From 7a060713a0d5cb9f074c02c2f627f5c88dc02145 Mon Sep 17 00:00:00 2001
From: Rafal Bielski <rafal.bielski@cern.ch>
Date: Mon, 7 Dec 2020 18:15:36 +0100
Subject: [PATCH] TrigOnlineMonitor: Add histograms to monitor number of forks,
threads and slots
---
.../TrigOnlineMonitor/src/TrigOpMonitor.cxx | 47 ++++++++++++++++++-
.../TrigOnlineMonitor/src/TrigOpMonitor.h | 3 ++
2 files changed, 49 insertions(+), 1 deletion(-)
diff --git a/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.cxx b/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.cxx
index 16ba506264e..8c529f4b1d0 100644
--- a/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.cxx
+++ b/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.cxx
@@ -76,6 +76,8 @@ void TrigOpMonitor::handle( const Incident& incident ) {
fillMagFieldHist();
fillIOVDbHist();
fillSubDetHist();
+ const AthenaInterprocess::UpdateAfterFork& updinc = dynamic_cast<const AthenaInterprocess::UpdateAfterFork&>(incident);
+ fillProcThreadHist(updinc.workerID());
}
}
@@ -128,6 +130,14 @@ StatusCode TrigOpMonitor::bookHists()
m_releaseHist = new TH1I("GeneralOpInfo", "General operational info;;Applications", 1, 0, 1);
+ m_mtConfigHist = new TH2I("MTConfig", "Multi-threading configuration", 2, 0, 2, 10, 0, 10);
+ m_mtConfigHist->SetCanExtend(TH1::kYaxis);
+ m_mtConfigHist->GetXaxis()->SetBinLabel(1, "N threads");
+ m_mtConfigHist->GetXaxis()->SetBinLabel(2, "N slots");
+
+ m_workersHist = new TH1I("MPWorkers", "Worker IDs;Worker ID;Number of workers", 10, 0, 10);
+ m_workersHist->SetCanExtend(TH1::kXaxis);
+
m_subdetHist = new TH2I("Subdetectors", "State of subdetectors", 1, 0, 1, 3, 0, 3);
m_subdetHist->SetCanExtend(TH1::kXaxis);
m_subdetHist->GetYaxis()->SetBinLabel(1, "# ROB");
@@ -140,7 +150,10 @@ StatusCode TrigOpMonitor::bookHists()
m_muHist = new TProfile("Pileup", "Pileup;Lumiblock;Interactions per BX", m_maxLB, 0, m_maxLB);
// Register histograms
- TH1* hist[] = {m_releaseHist, m_subdetHist, m_iovChangeHist, m_magFieldHist, m_lumiHist, m_muHist};
+ TH1* hist[] = {
+ m_releaseHist, m_mtConfigHist, m_workersHist, m_subdetHist,
+ m_iovChangeHist, m_magFieldHist, m_lumiHist, m_muHist
+ };
for (TH1* h : hist) {
if (h) ATH_CHECK(m_histSvc->regHist(m_histPath + h->GetName(), h));
}
@@ -347,6 +360,38 @@ void TrigOpMonitor::fillReleaseDataHist()
}
}
+void TrigOpMonitor::fillProcThreadHist(const int workerID)
+{
+ m_workersHist->Fill(workerID, 1);
+
+ auto getIntProp = [this](std::string_view name, const std::string& prop) -> std::optional<int> {
+ IProperty* svc = serviceLocator()->service(name).as<IProperty>();
+ if (svc == nullptr) return std::nullopt;
+ try {
+ return std::stoi(svc->getProperty(prop).toString());
+ }
+ catch (...) {
+ return std::nullopt;
+ }
+ };
+
+ const std::optional<int> numThreads = getIntProp("AvalancheSchedulerSvc","ThreadPoolSize");
+ if (numThreads.has_value()) {
+ m_mtConfigHist->Fill("N threads", numThreads.value(), 1.0);
+ }
+ else {
+ ATH_MSG_WARNING("Could not retrieve the number of threads to fill the monitoring histogram");
+ }
+
+ const std::optional<int> numSlots = getIntProp("EventDataSvc","NSlots");
+ if (numSlots.has_value()) {
+ m_mtConfigHist->Fill("N slots", numSlots.value(), 1.0);
+ }
+ else {
+ ATH_MSG_WARNING("Could not retrieve the number of slots to fill the monitoring histogram");
+ }
+}
+
void TrigOpMonitor::fillSubDetHist()
{
// Retrieve the enabled ROBs/SubDets list from DataFlowConfig which is a special object
diff --git a/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.h b/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.h
index de316dcc928..5f4442dd042 100644
--- a/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.h
+++ b/HLT/Trigger/TrigMonitoring/TrigOnlineMonitor/src/TrigOpMonitor.h
@@ -51,6 +51,7 @@ public:
private:
void fillMagFieldHist();
void fillReleaseDataHist();
+ void fillProcThreadHist(int workerID);
void fillSubDetHist();
void fillIOVDbHist();
void fillIOVDbChangeHist(const EventContext& ctx);
@@ -75,6 +76,8 @@ private:
TH2I* m_magFieldHist{nullptr};
TH2I* m_iovChangeHist{nullptr};
TH1I* m_releaseHist{nullptr};
+ TH2I* m_mtConfigHist{nullptr};
+ TH1I* m_workersHist{nullptr};
TH2I* m_subdetHist{nullptr};
TProfile* m_lumiHist{nullptr};
TProfile* m_muHist{nullptr};
--
GitLab