Implement support for limiting the number of concurrent checks

fixes #8137

diff --git a/doc/6-object-types.md b/doc/6-object-types.md
index 54fb0e0e9a7ff1f7bf5ccf434965c2be82cc9d90..1518a62de285da1060bab36cc902e0187ea7e889 100644 (file)
--- a/doc/6-object-types.md
+++ b/doc/6-object-types.md
@@ -189,7 +189,7 @@ Argument array `repeat_key = false`:
 
 ## <a id="objecttype-checkcomponent"></a> CheckerComponent
 
-The checker component is responsible for scheduling active checks. There are no configurable options.
+The checker component is responsible for scheduling active checks.
 
 Example:
 
@@ -197,6 +197,12 @@ Example:
 
     object CheckerComponent "checker" { }
 
+Configuration Attributes:
+
+  Name                |Description
+  --------------------|----------------
+  concurrent\_checks  |**Optional.** The maximum number of concurrent checks. Defaults to 512.
+
 ## <a id="objecttype-checkresultreader"></a> CheckResultReader
 
 Reads Icinga 1.x check results from a directory. This functionality is provided

diff --git a/lib/checker/checkercomponent.cpp b/lib/checker/checkercomponent.cpp
index 64c37c7e9c1e08e867e9958aa0d7c53dc3a2673a..6b20cc471dac7c73992cb78064cc9f2e969b37a2 100644 (file)
--- a/lib/checker/checkercomponent.cpp
+++ b/lib/checker/checkercomponent.cpp
@@ -69,6 +69,7 @@ void CheckerComponent::OnConfigLoaded(void)
        ConfigObject::OnActiveChanged.connect(bind(&CheckerComponent::ObjectHandler, this, _1));
        ConfigObject::OnPausedChanged.connect(bind(&CheckerComponent::ObjectHandler, this, _1));
 
+       Checkable::OnNewCheckResult.connect(bind(&CheckerComponent::CheckResultHandler, this, _1));
        Checkable::OnNextCheckChanged.connect(bind(&CheckerComponent::NextCheckChangedHandler, this, _1));
 }
 
@@ -121,7 +122,7 @@ void CheckerComponent::CheckThreadProc(void)
 
                double wait = checkable->GetNextCheck() - Utility::GetTime();
 
-               if (wait > 0) {
+               if (wait > 0 || m_PendingCheckables.size() >= GetConcurrentChecks()) {
                        /* Wait for the next check. */
                        m_CV.timed_wait(lock, boost::posix_time::milliseconds(wait * 1000));
 
@@ -215,27 +216,6 @@ void CheckerComponent::ExecuteCheckHelper(const Checkable::Ptr& checkable)
 
                Log(LogCritical, "checker", output);
        }
-
-       {
-               boost::mutex::scoped_lock lock(m_Mutex);
-
-               /* remove the object from the list of pending objects; if it's not in the
-                * list this was a manual (i.e. forced) check and we must not re-add the
-                * object to the list because it's already there. */
-               CheckerComponent::CheckableSet::iterator it;
-               it = m_PendingCheckables.find(checkable);
-               if (it != m_PendingCheckables.end()) {
-                       m_PendingCheckables.erase(it);
-
-                       if (checkable->IsActive())
-                               m_IdleCheckables.insert(checkable);
-
-                       m_CV.notify_all();
-               }
-       }
-
-       Log(LogDebug, "CheckerComponent")
-           << "Check finished for object '" << checkable->GetName() << "'";
 }
 
 void CheckerComponent::ResultTimerHandler(void)
@@ -279,6 +259,30 @@ void CheckerComponent::ObjectHandler(const ConfigObject::Ptr& object)
        }
 }
 
+void CheckerComponent::CheckResultHandler(const Checkable::Ptr& checkable)
+{
+       {
+               boost::mutex::scoped_lock lock(m_Mutex);
+
+               /* remove the object from the list of pending objects; if it's not in the
+                * list this was a manual (i.e. forced) check and we must not re-add the
+                * object to the list because it's already there. */
+               CheckerComponent::CheckableSet::iterator it;
+               it = m_PendingCheckables.find(checkable);
+               if (it != m_PendingCheckables.end()) {
+                       m_PendingCheckables.erase(it);
+
+                       if (checkable->IsActive())
+                               m_IdleCheckables.insert(checkable);
+
+                       m_CV.notify_all();
+               }
+       }
+
+       Log(LogDebug, "CheckerComponent")
+           << "Check finished for object '" << checkable->GetName() << "'";
+}
+
 void CheckerComponent::NextCheckChangedHandler(const Checkable::Ptr& checkable)
 {
        boost::mutex::scoped_lock lock(m_Mutex);

diff --git a/lib/checker/checkercomponent.hpp b/lib/checker/checkercomponent.hpp
index 9896685233a8d679526e7286216ae8c621e622f2..f132fe16cda34f6ab769fc5a93f3928bfcba8823 100644 (file)
--- a/lib/checker/checkercomponent.hpp
+++ b/lib/checker/checkercomponent.hpp
@@ -97,6 +97,7 @@ private:
        void AdjustCheckTimer(void);
 
        void ObjectHandler(const ConfigObject::Ptr& object);
+       void CheckResultHandler(const Checkable::Ptr& checkable);
        void NextCheckChangedHandler(const Checkable::Ptr& checkable);
 
        void RescheduleCheckTimer(void);

diff --git a/lib/checker/checkercomponent.ti b/lib/checker/checkercomponent.ti
index c35993493fbfb4c7ce0f8afa22cb458917abb448..6746b418d2648e459e18deca370c6cc36ffbc3f2 100644 (file)
--- a/lib/checker/checkercomponent.ti
+++ b/lib/checker/checkercomponent.ti
@@ -26,6 +26,11 @@ namespace icinga
 
 class CheckerComponent : ConfigObject
 {
+       [config] int concurrent_checks {
+               default {{{
+                       return 512;
+               }}}
+       };
 };
 
 }

diff --git a/lib/icinga/checkable-check.cpp b/lib/icinga/checkable-check.cpp
index 1d5153d57f5930a03a3fea9f71984ed14af8482e..6e69a8ffa121c7f65cc8362e5794f986c2de07fa 100644 (file)
--- a/lib/icinga/checkable-check.cpp
+++ b/lib/icinga/checkable-check.cpp
@@ -406,6 +406,10 @@ void Checkable::ExecuteCheck(void)
 {
        CONTEXT("Executing check for object '" + GetName() + "'");
 
+       /* keep track of scheduling info in case the check type doesn't provide its own information */
+       double scheduled_start = GetNextCheck();
+       double before_check = Utility::GetTime();
+
        UpdateNextCheck();
 
        bool reachable = IsReachable();
@@ -424,10 +428,6 @@ void Checkable::ExecuteCheck(void)
                SetLastReachable(reachable);
        }
 
-       /* keep track of scheduling info in case the check type doesn't provide its own information */
-       double scheduled_start = GetNextCheck();
-       double before_check = Utility::GetTime();
-
        CheckResult::Ptr cr = new CheckResult();
 
        cr->SetScheduleStart(scheduled_start);