add GpuCache Metrics

Former-commit-id: 9da6fcb11f69302fb1c00bb16fe4c6f7b7699ee5

cpp/src/db/DBImpl.cpp

@@ -312,6 +312,7 @@ void DBImpl::StartMetricTask() {
     int64_t cache_usage = cache::CpuCacheMgr::GetInstance()->CacheUsage();
     int64_t cache_total = cache::CpuCacheMgr::GetInstance()->CacheCapacity();
     server::Metrics::GetInstance().CpuCacheUsageGaugeSet(cache_usage*100/cache_total);
+    server::Metrics::GetInstance().GpuCacheUsageGaugeSet();
     uint64_t size;
     Size(size);
     server::Metrics::GetInstance().DataFileSizeGaugeSet(size);

cpp/src/grpc/gen-milvus/milvus.pb.cc

@@ -576,7 +576,7 @@ const char descriptor_table_protodef_milvus_2eproto[] PROTOBUF_SECTION_VARIABLE(
   "rpc.Status\"\000\022B\n\rDescribeIndex\022\026.milvus.g"
   "rpc.TableName\032\027.milvus.grpc.IndexParam\"\000"
   "\022:\n\tDropIndex\022\026.milvus.grpc.TableName\032\023."
-  "milvus.grpc.Status\"\000b\006proto3"
+  "milvus.grpc.Status\"\000B\002H\001b\006proto3"
   ;
 static const ::PROTOBUF_NAMESPACE_ID::internal::DescriptorTable*const descriptor_table_milvus_2eproto_deps[1] = {
   &::descriptor_table_status_2eproto,
@@ -603,7 +603,7 @@ static ::PROTOBUF_NAMESPACE_ID::internal::SCCInfoBase*const descriptor_table_mil
 static ::PROTOBUF_NAMESPACE_ID::internal::once_flag descriptor_table_milvus_2eproto_once;
 static bool descriptor_table_milvus_2eproto_initialized = false;
 const ::PROTOBUF_NAMESPACE_ID::internal::DescriptorTable descriptor_table_milvus_2eproto = {
-  &descriptor_table_milvus_2eproto_initialized, descriptor_table_protodef_milvus_2eproto, "milvus.proto", 2388,
+  &descriptor_table_milvus_2eproto_initialized, descriptor_table_protodef_milvus_2eproto, "milvus.proto", 2392,
   &descriptor_table_milvus_2eproto_once, descriptor_table_milvus_2eproto_sccs, descriptor_table_milvus_2eproto_deps, 17, 1,
   schemas, file_default_instances, TableStruct_milvus_2eproto::offsets,
   file_level_metadata_milvus_2eproto, 17, file_level_enum_descriptors_milvus_2eproto, file_level_service_descriptors_milvus_2eproto,

cpp/src/grpc/milvus.proto

 syntax = "proto3";
+option optimize_for = SPEED;
 
 import "status.proto";
 

cpp/src/metrics/MetricBase.h

@@ -32,7 +32,7 @@ class MetricsBase{
     virtual void BuildIndexDurationSecondsHistogramObserve(double value) {};
 
     virtual void CpuCacheUsageGaugeSet(double value) {};
-    virtual void GpuCacheUsageGaugeSet(double value) {};
+    virtual void GpuCacheUsageGaugeSet() {};
 
     virtual void MetaAccessTotalIncrement(double value = 1) {};
     virtual void MetaAccessDurationSecondsHistogramObserve(double value) {};

cpp/src/metrics/PrometheusMetrics.cpp

@@ -167,15 +167,26 @@ void PrometheusMetrics::CPUTemperature() {
     }
 }
 
-void PrometheusMetrics::GpuCacheUsageGaugeSet(double value) {
+void PrometheusMetrics::GpuCacheUsageGaugeSet() {
     if(!startup_) return;
-    int64_t num_processors = server::SystemInfo::GetInstance().num_processor();
+    server::ConfigNode& config = server::ServerConfig::GetInstance().GetConfig(server::CONFIG_CACHE);
+    std::string gpu_ids_str = config.GetValue(server::CONFIG_GPU_IDS, "0,1");
 
-    for (auto i = 0; i < num_processors; ++i) {
-//        int gpu_cache_usage = cache::GpuCacheMgr::GetInstance(i)->CacheUsage();
-//        int gpu_cache_total = cache::GpuCacheMgr::GetInstance(i)->CacheCapacity();
-//        prometheus::Gauge &gpu_cache = gpu_cache_usage_.Add({{"GPU_Cache", std::to_string(i)}});
-//        gpu_cache.Set(gpu_cache_usage * 100 / gpu_cache_total);
+    std::vector<uint64_t > gpu_ids;
+
+    std::stringstream ss(gpu_ids_str);
+    for (int i; ss >> i;) {
+        gpu_ids.push_back(i);
+        if (ss.peek() == ',') {
+            ss.ignore();
+        }
+    }
+
+    for(auto i = 0; i < gpu_ids.size(); ++i) {
+        uint64_t cache_usage = cache::GpuCacheMgr::GetInstance(gpu_ids[i])->CacheUsage();
+        uint64_t cache_capacity = cache::GpuCacheMgr::GetInstance(gpu_ids[i])->CacheCapacity();
+        prometheus::Gauge &gpu_cache = gpu_cache_usage_.Add({{"GPU_Cache", std::to_string(i)}});
+        gpu_cache.Set(cache_usage * 100 / cache_capacity);
     }
 }
 

cpp/src/metrics/PrometheusMetrics.h

@@ -55,7 +55,7 @@ class PrometheusMetrics: public MetricsBase {
     void IndexFileSizeHistogramObserve(double value) override { if(startup_) index_files_size_histogram_.Observe(value);};
     void BuildIndexDurationSecondsHistogramObserve(double value) override { if(startup_) build_index_duration_seconds_histogram_.Observe(value);};
     void CpuCacheUsageGaugeSet(double value) override { if(startup_) cpu_cache_usage_gauge_.Set(value);};
-    void GpuCacheUsageGaugeSet(double value) override; 
+    void GpuCacheUsageGaugeSet() override;
 
     void MetaAccessTotalIncrement(double value = 1) override { if(startup_) meta_access_total_.Increment(value);};
     void MetaAccessDurationSecondsHistogramObserve(double value) override { if(startup_) meta_access_duration_seconds_histogram_.Observe(value);};
@@ -343,7 +343,7 @@ class PrometheusMetrics: public MetricsBase {
         .Help("current cache usage by bytes")
         .Register(*registry_);
     prometheus::Gauge &cpu_cache_usage_gauge_ = cpu_cache_usage_.Add({});
-    
+
     //record GPU cache usage and %
     prometheus::Family<prometheus::Gauge> &gpu_cache_usage_ = prometheus::BuildGauge()
             .Name("gpu_cache_usage_bytes")

cpp/src/scheduler/ResourceMgr.cpp

@@ -60,7 +60,7 @@ ResourceMgr::Connect(const std::string &name1, const std::string &name2, Connect
 
 void
 ResourceMgr::Connect(ResourceWPtr &res1, ResourceWPtr &res2, Connection &connection) {
-    if (auto observe_a = res1.lock()) {
+    if (auto observe_a = res1.loc   k()) {
         if (auto observe_b = res2.lock()) {
             observe_a->AddNeighbour(std::static_pointer_cast<Node>(observe_b), connection);
             observe_b->AddNeighbour(std::static_pointer_cast<Node>(observe_a), connection);

cpp/src/sdk/grpc/ClientProxy.cpp

@@ -184,6 +184,7 @@ ClientProxy::Insert(const std::string &table_name,
 
         for (auto &record : record_array) {
             ::milvus::grpc::RowRecord *grpc_record = insert_param.add_row_record_array();
+//            memcpy(grpc_record->vector_data().data(), record.data.data(), record.data.size() * sizeof(float));
             for (size_t i = 0; i < record.data.size(); i++) {
                 grpc_record->add_vector_data(record.data[i]);
             }