VPC: check only ssh services if VPC VR does not have any guest network

Author: weizhouapache

core/src/main/java/com/cloud/agent/api/routing/SetMonitorServiceCommand.java

@@ -37,6 +37,7 @@ public class SetMonitorServiceCommand extends NetworkElementCommand {
     public static final String ROUTER_HEALTH_CHECKS_BASIC_INTERVAL = "router.health.checks.basic.interval";
     public static final String ROUTER_HEALTH_CHECKS_ADVANCED_INTERVAL = "router.health.checks.advanced.interval";
     public static final String ROUTER_HEALTH_CHECKS_EXCLUDED = "router.health.checks.excluded";
+    public static final String ROUTER_HEALTH_CHECKS_INCLUDED_SERVICES = "router.health.checks.included.services";
 
     private MonitorServiceTO[] services;
     private Map<String, String> healthChecksConfig;

core/src/main/java/com/cloud/agent/resource/virtualnetwork/facade/SetMonitorServiceConfigItem.java

@@ -70,6 +70,7 @@ private void setupHealthChecksRelatedInfo(MonitorService monitorService, SetMoni
         }
 
         monitorService.setExcludedHealthChecks(command.getAccessDetail(SetMonitorServiceCommand.ROUTER_HEALTH_CHECKS_EXCLUDED));
+        monitorService.setIncludedServices(command.getAccessDetail(SetMonitorServiceCommand.ROUTER_HEALTH_CHECKS_INCLUDED_SERVICES));
         monitorService.setHealthChecksConfig(command.getHealthChecksConfig());
     }
 

core/src/main/java/com/cloud/agent/resource/virtualnetwork/model/MonitorService.java

@@ -27,6 +27,7 @@ public class MonitorService extends ConfigBase {
     public Integer healthChecksBasicRunInterval;
     public Integer healthChecksAdvancedRunInterval;
     public String excludedHealthChecks;
+    public String includedServices;
     public Map<String, String> healthChecksConfig;
 
     public MonitorService() {
@@ -92,6 +93,10 @@ public void setExcludedHealthChecks(String excludedHealthChecks) {
         this.excludedHealthChecks = excludedHealthChecks;
     }
 
+    public void setIncludedServices(String includedServices) {
+        this.includedServices = includedServices;
+    }
+
     public void setHealthChecksConfig(Map<String, String> healthChecksConfig) {
         this.healthChecksConfig = healthChecksConfig;
     }

server/src/main/java/com/cloud/network/router/VirtualNetworkApplianceManagerImpl.java

@@ -70,6 +70,7 @@
 import org.apache.cloudstack.utils.CloudStackVersion;
 import org.apache.cloudstack.utils.identity.ManagementServerNode;
 import org.apache.cloudstack.utils.usage.UsageUtils;
+import org.apache.commons.collections.CollectionUtils;
 import org.apache.commons.lang3.ObjectUtils;
 import org.apache.commons.lang3.StringUtils;
 import org.apache.log4j.Logger;
@@ -1595,17 +1596,22 @@ private SetMonitorServiceCommand createMonitorServiceCommand(DomainRouterVO rout
         command.setAccessDetail(SetMonitorServiceCommand.ROUTER_HEALTH_CHECKS_ENABLED, RouterHealthChecksEnabled.value().toString());
         command.setAccessDetail(SetMonitorServiceCommand.ROUTER_HEALTH_CHECKS_BASIC_INTERVAL, RouterHealthChecksBasicInterval.value().toString());
         command.setAccessDetail(SetMonitorServiceCommand.ROUTER_HEALTH_CHECKS_ADVANCED_INTERVAL, RouterHealthChecksAdvancedInterval.value().toString());
+
+        final List<Long> routerGuestNtwkIds = _routerDao.getRouterNetworks(router.getId());
         String excludedTests = RouterHealthChecksToExclude.valueIn(router.getDataCenterId());
         if (router.getIsRedundantRouter()) {
             // Disable gateway check if VPC has no tiers or no active VM's in it
-            final List<Long> routerGuestNtwkIds = _routerDao.getRouterNetworks(router.getId());
             if (RedundantState.BACKUP.equals(router.getRedundantState()) ||
                     routerGuestNtwkIds == null || routerGuestNtwkIds.isEmpty()) {
                 excludedTests = excludedTests.isEmpty() ? BACKUP_ROUTER_EXCLUDED_TESTS : excludedTests + "," + BACKUP_ROUTER_EXCLUDED_TESTS;
             }
         }
 
         command.setAccessDetail(SetMonitorServiceCommand.ROUTER_HEALTH_CHECKS_EXCLUDED, excludedTests);
+        if (router.getVpcId() != null && CollectionUtils.isEmpty(routerGuestNtwkIds)) {
+            command.setAccessDetail(SetMonitorServiceCommand.ROUTER_HEALTH_CHECKS_INCLUDED_SERVICES, MonitoringService.Service.Ssh.toString().toLowerCase());
+        }
+
         command.setHealthChecksConfig(routerHealthCheckConfig);
         command.setReconfigureAfterUpdate(reconfigure);
         command.setDeleteFromProcessedCache(deleteFromProcessedCache); // As part of updating

systemvm/debian/opt/cloud/bin/cs/CsMonitor.py

@@ -69,6 +69,12 @@ def setupHealthChecksConfigFile(self):
         else:
             hc_data["excluded_health_checks"] = []
 
+        if "included_services" in self.dbag:
+            included_services = self.dbag["included_services"]
+            hc_data["included_services"] = [ch.strip() for ch in included_services.split(",")] if len(included_services) > 0 else []
+        else:
+            hc_data["included_services"] = []
+
         if "health_checks_config" in self.dbag:
             hc_data["health_checks_config"] = self.dbag["health_checks_config"]
         else:

systemvm/debian/opt/cloud/bin/cs_monitorservice.py

@@ -33,4 +33,9 @@ def merge(dbag, data):
     if "health_checks_config" in data:
         dbag["health_checks_config"] = data["health_checks_config"]
 
+    if "included_services" in data:
+        dbag["included_services"] = data["included_services"]
+    else:
+        dbag["included_services"] = ""
+
     return dbag

systemvm/debian/root/monitorServices.py

@@ -247,7 +247,7 @@ def checkProcessStatus( process ):
         return StatusCodes.RUNNING, True
 
 
-def monitProcess( processes_info ):
+def monitProcess( processes_info, included_services ):
     """
     Monitors the processes which got from the config file
     """
@@ -263,7 +263,10 @@ def monitProcess( processes_info ):
     #time for noting process down time
     csec = repr(time.time()).split('.')[0]
 
-    for process,properties in processes_info.items():
+    for process,properties in list(processes_info.items()):
+        if included_services and len(included_services) > 0 and process not in included_services:
+            printd ("---------------------------\nskipping the service %s\n---------------------------- " %process)
+            continue
         printd ("---------------------------\nchecking the service %s\n---------------------------- " %process)
         serviceName = process + ".service"
         processStatus, wasRestarted = checkProcessStatus(properties)
@@ -325,20 +328,20 @@ def main(checkType = "basic"):
     '''
     printd("monitoring started")
     configDict = getServicesConfig()
+    hc_data = getHealthChecksData()
 
     '''
     Step2: Monitor services and Raise Alerts
     '''
     monitResult = {}
     failingChecks = []
     if checkType == "basic":
-        monitResult, failingChecks = monitProcess(configDict)
+        included_services = hc_data["included_services"] if "included_services" in hc_data else []
+        monitResult, failingChecks = monitProcess(configDict, included_services)
 
     '''
     Step3: Run health check scripts as needed
     '''
-    hc_data = getHealthChecksData()
-
     if hc_data is not None and "health_checks_enabled" in hc_data and hc_data['health_checks_enabled']:
         hc_exclude = hc_data["excluded_health_checks"] if "excluded_health_checks" in hc_data else []
         for f in os.listdir(Config.HEALTH_CHECKS_DIR):