Merge pull request #8600 from Icinga/feature/flapping-ignore-unknown

Flapping: Allow to ignore states in flapping detection

doc/08-advanced-topics.md

@@ -481,6 +481,8 @@ when a [host](09-object-types.md#objecttype-host) or [service](09-object-types.m
 The default thresholds are 30% for high and 25% for low. If the computed flapping value exceeds the high threshold a
 host or service is considered flapping until it drops below the low flapping threshold.
 
+The attribute `flapping_ignore_states` allows to ignore state changes to specified states during the flapping calculation.
+
 `FlappingStart` and `FlappingEnd` notifications will be sent out accordingly, if configured. See the chapter on
 [notifications](alert-notifications) for details
 

doc/09-object-types.md

@@ -359,6 +359,7 @@ Configuration Attributes:
   event\_command            | Object name           | **Optional.** The name of an event command that should be executed every time the host's state changes or the host is in a `SOFT` state.
   flapping\_threshold\_high | Number                | **Optional.** Flapping upper bound in percent for a host to be considered flapping. Default `30.0`
   flapping\_threshold\_low  | Number                | **Optional.** Flapping lower bound in percent for a host to be considered  not flapping. Default `25.0`
+  flapping\_ignore\_states  | Array                 | **Optional.** A list of states that should be ignored during flapping calculation. By default no state is ignored.
   volatile                  | Boolean               | **Optional.** Treat all state changes as HARD changes. See [here](08-advanced-topics.md#volatile-services-hosts) for details. Defaults to `false`.
   zone                      | Object name           | **Optional.** The zone this object is a member of. Please read the [distributed monitoring](06-distributed-monitoring.md#distributed-monitoring) chapter for details.
   command\_endpoint         | Object name           | **Optional.** The endpoint where commands are executed on.
@@ -723,6 +724,7 @@ Configuration Attributes:
   enable\_flapping          | Boolean               | **Optional.** Whether flap detection is enabled. Defaults to `false`.
   flapping\_threshold\_high | Number                | **Optional.** Flapping upper bound in percent for a service to be considered flapping. `30.0`
   flapping\_threshold\_low  | Number                | **Optional.** Flapping lower bound in percent for a service to be considered  not flapping. `25.0`
+  flapping\_ignore\_states  | Array                 | **Optional.** A list of states that should be ignored during flapping calculation. By default no state is ignored.
   enable\_perfdata          | Boolean               | **Optional.** Whether performance data processing is enabled. Defaults to `true`.
   event\_command            | Object name           | **Optional.** The name of an event command that should be executed every time the service's state changes or the service is in a `SOFT` state.
   volatile                  | Boolean               | **Optional.** Treat all state changes as HARD changes. See [here](08-advanced-topics.md#volatile-services-hosts) for details. Defaults to `false`.

lib/icinga/checkable-check.cpp

@@ -383,7 +383,7 @@ void Checkable::ProcessCheckResult(const CheckResult::Ptr& cr, const MessageOrig
 
 	bool was_flapping = IsFlapping();
 
-	UpdateFlappingStatus(old_state != cr->GetState());
+	UpdateFlappingStatus(cr->GetState());
 
 	bool is_flapping = IsFlapping();
 

lib/icinga/checkable-flapping.cpp

@@ -36,11 +36,22 @@ private:
 	T m_Data{0};
 };
 
-void Checkable::UpdateFlappingStatus(bool stateChange)
+void Checkable::UpdateFlappingStatus(ServiceState newState)
 {
 	Bitset<unsigned long> stateChangeBuf = GetFlappingBuffer();
 	int oldestIndex = GetFlappingIndex();
 
+	ServiceState lastState = GetFlappingLastState();
+	bool stateChange = false;
+
+	int stateFilter = GetFlappingIgnoreStatesFilter();
+
+	/* Only count as state change if no state filter is set or the new state isn't filtered out */
+	if (stateFilter == -1 || !(ServiceStateToFlappingFilter(newState) & stateFilter)) {
+		stateChange = newState != lastState;
+		SetFlappingLastState(newState);
+	}
+
 	stateChangeBuf.Modify(oldestIndex, stateChange);
 	oldestIndex = (oldestIndex + 1) % 20;
 
@@ -85,3 +96,19 @@ bool Checkable::IsFlapping() const
 	else
 		return GetFlapping();
 }
+
+int Checkable::ServiceStateToFlappingFilter(ServiceState state)
+{
+	switch (state) {
+		case ServiceOK:
+			return StateFilterOK;
+		case ServiceWarning:
+			return StateFilterWarning;
+		case ServiceCritical:
+			return StateFilterCritical;
+		case ServiceUnknown:
+			return StateFilterUnknown;
+		default:
+			VERIFY(!"Invalid state type.");
+	}
+}

lib/icinga/checkable.cpp

@@ -15,6 +15,15 @@ using namespace icinga;
 REGISTER_TYPE_WITH_PROTOTYPE(Checkable, Checkable::GetPrototype());
 INITIALIZE_ONCE(&Checkable::StaticInitialize);
 
+const std::map<String, int> Checkable::m_FlappingStateFilterMap ({
+	{"OK", FlappingStateFilterOk},
+	{"Warning", FlappingStateFilterWarning},
+	{"Critical", FlappingStateFilterCritical},
+	{"Unknown", FlappingStateFilterUnknown},
+	{"Up", FlappingStateFilterOk},
+	{"Down", FlappingStateFilterCritical},
+});
+
 boost::signals2::signal<void (const Checkable::Ptr&, const String&, const String&, AcknowledgementType, bool, bool, double, double, const MessageOrigin::Ptr&)> Checkable::OnAcknowledgementSet;
 boost::signals2::signal<void (const Checkable::Ptr&, const String&, double, const MessageOrigin::Ptr&)> Checkable::OnAcknowledgementCleared;
 boost::signals2::signal<void (const Checkable::Ptr&, double)> Checkable::OnFlappingChange;
@@ -39,6 +48,13 @@ Checkable::Checkable()
 	SetSchedulingOffset(Utility::Random());
 }
 
+void Checkable::OnConfigLoaded()
+{
+	ObjectImpl<Checkable>::OnConfigLoaded();
+
+	SetFlappingIgnoreStatesFilter(FilterArrayToInt(GetFlappingIgnoreStates(), m_FlappingStateFilterMap, ~0));
+}
+
 void Checkable::OnAllConfigLoaded()
 {
 	ObjectImpl<Checkable>::OnAllConfigLoaded();

lib/icinga/checkable.hpp

@@ -41,6 +41,17 @@ enum CheckableType
 	CheckableService
 };
 
+/**
+ * @ingroup icinga
+ */
+enum FlappingStateFilter
+{
+	FlappingStateFilterOk = 1,
+	FlappingStateFilterWarning = 2,
+	FlappingStateFilterCritical = 4,
+	FlappingStateFilterUnknown = 8,
+};
+
 class CheckCommand;
 class EventCommand;
 class Dependency;
@@ -186,6 +197,7 @@ public:
 
 protected:
 	void Start(bool runtimeCreated) override;
+	void OnConfigLoaded() override;
 	void OnAllConfigLoaded() override;
 
 private:
@@ -226,7 +238,10 @@ private:
 	void GetAllChildrenInternal(std::set<Checkable::Ptr>& children, int level = 0) const;
 
 	/* Flapping */
-	void UpdateFlappingStatus(bool stateChange);
+	static const std::map<String, int> m_FlappingStateFilterMap;
+
+	void UpdateFlappingStatus(ServiceState newState);
+	static int ServiceStateToFlappingFilter(ServiceState state);
 };
 
 }

lib/icinga/checkable.ti

@@ -73,6 +73,9 @@ abstract class Checkable : CustomVarObject
 		default {{{ return true; }}}
 	};
 
+	[config] array(String) flapping_ignore_states;
+	[no_user_view, no_user_modify] int flapping_ignore_states_filter_real (FlappingIgnoreStatesFilter);
+
 	[config, deprecated] double flapping_threshold;
 
 	[config] double flapping_threshold_low {
@@ -163,6 +166,9 @@ abstract class Checkable : CustomVarObject
 	};
 	[state] Timestamp flapping_last_change;
 
+	[state, enum, no_user_view, no_user_modify] ServiceState flapping_last_state {
+		default {{{ return ServiceUnknown; }}}
+	};
 	[state, no_user_view, no_user_modify] int flapping_buffer;
 	[state, no_user_view, no_user_modify] int flapping_index;
 	[state, protected] bool flapping;