Merge pull request #242 from atc0005/i220-add-support-for-filtering-b…

…y-entity-name

Add support for filtering via entity name

cmd/check_vmware_alarms/main.go

@@ -189,6 +189,8 @@ func main() {
 	triggeredAlarmFilters := vsphere.TriggeredAlarmFilters{
 		IncludedAlarmEntityTypes:   cfg.IncludedAlarmEntityTypes,
 		ExcludedAlarmEntityTypes:   cfg.ExcludedAlarmEntityTypes,
+		IncludedAlarmEntityNames:   cfg.IncludedAlarmEntityNames,
+		ExcludedAlarmEntityNames:   cfg.ExcludedAlarmEntityNames,
 		IncludedAlarmNames:         cfg.IncludedAlarmNames,
 		ExcludedAlarmNames:         cfg.ExcludedAlarmNames,
 		IncludedAlarmDescriptions:  cfg.IncludedAlarmDescriptions,

contrib/nagios/etc/nagios-plugins/config/vmware-alarms.cfg

@@ -25,15 +25,15 @@ define command{
 # alarms which have been previously acknowledged.
 define command{
     command_name    check_vmware_alarms_eval_acknowledged
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-type '$ARG4$' --eval-acknowledged --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-entity-type '$ARG4$' --eval-acknowledged --trust-cert --log-level info
     }
 
 # Look at triggered alarms for specified managed object types (e.g., Datastore
 # and VirtualMachine), across all detected datacenters, do not evaluate any
 # triggered alarms which have been previously acknowledged.
 define command{
     command_name    check_vmware_alarms_include_types
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-type '$ARG4$' --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-entity-type '$ARG4$' --trust-cert --log-level info
     }
 
 # Look at triggered alarms for specified managed object types (e.g., Datastore
@@ -42,7 +42,7 @@ define command{
 # resolved.
 define command{
     command_name    check_vmware_alarms_include_types_eval_acknowledged
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-type '$ARG4$' --eval-acknowledged --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-entity-type '$ARG4$' --eval-acknowledged --trust-cert --log-level info
     }
 
 # Look at triggered alarms associated with managed objects which do not match
@@ -51,7 +51,7 @@ define command{
 # triggered alarms which have been previously acknowledged.
 define command{
     command_name    check_vmware_alarms_exclude_types
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-type '$ARG4$' --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-entity-type '$ARG4$' --trust-cert --log-level info
     }
 
 # Look at triggered alarms associated with managed objects which do not match
@@ -61,7 +61,7 @@ define command{
 # resolved.
 define command{
     command_name    check_vmware_alarms_exclude_types_eval_acknowledged
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-type '$ARG4$' --eval-acknowledged --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-entity-type '$ARG4$' --eval-acknowledged --trust-cert --log-level info
     }
 
 # Look at triggered alarms whose name matches the specified list of alarm name
@@ -162,15 +162,15 @@ define command{
 # triggered alarms which have been previously acknowledged.
 define command{
     command_name    check_vmware_alarms_specific_dc_include_types
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-type '$ARG4$' --dc-name '$ARG5$' --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-entity-type '$ARG4$' --dc-name '$ARG5$' --trust-cert --log-level info
     }
 
 # Look at triggered alarms for specified managed object types (e.g., Datastore
 # and VirtualMachine), within specified datacenters. Evaluate any triggered
 # alarms which have been previously acknowledged, but not yet resolved.
 define command{
     command_name    check_vmware_alarms_specific_dc_include_types_eval_acknowledged
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-type '$ARG4$' --dc-name '$ARG5$' --eval-acknowledged --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --include-entity-type '$ARG4$' --dc-name '$ARG5$' --eval-acknowledged --trust-cert --log-level info
     }
 
 # Look at triggered alarms associated with managed objects which do not match
@@ -179,7 +179,7 @@ define command{
 # alarms which have been previously acknowledged.
 define command{
     command_name    check_vmware_alarms_specific_dc_exclude_types
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-type '$ARG4$' --dc-name '$ARG5$' --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-entity-type '$ARG4$' --dc-name '$ARG5$' --trust-cert --log-level info
     }
 
 # Look at triggered alarms associated with managed objects which do not match
@@ -188,7 +188,7 @@ define command{
 # which have been previously acknowledged, but not yet resolved.
 define command{
     command_name    check_vmware_alarms_specific_dc_exclude_types_eval_acknowledged
-    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-type '$ARG4$' --dc-name '$ARG5$' --eval-acknowledged --trust-cert --log-level info
+    command_line    /usr/lib/nagios/plugins/check_vmware_alarms --server '$HOSTNAME$' --domain '$ARG1$' --username '$ARG2$' --password '$ARG3$' --exclude-entity-type '$ARG4$' --dc-name '$ARG5$' --eval-acknowledged --trust-cert --log-level info
     }
 
 # Look at triggered alarms whose name matches the specified list of alarm name

internal/config/config.go

@@ -207,6 +207,19 @@ type Config struct {
 	// inclusions.
 	ExcludedAlarmEntityTypes multiValueStringFlag
 
+	// IncludedAlarmEntityNames is a list of entity names for Alarms that will
+	// be explicitly included for evaluation. Unless included by later
+	// filtering logic, unmatched Triggered Alarms will be excluded from final
+	// evaluation. Explicitly included Triggered Alarms are still subject to
+	// permanent exclusion if a an explicit exclusion match is made.
+	IncludedAlarmEntityNames multiValueStringFlag
+
+	// ExcludedAlarmEntityTypes is a list of entity names for Alarms that will
+	// be explicitly excluded from further evaluation by other stages in the
+	// filtering pipeline. Explicit exclusions have precedence over explicit
+	// inclusions.
+	ExcludedAlarmEntityNames multiValueStringFlag
+
 	// IncludedAlarmNames is a list of names for defined Alarms that will be
 	// explicitly included for evaluation. Unless included by later filtering
 	// logic, unmatched Triggered Alarms will be excluded from final

internal/config/constants.go

@@ -64,6 +64,8 @@ const (
 	virtualHardwareDefaultIsMinimumFlagHelp         string = "If specified, the host or cluster default virtual hardware version is the minimum hardware version allowed. Any Virtual Machine not meeting this minimum value is considered to be in a WARNING state."
 	includedAlarmEntityTypesFlagHelp                string = "If specified, triggered alarms will only be evaluated if the associated entity type (e.g., Datastore) matches one of the provided values."
 	excludedAlarmEntityTypesFlagHelp                string = "If specified, triggered alarms will only be evaluated if the associated entity type (e.g., Datastore) does NOT match one of the provided values."
+	includedAlarmEntityNamesFlagHelp                string = "If specified, triggered alarms will only be evaluated if the associated entity name (e.g., \"node1.example.com\") matches one of the provided values."
+	excludedAlarmEntityNamesFlagHelp                string = "If specified, triggered alarms will only be evaluated if the associated entity name (e.g., \"node1.example.com\") does NOT match one of the provided values."
 	evaluateAcknowledgedTriggeredAlarmFlagHelp      string = "Toggles evaluation of acknowledged triggered alarms in addition to unacknowledged triggered alarms. Evaluation of acknowledged alarms is disabled by default."
 	includedAlarmNamesFlagHelp                      string = "If specified, triggered alarms will only be evaluated if the alarm name (e.g., \"Datastore usage on disk\") case-insensitively matches one of the provided substring values (e.g., \"datastore\" or \"datastore usage\") and is not explicitly excluded by another filter in the pipeline; while multiple explicit inclusions are allowed, explicit exclusions have precedence over explicit inclusions and will exclude the triggered alarm from further evaluation."
 	excludedAlarmNamesFlagHelp                      string = "If specified, triggered alarms will only be evaluated if the alarm name (e.g., \"Datastore usage on disk\") DOES NOT case-insensitively match one of the provided substring values (e.g., \"datastore\" or \"datastore usage\") and is not explicitly excluded by another filter in the pipeline; while multiple explicit inclusions are allowed, explicit exclusions have precedence over explicit inclusions and will exclude the triggered alarm from further evaluation."

internal/config/flags.go

@@ -126,8 +126,8 @@ func (c *Config) handleFlagsConfig(pluginType PluginType) {
 	case pluginType.Alarms:
 
 		flag.Var(&c.DatacenterNames, "dc-name", datacenterNamesFlagHelp)
-		flag.Var(&c.IncludedAlarmEntityTypes, "include-type", includedAlarmEntityTypesFlagHelp)
-		flag.Var(&c.ExcludedAlarmEntityTypes, "exclude-type", excludedAlarmEntityTypesFlagHelp)
+		flag.Var(&c.IncludedAlarmEntityTypes, "include-entity-type", includedAlarmEntityTypesFlagHelp)
+		flag.Var(&c.ExcludedAlarmEntityTypes, "exclude-entity-type", excludedAlarmEntityTypesFlagHelp)
 
 		flag.BoolVar(&c.EvaluateAcknowledgedAlarms, "eval-acknowledged", defaultEvaluateAcknowledgedAlarms, evaluateAcknowledgedTriggeredAlarmFlagHelp)
 
@@ -140,6 +140,9 @@ func (c *Config) handleFlagsConfig(pluginType PluginType) {
 		flag.Var(&c.includedAlarmStatuses, "include-status", includedAlarmStatusesFlagHelp)
 		flag.Var(&c.excludedAlarmStatuses, "exclude-status", excludedAlarmStatusesFlagHelp)
 
+		flag.Var(&c.IncludedAlarmEntityNames, "include-entity-name", includedAlarmEntityNamesFlagHelp)
+		flag.Var(&c.ExcludedAlarmEntityNames, "exclude-entity-name", excludedAlarmEntityNamesFlagHelp)
+
 	case pluginType.DatastoresSize:
 
 		flag.StringVar(&c.DatacenterName, "dc-name", defaultDatacenterName, datacenterNameFlagHelp)

internal/config/validate.go

@@ -428,8 +428,17 @@ func (c Config) validate(pluginType PluginType) error {
 		if len(c.IncludedAlarmEntityTypes) > 0 && len(c.ExcludedAlarmEntityTypes) > 0 {
 			return fmt.Errorf(
 				"only one of %q or %q flags may be specified",
-				"include-type",
-				"exclude-type",
+				"include-entity-type",
+				"exclude-entity-type",
+			)
+		}
+
+		// only one of these options may be used
+		if len(c.IncludedAlarmEntityNames) > 0 && len(c.ExcludedAlarmEntityNames) > 0 {
+			return fmt.Errorf(
+				"only one of %q or %q flags may be specified",
+				"include-entity-name",
+				"exclude-entity-name",
 			)
 		}
 

internal/vsphere/alarms.go

@@ -117,6 +117,8 @@ type TriggeredAlarms []TriggeredAlarm
 type TriggeredAlarmFilters struct {
 	IncludedAlarmEntityTypes   []string
 	ExcludedAlarmEntityTypes   []string
+	IncludedAlarmEntityNames   []string
+	ExcludedAlarmEntityNames   []string
 	IncludedAlarmNames         []string
 	ExcludedAlarmNames         []string
 	IncludedAlarmDescriptions  []string
@@ -529,7 +531,7 @@ func logTriggeredAlarmMarked(triggeredAlarm TriggeredAlarm, keep bool, explicit
 	switch {
 	case keep:
 		logger.Printf(
-			"Alarm (%s) for %q of type %q with name %q %s marked for inclusion",
+			"Alarm (%s) for entity name %q of type %q with alarm name %q %s marked for inclusion",
 			triggeredAlarm.OverallStatus,
 			triggeredAlarm.Entity.Name,
 			triggeredAlarm.Entity.MOID.Type,
@@ -539,7 +541,7 @@ func logTriggeredAlarmMarked(triggeredAlarm TriggeredAlarm, keep bool, explicit
 
 	default:
 		logger.Printf(
-			"Alarm (%s) for %q of type %q with name %q %s marked for exclusion",
+			"Alarm (%s) for entity name %q of type %q with alarm name %q %s marked for exclusion",
 			triggeredAlarm.OverallStatus,
 			triggeredAlarm.Entity.Name,
 			triggeredAlarm.Entity.MOID.Type,
@@ -678,6 +680,16 @@ func EntityStatusToNagiosState(entityStatus types.ManagedEntityStatus) (string,
 
 }
 
+// getSubstringFilterKeywords is a helper function that returns a map of all
+// valid keywords used by the TriggeredAlarms.filterByString method.
+// func getSubstringFilterKeywords() map[string]struct{} {
+// 	return map[string]struct{}{
+// 		AlarmDescription: struct{}{},
+// 		AlarmName:        struct{}{},
+// 		EntityName:       struct{}{},
+// 	}
+// }
+
 // Filter explicitly includes or excludes TriggeredAlarms based on specified
 // filter settings.
 func (tas *TriggeredAlarms) Filter(filters TriggeredAlarmFilters) {
@@ -689,14 +701,17 @@ func (tas *TriggeredAlarms) Filter(filters TriggeredAlarmFilters) {
 	tas.filterByEntityType(filters.IncludedAlarmEntityTypes, filters.ExcludedAlarmEntityTypes)
 
 	logger.Println("Filtering triggered alarms by name")
-	tas.filterBySubstring(false, filters.IncludedAlarmNames, filters.ExcludedAlarmNames)
+	tas.filterBySubstring(alarmName, filters.IncludedAlarmNames, filters.ExcludedAlarmNames)
 
 	logger.Println("Filtering triggered alarms by description")
-	tas.filterBySubstring(true, filters.IncludedAlarmDescriptions, filters.ExcludedAlarmDescriptions)
+	tas.filterBySubstring(alarmDescription, filters.IncludedAlarmDescriptions, filters.ExcludedAlarmDescriptions)
 
 	logger.Println("Filtering triggered alarms by status")
 	tas.filterByStatus(filters.IncludedAlarmStatuses, filters.ExcludedAlarmStatuses)
 
+	logger.Println("Filtering triggered alarms by entity name")
+	tas.filterBySubstring(entityName, filters.IncludedAlarmEntityNames, filters.ExcludedAlarmEntityNames)
+
 }
 
 // FilterByIncludedEntityType accepts a slice of entity type keywords to use
@@ -767,11 +782,13 @@ func (tas *TriggeredAlarms) filterByEntityType(include []string, exclude []strin
 	switch {
 	// if the collection of TriggeredAlarms is empty, skip filtering attempts.
 	case len(*tas) == 0:
+		logger.Println("Triggered Alarms list is empty, aborting")
 		return
 
 	// if we're not limiting TriggeredAlarms by entity type, skip filtering
 	// attempts.
 	case len(include) == 0 && len(exclude) == 0:
+		logger.Println("Triggered Alarms status inclusion and exclusion lists are empty, aborting")
 		return
 	}
 
@@ -902,7 +919,7 @@ func (tas *TriggeredAlarms) FilterByIncludedNameSubstring(include []string) {
 		)
 	}()
 
-	tas.filterBySubstring(false, include, []string{})
+	tas.filterBySubstring(alarmName, include, []string{})
 
 }
 
@@ -923,7 +940,7 @@ func (tas *TriggeredAlarms) FilterByExcludedNameSubstring(exclude []string) {
 		)
 	}()
 
-	tas.filterBySubstring(false, []string{}, exclude)
+	tas.filterBySubstring(alarmName, []string{}, exclude)
 
 }
 
@@ -946,7 +963,7 @@ func (tas *TriggeredAlarms) FilterByIncludedDescriptionSubstring(include []strin
 		)
 	}()
 
-	tas.filterBySubstring(true, include, []string{})
+	tas.filterBySubstring(alarmDescription, include, []string{})
 
 }
 
@@ -967,19 +984,20 @@ func (tas *TriggeredAlarms) FilterByExcludedDescriptionSubstring(exclude []strin
 		)
 	}()
 
-	tas.filterBySubstring(true, []string{}, exclude)
+	tas.filterBySubstring(alarmDescription, []string{}, exclude)
 
 }
 
-// filterBySubstring accepts slices of substrings to use in comparisons
-// against TriggeredAlarm names or descriptions in order to explicitly mark
-// TriggeredAlarms for inclusion or exclusion in the final evaluation. A
-// boolean value indicates whether the comparison should be against the
-// defined Alarm's description or name field.
+// filterBySubstring accepts a field keyword and slices of substrings to use
+// in comparisons against TriggeredAlarm fields in order to explicitly mark
+// TriggeredAlarms for inclusion or exclusion in the final evaluation. The
+// provided field keyword indicates which field the comparison should be
+// against. If an invlaid field keyword is supplied the field comparison will
+// default to using the alarm name.
 //
 // Flag evaluation logic prevents sysadmins from providing both an inclusion
 // and exclusion list.
-func (tas *TriggeredAlarms) filterBySubstring(useDescription bool, include []string, exclude []string) {
+func (tas *TriggeredAlarms) filterBySubstring(fieldKeyword string, include []string, exclude []string) {
 
 	funcTimeStart := time.Now()
 
@@ -988,23 +1006,26 @@ func (tas *TriggeredAlarms) filterBySubstring(useDescription bool, include []str
 	// been applied in order to show the results of this filter.
 	nonExcludedStart := len(*tas) - tas.NumExcluded()
 
-	defer func(start *int) {
+	defer func(start *int, keyword string) {
 		logger.Printf(
-			"It took %v to execute filterBySubstring func (for %d non-excluded TriggeredAlarms, yielding %d non-excluded TriggeredAlarms)\n",
+			"It took %v to execute filterBySubstring func (for %d non-excluded TriggeredAlarms, using keyword %s, yielding %d non-excluded TriggeredAlarms)\n",
 			time.Since(funcTimeStart),
 			*start,
+			keyword,
 			len(*tas)-tas.NumExcluded(),
 		)
-	}(&nonExcludedStart)
+	}(&nonExcludedStart, fieldKeyword)
 
 	switch {
 	// if the collection of TriggeredAlarms is empty, skip filtering attempts.
 	case len(*tas) == 0:
+		logger.Println("Triggered Alarms list is empty, aborting")
 		return
 
 	// if we're not limiting TriggeredAlarms by entity type, skip filtering
 	// attempts.
 	case len(include) == 0 && len(exclude) == 0:
+		logger.Println("Triggered Alarms status inclusion and exclusion lists are empty, aborting")
 		return
 	}
 
@@ -1022,13 +1043,27 @@ func (tas *TriggeredAlarms) filterBySubstring(useDescription bool, include []str
 		)
 	}
 
+	// validKeywords := getSubstringFilterKeywords()
+	// if _, ok := validKeywords[fieldKeyword]; !ok {
+	// 	logger.Printf("")
+	// }
+
+	logger.Printf("substring field keyword %q specified", fieldKeyword)
 	for i := range *tas {
 
 		var substrField string
-		switch {
-		case useDescription:
+		switch fieldKeyword {
+		case alarmDescription:
 			substrField = (*tas)[i].Description
+		case alarmName:
+			substrField = (*tas)[i].Name
+		case entityName:
+			substrField = (*tas)[i].Entity.Name
 		default:
+			logger.Printf(
+				"substring field %q not recognized, defaulting to alarm name",
+				fieldKeyword,
+			)
 			substrField = (*tas)[i].Name
 		}
 
@@ -1440,6 +1475,22 @@ func AlarmsReport(
 		nagios.CheckOutputEOL,
 	)
 
+	fmt.Fprintf(
+		&report,
+		"* Specified Triggered Alarm entity names to explicitly include (%d): [%v]%s",
+		len(triggeredAlarmFilters.IncludedAlarmEntityNames),
+		strings.Join(triggeredAlarmFilters.IncludedAlarmEntityNames, ", "),
+		nagios.CheckOutputEOL,
+	)
+
+	fmt.Fprintf(
+		&report,
+		"* Specified Triggered Alarm entity names to explicitly exclude (%d): [%v]%s",
+		len(triggeredAlarmFilters.ExcludedAlarmEntityNames),
+		strings.Join(triggeredAlarmFilters.ExcludedAlarmEntityNames, ", "),
+		nagios.CheckOutputEOL,
+	)
+
 	fmt.Fprintf(
 		&report,
 		"* Specified Triggered Alarm names to explicitly include (%d): [%v]%s",