The GPU cache holds much more data than the configured value (#4720)

* The GPU cache holds much more data than the configured value
Signed-off-by: Ncmli <chengming.li@zilliz.com>

* update changelog and remove the useless variable G_BYTE and unit
Signed-off-by: Ncmli <chengming.li@zilliz.com>

CHANGELOG.md

@@ -6,6 +6,7 @@ Please mark all change in change log and use the issue from GitHub
 ## Bug
 -   \#4683 A negative zero may be returned if the metric_type is Tanimoto
 -   \#4678 Server crash on BinaryFlat if dimension is not a power of 2
+-   \#4720 The GPU cache holds much more data than the configured value
 
 ## Feature
 

core/src/cache/CpuCacheMgr.cpp

@@ -21,18 +21,13 @@
 namespace milvus {
 namespace cache {
 
-namespace {
-// constexpr int64_t unit = 1024 * 1024 * 1024;
-constexpr int64_t unit = 1;
-}  // namespace
-
 CpuCacheMgr::CpuCacheMgr() {
     // All config values have been checked in Config::ValidateConfig()
     server::Config& config = server::Config::GetInstance();
 
     int64_t cpu_cache_cap;
     config.GetCacheConfigCpuCacheCapacity(cpu_cache_cap);
-    int64_t cap = cpu_cache_cap * unit;
+    int64_t cap = cpu_cache_cap;
     LOG_SERVER_DEBUG_ << "cpu cache.size: " << cap;
     LOG_SERVER_INFO_ << "cpu cache.size: " << cap;
     cache_ = std::make_shared<Cache<DataObjPtr>>(cap, 1UL << 32, "[CACHE CPU]");
@@ -59,7 +54,7 @@ CpuCacheMgr::GetIndex(const std::string& key) {
 
 void
 CpuCacheMgr::OnCpuCacheCapacityChanged(int64_t value) {
-    SetCapacity(value * unit);
+    SetCapacity(value);
 }
 
 }  // namespace cache

core/src/cache/GpuCacheMgr.cpp

@@ -24,17 +24,13 @@ namespace cache {
 std::mutex GpuCacheMgr::global_mutex_;
 std::unordered_map<int64_t, GpuCacheMgrPtr> GpuCacheMgr::instance_;
 
-namespace {
-constexpr int64_t G_BYTE = 1024 * 1024 * 1024;
-}
-
 GpuCacheMgr::GpuCacheMgr(int64_t gpu_id) : gpu_id_(gpu_id) {
     // All config values have been checked in Config::ValidateConfig()
     server::Config& config = server::Config::GetInstance();
 
     int64_t gpu_cache_cap;
     config.GetGpuResourceConfigCacheCapacity(gpu_cache_cap);
-    int64_t cap = gpu_cache_cap * G_BYTE;
+    int64_t cap = gpu_cache_cap;
     std::string header = "[CACHE GPU" + std::to_string(gpu_id) + "]";
     cache_ = std::make_shared<Cache<DataObjPtr>>(cap, 1UL << 32, header);
 
@@ -84,7 +80,7 @@ GpuCacheMgr::GetInstance(int64_t gpu_id) {
 void
 GpuCacheMgr::OnGpuCacheCapacityChanged(int64_t capacity) {
     for (auto& iter : instance_) {
-        iter.second->SetCapacity(capacity * G_BYTE);
+        iter.second->SetCapacity(capacity);
     }
 }