Fix incorrect re-scheduling behavior for command_endpoint checks

refs #8137
This commit is contained in:
Gunnar Beutner 2016-05-12 13:46:22 +02:00
parent 61c94cb98f
commit 97a5091abc
5 changed files with 54 additions and 27 deletions

View File

@ -69,7 +69,6 @@ void CheckerComponent::OnConfigLoaded(void)
ConfigObject::OnActiveChanged.connect(bind(&CheckerComponent::ObjectHandler, this, _1)); ConfigObject::OnActiveChanged.connect(bind(&CheckerComponent::ObjectHandler, this, _1));
ConfigObject::OnPausedChanged.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)); Checkable::OnNextCheckChanged.connect(bind(&CheckerComponent::NextCheckChangedHandler, this, _1));
} }
@ -122,7 +121,7 @@ void CheckerComponent::CheckThreadProc(void)
double wait = checkable->GetNextCheck() - Utility::GetTime(); double wait = checkable->GetNextCheck() - Utility::GetTime();
if (wait > 0 || m_PendingCheckables.size() >= GetConcurrentChecks()) { if (wait > 0 || Checkable::GetPendingChecks() >= GetConcurrentChecks()) {
/* Wait for the next check. */ /* Wait for the next check. */
m_CV.timed_wait(lock, boost::posix_time::milliseconds(wait * 1000)); m_CV.timed_wait(lock, boost::posix_time::milliseconds(wait * 1000));
@ -216,6 +215,27 @@ void CheckerComponent::ExecuteCheckHelper(const Checkable::Ptr& checkable)
Log(LogCritical, "checker", output); 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) void CheckerComponent::ResultTimerHandler(void)
@ -259,30 +279,6 @@ 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) void CheckerComponent::NextCheckChangedHandler(const Checkable::Ptr& checkable)
{ {
boost::mutex::scoped_lock lock(m_Mutex); boost::mutex::scoped_lock lock(m_Mutex);

View File

@ -97,7 +97,6 @@ private:
void AdjustCheckTimer(void); void AdjustCheckTimer(void);
void ObjectHandler(const ConfigObject::Ptr& object); void ObjectHandler(const ConfigObject::Ptr& object);
void CheckResultHandler(const Checkable::Ptr& checkable);
void NextCheckChangedHandler(const Checkable::Ptr& checkable); void NextCheckChangedHandler(const Checkable::Ptr& checkable);
void RescheduleCheckTimer(void); void RescheduleCheckTimer(void);

View File

@ -41,6 +41,9 @@ boost::signals2::signal<void (const Checkable::Ptr&, const CheckResult::Ptr&, st
boost::signals2::signal<void (const Checkable::Ptr&, NotificationType, const CheckResult::Ptr&, const String&, const String&)> Checkable::OnNotificationsRequested; boost::signals2::signal<void (const Checkable::Ptr&, NotificationType, const CheckResult::Ptr&, const String&, const String&)> Checkable::OnNotificationsRequested;
boost::signals2::signal<void (const Checkable::Ptr&)> Checkable::OnNextCheckUpdated; boost::signals2::signal<void (const Checkable::Ptr&)> Checkable::OnNextCheckUpdated;
boost::mutex Checkable::m_StatsMutex;
int Checkable::m_PendingChecks = 0;
CheckCommand::Ptr Checkable::GetCheckCommand(void) const CheckCommand::Ptr Checkable::GetCheckCommand(void) const
{ {
return dynamic_pointer_cast<CheckCommand>(NavigateCheckCommandRaw()); return dynamic_pointer_cast<CheckCommand>(NavigateCheckCommandRaw());
@ -515,3 +518,21 @@ void Checkable::UpdateStatistics(const CheckResult::Ptr& cr, CheckableType type)
Log(LogWarning, "Checkable", "Unknown checkable type for statistic update."); Log(LogWarning, "Checkable", "Unknown checkable type for statistic update.");
} }
} }
void Checkable::IncreasePendingChecks(void)
{
boost::mutex::scoped_lock lock(m_StatsMutex);
m_PendingChecks++;
}
void Checkable::DecreasePendingChecks(void)
{
boost::mutex::scoped_lock lock(m_StatsMutex);
m_PendingChecks--;
}
int Checkable::GetPendingChecks(void)
{
boost::mutex::scoped_lock lock(m_StatsMutex);
return m_PendingChecks;
}

View File

@ -178,6 +178,10 @@ public:
virtual void ValidateCheckInterval(double value, const ValidationUtils& utils) override; virtual void ValidateCheckInterval(double value, const ValidationUtils& utils) override;
static void IncreasePendingChecks(void);
static void DecreasePendingChecks(void);
static int GetPendingChecks(void);
protected: protected:
virtual void Start(bool runtimeCreated) override; virtual void Start(bool runtimeCreated) override;
@ -186,6 +190,9 @@ private:
bool m_CheckRunning; bool m_CheckRunning;
long m_SchedulingOffset; long m_SchedulingOffset;
static boost::mutex m_StatsMutex;
static int m_PendingChecks;
/* Downtimes */ /* Downtimes */
std::set<Downtime::Ptr> m_Downtimes; std::set<Downtime::Ptr> m_Downtimes;
mutable boost::mutex m_DowntimeMutex; mutable boost::mutex m_DowntimeMutex;

View File

@ -52,6 +52,8 @@ void PluginCheckTask::ScriptFunc(const Checkable::Ptr& checkable, const CheckRes
resolvers.push_back(std::make_pair("command", commandObj)); resolvers.push_back(std::make_pair("command", commandObj));
resolvers.push_back(std::make_pair("icinga", IcingaApplication::GetInstance())); resolvers.push_back(std::make_pair("icinga", IcingaApplication::GetInstance()));
Checkable::IncreasePendingChecks();
PluginUtility::ExecuteCommand(commandObj, checkable, checkable->GetLastCheckResult(), PluginUtility::ExecuteCommand(commandObj, checkable, checkable->GetLastCheckResult(),
resolvers, resolvedMacros, useResolvedMacros, resolvers, resolvedMacros, useResolvedMacros,
boost::bind(&PluginCheckTask::ProcessFinishedHandler, checkable, cr, _1, _2)); boost::bind(&PluginCheckTask::ProcessFinishedHandler, checkable, cr, _1, _2));
@ -59,6 +61,8 @@ void PluginCheckTask::ScriptFunc(const Checkable::Ptr& checkable, const CheckRes
void PluginCheckTask::ProcessFinishedHandler(const Checkable::Ptr& checkable, const CheckResult::Ptr& cr, const Value& commandLine, const ProcessResult& pr) void PluginCheckTask::ProcessFinishedHandler(const Checkable::Ptr& checkable, const CheckResult::Ptr& cr, const Value& commandLine, const ProcessResult& pr)
{ {
Checkable::DecreasePendingChecks();
if (pr.ExitStatus > 3) { if (pr.ExitStatus > 3) {
Process::Arguments parguments = Process::PrepareCommand(commandLine); Process::Arguments parguments = Process::PrepareCommand(commandLine);
Log(LogWarning, "PluginCheckTask") Log(LogWarning, "PluginCheckTask")