);
@@ -353,12 +351,14 @@ function hdp_mon_generate_response( $response_data )
         }
         break;
       }
+      
       if (!empty($servicestatus)) {
         $services_objects[$i] = $servicestatus;
         $i++;
       }
     }
-    /* echo "COUNT : " . count ($services_objects) . "\n"; */
+
+    // echo "COUNT : " . count ($services_objects) . "\n";
     return $services_objects;
   }
 
@@ -459,6 +459,7 @@ function hdp_mon_generate_response( $response_data )
 
     return $map;
   }
+  
 function indent($json) {
 
     $result      = '';