Merge pull request #5732 from Icinga/fix/flapping

Fix flapping calculation and events

fixes #5720

doc/12-icinga2-api.md

@@ -1306,15 +1306,18 @@ Example for all downtime events:
 
 #### <a id="icinga2-api-event-streams-type-flapping"></a> Event Stream Type: Flapping
 
-  Name 		| Type          | Description
+  Name              | Type          | Description
-  --------------|---------------|--------------------------
+  ------------------|---------------|--------------------------
-  type 	        | String        | Event type `Flapping`.
+  type              | String        | Event type `Flapping`.
-  timestamp     | Timestamp     | Unix timestamp when the event happened.
+  timestamp         | Timestamp     | Unix timestamp when the event happened.
-  host	        | String        | [Host](09-object-types.md#objecttype-host) name.
+  host              | String        | [Host](09-object-types.md#objecttype-host) name.
-  service       | String        | [Service](09-object-types.md#objecttype-service) name. Optional if this is a host flapping event.
+  service           | String        | [Service](09-object-types.md#objecttype-service) name. Optional if this is a host flapping event.
-  state	        | Number        | [Host](09-object-types.md#objecttype-host) or [service](09-object-types.md#objecttype-service) state.
+  state             | Number        | [Host](09-object-types.md#objecttype-host) or [service](09-object-types.md#objecttype-service) state.
-  state\_type   | Number        | [Host](09-object-types.md#objecttype-host) or [service](09-object-types.md#objecttype-service) state type.
+  state\_type       | Number        | [Host](09-object-types.md#objecttype-host) or [service](09-object-types.md#objecttype-service) state type.
-  is\_flapping  | Boolean       | Whether this object is flapping.
+  is\_flapping      | Boolean       | Whether this object is flapping.
+  current\_flapping | Number        | Current flapping value in percent (added in 2.8).
+  threshold\_low    | Number        | Low threshold in percent (added in 2.8).
+  threshold\_high   | Number        | High threshold in percent (added in 2.8).
 
 #### <a id="icinga2-api-event-streams-type-acknowledgementset"></a> Event Stream Type: AcknowledgementSet
 

lib/icinga/apievents.cpp

@@ -170,6 +170,9 @@ void ApiEvents::FlappingChangedHandler(const Checkable::Ptr& checkable, const Me
 	result->Set("state", service ? static_cast<int>(service->GetState()) : static_cast<int>(host->GetState()));
 	result->Set("state_type", checkable->GetStateType());
 	result->Set("is_flapping", checkable->IsFlapping());
+	result->Set("flapping_current", checkable->GetFlappingCurrent());
+	result->Set("threshold_low", checkable->GetFlappingThresholdLow());
+	result->Set("threshold_high", checkable->GetFlappingThresholdHigh());
 
 	for (const EventQueue::Ptr& queue : queues) {
 		queue->ProcessEvent(result);

lib/icinga/checkable-check.cpp

@@ -332,12 +332,15 @@ void Checkable::ProcessCheckResult(const CheckResult::Ptr& cr, const MessageOrig
 
 	olock.Unlock();
 
-//	Log(LogDebug, "Checkable")
+#ifdef I2_DEBUG /* I2_DEBUG */
-//	    << "Flapping: Checkable " << GetName()
+	Log(LogDebug, "Checkable")
-//	    << " was: " << (was_flapping)
+	    << "Flapping: Checkable " << GetName()
-//	    << " is: " << is_flapping)
+	    << " was: " << was_flapping
-//	    << " threshold: " << GetFlappingThreshold()
+	    << " is: " << is_flapping
-//	    << "% current: " + GetFlappingCurrent()) << "%.";
+	    << " threshold low: " << GetFlappingThresholdLow()
+	    << " threshold high: " << GetFlappingThresholdHigh()
+	    << "% current: " << GetFlappingCurrent() << "%.";
+#endif /* I2_DEBUG */
 
 	OnNewCheckResult(this, cr, origin);
 
@@ -371,7 +374,8 @@ void Checkable::ProcessCheckResult(const CheckResult::Ptr& cr, const MessageOrig
 			OnNotificationsRequested(this, NotificationFlappingStart, cr, "", "", MessageOrigin::Ptr());
 
 		Log(LogNotice, "Checkable")
-			<< "Flapping: Checkable '" << GetName() << "' started flapping (Current flapping value " << GetFlappingCurrent() << "% > threshold " << GetFlappingThresholdHigh() << "%).";
+		    << "Flapping Start: Checkable '" << GetName() << "' started flapping (Current flapping value "
+		    << GetFlappingCurrent() << "% > high threshold " << GetFlappingThresholdHigh() << "%).";
 
 		NotifyFlapping(origin);
 	} else if (!in_downtime && was_flapping && !is_flapping) {
@@ -380,7 +384,8 @@ void Checkable::ProcessCheckResult(const CheckResult::Ptr& cr, const MessageOrig
 			OnNotificationsRequested(this, NotificationFlappingEnd, cr, "", "", MessageOrigin::Ptr());
 
 		Log(LogNotice, "Checkable")
-			<< "Flapping: Checkable '" << GetName() << "' stopped flapping (Current flapping value " << GetFlappingCurrent() << "% < threshold " << GetFlappingThresholdLow() << "%).";
+		    << "Flapping Stop: Checkable '" << GetName() << "' stopped flapping (Current flapping value "
+		    << GetFlappingCurrent() << "% < low threshold " << GetFlappingThresholdLow() << "%).";
 
 		NotifyFlapping(origin);
 	}

lib/icinga/checkable-flapping.cpp

@@ -27,13 +27,14 @@ using namespace icinga;
 void Checkable::UpdateFlappingStatus(bool stateChange)
 {
 	std::bitset<20> stateChangeBuf = GetFlappingBuffer();
-	int oldestIndex = (GetFlappingBuffer() & 0xFF00000) >> 20;
+	int oldestIndex = GetFlappingIndex();
 
 	stateChangeBuf[oldestIndex] = stateChange;
 	oldestIndex = (oldestIndex + 1) % 20;
 
 	double stateChanges = 0;
 
+	/* Iterate over our state array and compute a weighted total */
 	for (int i = 0; i < 20; i++) {
 		if (stateChangeBuf[(oldestIndex + i) % 20])
 			stateChanges += 0.8 + (0.02 * i);
@@ -48,12 +49,13 @@ void Checkable::UpdateFlappingStatus(bool stateChange)
 	else
 		flapping = flappingValue > GetFlappingThresholdHigh();
 
+	SetFlappingBuffer(stateChangeBuf.to_ulong());
+	SetFlappingIndex(oldestIndex);
+	SetFlappingCurrent(flappingValue);
+	SetFlapping(flapping, true);
+
 	if (flapping != GetFlapping())
 		SetFlappingLastChange(Utility::GetTime());
-
-	SetFlappingBuffer((stateChangeBuf.to_ulong() | (oldestIndex << 20)));
-	SetFlappingCurrent(flappingValue);
-	SetFlapping(flapping);
 }
 
 bool Checkable::IsFlapping(void) const

lib/icinga/checkable.hpp

@@ -181,7 +181,6 @@ public:
 
 	/* Flapping Detection */
 	bool IsFlapping(void) const;
-	void UpdateFlappingStatus(bool stateChange);
 
 	/* Dependencies */
 	void AddDependency(const intrusive_ptr<Dependency>& dep);
@@ -237,6 +236,9 @@ private:
 	std::set<intrusive_ptr<Dependency> > m_ReverseDependencies;
 
 	void GetAllChildrenInternal(std::set<Checkable::Ptr>& children, int level = 0) const;
+
+	/* Flapping */
+	void UpdateFlappingStatus(bool stateChange);
 };
 
 }

lib/icinga/checkable.ti

@@ -158,7 +158,9 @@ abstract class Checkable : CustomVarObject
 		default {{{ return 0; }}}
 	};
 	[state] Timestamp flapping_last_change;
+
 	[state, no_user_view, no_user_modify] int flapping_buffer;
+	[state, no_user_view, no_user_modify] int flapping_index;
 	[state, protected] bool flapping;
 
 	[config, navigation] name(Endpoint) command_endpoint (CommandEndpointRaw) {