modify prometheus

Former-commit-id: 062c830e0ce3c8c1b7d9acffdcae5002ea6e2dd1

cpp/src/metrics/PrometheusMetrics.cpp

@@ -60,47 +60,34 @@ PrometheusMetrics::RAMUsagePercentSet() {
 void
 PrometheusMetrics::GPUPercentGaugeSet() {
     if(!startup_) return;
-    int numDevide = server::SystemInfo::GetInstance().num_device();
-    std::vector<unsigned int> values = server::SystemInfo::GetInstance().GPUPercent();
+    int numDevice = server::SystemInfo::GetInstance().num_device();
+//    std::vector<unsigned int> values = server::SystemInfo::GetInstance().GPUPercent();
+    std::vector<unsigned long long > used_memory = server::SystemInfo::GetInstance().GPUMemoryUsed();
+    constexpr unsigned long long MtoB = 1024*1024;
+
 
-    for (int i = 0; i < values.size(); i++) {
+    for (int i = 0; i < numDevice; i++) {
         prometheus::Gauge &GPU_percent = GPU_percent_.Add({{"DeviceNum", std::to_string(i)}});
-        GPU_percent.Set(static_cast<double>(values[i]));
+//        std::cout << "nvmlDeviceGetUtilizationRates: " << values[i] << std::endl;
+//        GPU_percent.Set(static_cast<double>(values[i]));
+        double percent = (double)used_memory[i] / (double)MtoB;
+        double res = (percent / 6078) * 100;
+        GPU_percent.Set(res);
     }
 
-//    if(numDevide >= 1) GPU0_percent_gauge_.Set(static_cast<double>(values[0]));
-//    if(numDevide >= 2) GPU1_percent_gauge_.Set(static_cast<double>(values[1]));
-//    if(numDevide >= 3) GPU2_percent_gauge_.Set(static_cast<double>(values[2]));
-//    if(numDevide >= 4) GPU3_percent_gauge_.Set(static_cast<double>(values[3]));
-//    if(numDevide >= 5) GPU4_percent_gauge_.Set(static_cast<double>(values[4]));
-//    if(numDevide >= 6) GPU5_percent_gauge_.Set(static_cast<double>(values[5]));
-//    if(numDevide >= 7) GPU6_percent_gauge_.Set(static_cast<double>(values[6]));
-//    if(numDevide >= 8) GPU7_percent_gauge_.Set(static_cast<double>(values[7]));
-
-    // to do
 }
 
 void PrometheusMetrics::GPUMemoryUsageGaugeSet() {
     if(!startup_) return;
     std::vector<unsigned long long> values = server::SystemInfo::GetInstance().GPUMemoryUsed();
     constexpr unsigned long long MtoB = 1024*1024;
-    int numDevice = values.size();
+    int numDevice = server::SystemInfo::GetInstance().num_device();
 
     for (int i = 0; i < numDevice; i++) {
         prometheus::Gauge &GPU_memory = GPU_memory_usage_.Add({{"DeviceNum", std::to_string(i)}});
         GPU_memory.Set(values[i] / MtoB);
     }
 
-
-//    if(numDevice >=1) GPU0_memory_usage_gauge_.Set(values[0]/MtoB);
-//    if(numDevice >=2) GPU1_memory_usage_gauge_.Set(values[1]/MtoB);
-//    if(numDevice >=3) GPU2_memory_usage_gauge_.Set(values[2]/MtoB);
-//    if(numDevice >=4) GPU3_memory_usage_gauge_.Set(values[3]/MtoB);
-//    if(numDevice >=5) GPU4_memory_usage_gauge_.Set(values[4]/MtoB);
-//    if(numDevice >=6) GPU5_memory_usage_gauge_.Set(values[5]/MtoB);
-//    if(numDevice >=7) GPU6_memory_usage_gauge_.Set(values[6]/MtoB);
-//    if(numDevice >=8) GPU7_memory_usage_gauge_.Set(values[7]/MtoB);
-
 }
 void PrometheusMetrics::AddVectorsPerSecondGaugeSet(int num_vector, int dim, double time) {
     // MB/s
@@ -162,7 +149,6 @@ void PrometheusMetrics::CPUCoreUsagePercentSet() {
     for (int i = 0; i < cpu_core_percent.size(); i++) {
         prometheus::Gauge &core_percent = CPU_.Add({{"CPU", std::to_string(i)}});
         core_percent.Set(cpu_core_percent[i]);
-//        std::cout << cpu_core_percent[i] << "+";
     }
 }
 

cpp/src/metrics/PrometheusMetrics.h

@@ -397,7 +397,6 @@ class PrometheusMetrics: public MetricsBase {
         .Help("CPU usage percent by this this process")
         .Register(*registry_);
     prometheus::Gauge &CPU_usage_percent_ = CPU_.Add({{"CPU", "0"}});
-//    std::vector<prometheus::Gauge> &CPU_usage_percent_array;
 
 
     prometheus::Family<prometheus::Gauge> &RAM_ = prometheus::BuildGauge()
@@ -411,32 +410,12 @@ class PrometheusMetrics: public MetricsBase {
         .Name("Gpu_usage_percent")
         .Help("GPU_usage_percent ")
         .Register(*registry_);
-//    prometheus::Gauge &GPU0_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "0"}});
-//    prometheus::Gauge &GPU1_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "1"}});
-//    prometheus::Gauge &GPU2_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "2"}});
-//    prometheus::Gauge &GPU3_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "3"}});
-//    prometheus::Gauge &GPU4_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "4"}});
-//    prometheus::Gauge &GPU5_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "5"}});
-//    prometheus::Gauge &GPU6_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "6"}});
-//    prometheus::Gauge &GPU7_percent_gauge_ = GPU_percent_.Add({{"DeviceNum", "7"}});
-
-
-
 
     //GPU Mempry used
     prometheus::Family<prometheus::Gauge> &GPU_memory_usage_ = prometheus::BuildGauge()
         .Name("GPU_memory_usage_total")
         .Help("GPU memory usage total ")
         .Register(*registry_);
-//    prometheus::Gauge &GPU0_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "0"}});
-//    prometheus::Gauge &GPU1_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "1"}});
-//    prometheus::Gauge &GPU2_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "2"}});
-//    prometheus::Gauge &GPU3_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "3"}});
-//    prometheus::Gauge &GPU4_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "4"}});
-//    prometheus::Gauge &GPU5_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "5"}});
-//    prometheus::Gauge &GPU6_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "6"}});
-//    prometheus::Gauge &GPU7_memory_usage_gauge_ = GPU_memory_usage_.Add({{"DeviceNum", "7"}});
-
 
     prometheus::Family<prometheus::Gauge> &query_index_type_per_second_ = prometheus::BuildGauge()
         .Name("query_index_throughtout_per_microsecond")