DPL: refactor dumping of performanceMetrics.json

Framework/Core/CMakeLists.txt

@@ -240,6 +240,7 @@ add_executable(o2-test-framework-core
               test/test_OverrideLabels.cxx
               test/test_O2DataModelHelpers.cxx
               test/test_RootConfigParamHelpers.cxx
+              test/test_ResourcesMonitoringHelpers.cxx
               test/test_Services.cxx
               test/test_StringHelpers.cxx
               test/test_StaticFor.cxx

Framework/Core/include/Framework/DriverInfo.h

@@ -149,6 +149,8 @@ struct DriverInfo {
   std::string uniqueWorkflowId = "";
   /// Metrics gathering interval
   unsigned short resourcesMonitoringInterval = 0;
+  /// Where to dump the metrics
+  std::string resourcesMonitoringFilename = "performanceMetrics.json";
   /// Metrics gathering dump to disk interval
   unsigned short resourcesMonitoringDumpInterval = 0;
   /// Port used by the websocket control. 0 means not initialised.

Framework/Core/src/ResourcesMonitoringHelper.cxx

@@ -55,7 +55,8 @@ boost::property_tree::ptree fillNodeWithValue(const DeviceMetricsInfo& deviceMet
 bool ResourcesMonitoringHelper::dumpMetricsToJSON(const std::vector<DeviceMetricsInfo>& metrics,
                                                   const DeviceMetricsInfo& driverMetrics,
                                                   const std::vector<DeviceSpec>& specs,
-                                                  std::vector<std::regex> const& performanceMetricsRegex) noexcept
+                                                  std::vector<std::regex> const& performanceMetricsRegex,
+                                                  std::ostream& out) noexcept
 {
 
   assert(metrics.size() == specs.size());
@@ -76,7 +77,7 @@ bool ResourcesMonitoringHelper::dumpMetricsToJSON(const std::vector<DeviceMetric
       auto same = [metricLabel](std::regex const& matcher) -> bool {
         return std::regex_match(metricLabel.begin(), metricLabel.end(), matcher);
       };
-      //check if we are interested
+      // check if we are interested
       if (std::find_if(std::begin(performanceMetricsRegex), std::end(performanceMetricsRegex), same) == performanceMetricsRegex.end()) {
         continue;
       }
@@ -85,7 +86,7 @@ bool ResourcesMonitoringHelper::dumpMetricsToJSON(const std::vector<DeviceMetric
       if (deviceMetrics.metrics[mi].filledMetrics == 0) {
         continue;
       }
-      //if so
+      // if so
 
       boost::property_tree::ptree metricNode;
 
@@ -122,7 +123,7 @@ bool ResourcesMonitoringHelper::dumpMetricsToJSON(const std::vector<DeviceMetric
       return std::regex_match(metricLabel.begin(), metricLabel.end(), matcher);
     };
 
-    //check if we are interested
+    // check if we are interested
     if (std::find_if(std::begin(performanceMetricsRegex), std::end(performanceMetricsRegex), same) == performanceMetricsRegex.end()) {
       continue;
     }
@@ -133,7 +134,7 @@ bool ResourcesMonitoringHelper::dumpMetricsToJSON(const std::vector<DeviceMetric
       continue;
     }
 
-    //if so
+    // if so
     boost::property_tree::ptree metricNode;
 
     switch (driverMetrics.metrics[mi].type) {
@@ -161,14 +162,7 @@ bool ResourcesMonitoringHelper::dumpMetricsToJSON(const std::vector<DeviceMetric
 
   root.add_child("driver", driverRoot);
 
-  std::ofstream file("performanceMetrics.json", std::ios::out);
-  if (file.is_open()) {
-    boost::property_tree::json_parser::write_json(file, root);
-  } else {
-    return false;
-  }
-
-  file.close();
+  boost::property_tree::json_parser::write_json(out, root);
 
   return true;
 }

Framework/Core/src/ResourcesMonitoringHelper.h

@@ -13,27 +13,25 @@
 #define O2_FRAMEWORK_RESOURCESMONITORINGHELPER_H_
 
 #include "Framework/DeviceMetricsInfo.h"
-#include "Monitoring/ProcessMonitor.h"
 #include "Framework/DeviceSpec.h"
 
 #include <vector>
-#include <type_traits>
 #include <regex>
+#include <iosfwd>
 
 namespace o2::framework
 {
-
 struct ResourcesMonitoringHelper {
   /// Dump the metrics in @a metrics which match the names specified in @a metricsToDump
   /// @a specs are the DeviceSpecs associated to the metrics.
   static bool dumpMetricsToJSON(std::vector<DeviceMetricsInfo> const& metrics,
                                 DeviceMetricsInfo const& driverMetrics,
                                 std::vector<DeviceSpec> const& specs,
-                                std::vector<std::regex> const& metricsToDump) noexcept;
+                                std::vector<std::regex> const& metricsToDump,
+                                std::ostream& out) noexcept;
   static bool isResourcesMonitoringEnabled(unsigned short interval) noexcept { return interval > 0; }
 };
 
-
 } // namespace o2::framework
 
 #endif // O2_FRAMEWORK_RESOURCESMONITORINGHELPER_H_

Framework/Core/src/runDataProcessing.cxx

@@ -1247,8 +1247,10 @@ void dumpMetricsCallback(uv_timer_t* handle)
   auto* context = (DriverServerContext*)handle->data;
 
   static auto performanceMetrics = getDumpableMetrics();
+  std::ofstream file(context->driver->resourcesMonitoringFilename, std::ios::out);
   ResourcesMonitoringHelper::dumpMetricsToJSON(*(context->metrics),
-                                               context->driver->metrics, *(context->specs), performanceMetrics);
+                                               context->driver->metrics, *(context->specs), performanceMetrics,
+                                               file);
 }
 
 void dumpRunSummary(DriverServerContext& context, DriverInfo const& driverInfo, DeviceInfos const& infos, DeviceSpecs const& specs)
@@ -2035,6 +2037,7 @@ int runStateMachine(DataProcessorSpecs const& workflow,
             "--fairmq-ipc-prefix",
             "--readers",
             "--resources-monitoring",
+            "--resources-monitoring-file",
             "--resources-monitoring-dump-interval",
             "--time-limit",
           };
@@ -2268,7 +2271,7 @@ int runStateMachine(DataProcessorSpecs const& workflow,
           if (driverInfo.resourcesMonitoringDumpInterval) {
             uv_timer_stop(&metricDumpTimer);
           }
-          LOG(info) << "Dumping performance metrics to performanceMetrics.json file";
+          LOGP(info, "Dumping performance metrics to {}.json file", driverInfo.resourcesMonitoringFilename);
           dumpMetricsCallback(&metricDumpTimer);
         }
         dumpRunSummary(serverContext, driverInfo, infos, runningWorkflow.devices);
@@ -2916,6 +2919,7 @@ int doMain(int argc, char** argv, o2::framework::WorkflowSpec const& workflow,
     ("no-IPC", bpo::value<bool>()->zero_tokens()->default_value(false), "disable IPC topology optimization")                                                           //                                                                                                                                        //
     ("o2-control,o2", bpo::value<std::string>()->default_value(""), "dump O2 Control workflow configuration under the specified name")                                 //
     ("resources-monitoring", bpo::value<unsigned short>()->default_value(0), "enable cpu/memory monitoring for provided interval in seconds")                          //
+    ("resources-monitoring-file", bpo::value<std::string>()->default_value("performanceMetrics.json"), "file where to dump the metrics")                               //
     ("resources-monitoring-dump-interval", bpo::value<unsigned short>()->default_value(0), "dump monitoring information to disk every provided seconds");              //
   // some of the options must be forwarded by default to the device
   executorOptions.add(DeviceSpecHelpers::getForwardedDeviceOptions());
@@ -3186,6 +3190,7 @@ int doMain(int argc, char** argv, o2::framework::WorkflowSpec const& workflow,
   driverInfo.deployHostname = varmap["hostname"].as<std::string>();
   driverInfo.resources = varmap["resources"].as<std::string>();
   driverInfo.resourcesMonitoringInterval = varmap["resources-monitoring"].as<unsigned short>();
+  driverInfo.resourcesMonitoringFilename = varmap["resources-monitoring-file"].as<std::string>();
   driverInfo.resourcesMonitoringDumpInterval = varmap["resources-monitoring-dump-interval"].as<unsigned short>();
 
   // FIXME: should use the whole dataProcessorInfos, actually...