mirror of
https://gitee.com/milvus-io/milvus.git
synced 2026-01-06 19:02:18 +08:00
fix conflict
This commit is contained in:
commit
f1f90ec3e0
@ -20,7 +20,11 @@ Please mark all change in change log and use the ticket from JIRA.
|
||||
- \#440 - Query API in customization still uses old version
|
||||
- \#440 - Server cannot startup with gpu_resource_config.enable=false in GPU version
|
||||
- \#458 - Index data is not compatible between 0.5 and 0.6
|
||||
- \#465 - Server hang caused by searching with nsg index
|
||||
- \#486 - gpu no usage during index building
|
||||
- \#509 - IVF_PQ index build trapped into dead loop caused by invalid params
|
||||
- \#513 - Unittest DELETE_BY_RANGE sometimes failed
|
||||
- \#527 - faiss benchmark not compatible with faiss 1.6.0
|
||||
- \#530 - BuildIndex stop when do build index and search simultaneously
|
||||
|
||||
## Feature
|
||||
@ -31,6 +35,7 @@ Please mark all change in change log and use the ticket from JIRA.
|
||||
- \#227 - Support new index types SPTAG-KDT and SPTAG-BKT
|
||||
- \#346 - Support build index with multiple gpu
|
||||
- \#488 - Add log in scheduler/optimizer
|
||||
- \#502 - C++ SDK support IVFPQ and SPTAG
|
||||
|
||||
## Improvement
|
||||
- \#255 - Add ivfsq8 test report detailed version
|
||||
|
||||
@ -7,6 +7,7 @@
|
||||
[](https://codebeat.co/projects/github-com-jinhai-cn-milvus-master)
|
||||

|
||||

|
||||
[](https://codecov.io/gh/milvus-io/milvus)
|
||||
|
||||
[中文版](README_CN.md) | [日本語版](README_JP.md)
|
||||
|
||||
|
||||
@ -6,6 +6,8 @@
|
||||
[](https://codebeat.co/projects/github-com-jinhai-cn-milvus-master)
|
||||

|
||||

|
||||
[](https://codecov.io/gh/milvus-io/milvus)
|
||||
|
||||
|
||||
# 欢迎来到 Milvus
|
||||
|
||||
@ -15,7 +17,7 @@ Milvus 是一款开源的、针对海量特征向量的相似性搜索引擎。
|
||||
|
||||
若要了解 Milvus 详细介绍和整体架构,请访问 [Milvus 简介](https://www.milvus.io/docs/zh-CN/aboutmilvus/overview/)。
|
||||
|
||||
Milvus 提供稳定的 [Python](https://github.com/milvus-io/pymilvus)、[Java](https://github.com/milvus-io/milvus-sdk-java) 以及 C++ 的 API 接口。
|
||||
Milvus 提供稳定的 [Python](https://github.com/milvus-io/pymilvus)、[Java](https://github.com/milvus-io/milvus-sdk-java) 以及[C++](https://github.com/milvus-io/milvus/tree/master/core/src/sdk) 的 API 接口。
|
||||
|
||||
通过 [版本发布说明](https://milvus.io/docs/zh-CN/release/v0.5.3/) 获取最新版本的功能和更新。
|
||||
|
||||
|
||||
6
ci/jenkins/Jenkinsfile
vendored
6
ci/jenkins/Jenkinsfile
vendored
@ -114,7 +114,8 @@ pipeline {
|
||||
|
||||
stage("Deploy to Development") {
|
||||
environment {
|
||||
HELM_RELEASE_NAME = "${env.PIPELINE_NAME}-${env.SEMVER}-${env.BUILD_NUMBER}-single-${env.BINRARY_VERSION}".toLowerCase()
|
||||
FROMAT_SEMVER = "${env.SEMVER}".replaceAll("\\.", "-")
|
||||
HELM_RELEASE_NAME = "${env.PIPELINE_NAME}-${env.FROMAT_SEMVER}-${env.BUILD_NUMBER}-single-${env.BINRARY_VERSION}".toLowerCase()
|
||||
}
|
||||
|
||||
agent {
|
||||
@ -248,7 +249,8 @@ pipeline {
|
||||
|
||||
stage("Deploy to Development") {
|
||||
environment {
|
||||
HELM_RELEASE_NAME = "${env.PIPELINE_NAME}-${env.SEMVER}-${env.BUILD_NUMBER}-single-${env.BINRARY_VERSION}".toLowerCase()
|
||||
FROMAT_SEMVER = "${env.SEMVER}".replaceAll("\\.", "-")
|
||||
HELM_RELEASE_NAME = "${env.PIPELINE_NAME}-${env.FROMAT_SEMVER}-${env.BUILD_NUMBER}-single-${env.BINRARY_VERSION}".toLowerCase()
|
||||
}
|
||||
|
||||
agent {
|
||||
|
||||
@ -2,8 +2,9 @@ timeout(time: 5, unit: 'MINUTES') {
|
||||
dir ("ci/jenkins/scripts") {
|
||||
sh "pip3 install -r requirements.txt"
|
||||
sh "./yaml_processor.py merge -f /opt/milvus/conf/server_config.yaml -m ../yaml/update_server_config.yaml -i && rm /opt/milvus/conf/server_config.yaml.bak"
|
||||
sh "sed -i 's/\/tmp\/milvus/\/opt\/milvus/g' /opt/milvus/conf/log_config.conf"
|
||||
sh "sed -i 's/\\/tmp\\/milvus/\\/opt\\/milvus/g' /opt/milvus/conf/log_config.conf"
|
||||
}
|
||||
sh "rm -rf /opt/milvus/unittest"
|
||||
sh "tar -zcvf ./${PROJECT_NAME}-${PACKAGE_VERSION}.tar.gz -C /opt/ milvus"
|
||||
withCredentials([usernamePassword(credentialsId: "${params.JFROG_CREDENTIALS_ID}", usernameVariable: 'JFROG_USERNAME', passwordVariable: 'JFROG_PASSWORD')]) {
|
||||
def uploadStatus = sh(returnStatus: true, script: "curl -u${JFROG_USERNAME}:${JFROG_PASSWORD} -T ./${PROJECT_NAME}-${PACKAGE_VERSION}.tar.gz ${params.JFROG_ARTFACTORY_URL}/milvus/package/${PROJECT_NAME}-${PACKAGE_VERSION}.tar.gz")
|
||||
|
||||
@ -31,9 +31,21 @@ GET_CURRENT_TIME(BUILD_TIME)
|
||||
string(REGEX REPLACE "\n" "" BUILD_TIME ${BUILD_TIME})
|
||||
message(STATUS "Build time = ${BUILD_TIME}")
|
||||
|
||||
if (NOT DEFINED CMAKE_BUILD_TYPE)
|
||||
set(CMAKE_BUILD_TYPE Release CACHE STRING "Choose the type of build.")
|
||||
endif ()
|
||||
|
||||
set (GIT_BRANCH_NAME_REGEX "[0-9]+\\.[0-9]+\\.[0-9]")
|
||||
|
||||
MACRO(GET_GIT_BRANCH_NAME GIT_BRANCH_NAME)
|
||||
execute_process(COMMAND sh "-c" "git log --decorate | head -n 1 | sed 's/.*(\\(.*\\))/\\1/' | sed 's/.* \\(.*\\),.*/\\1/' | sed 's=[a-zA-Z]*\/==g'"
|
||||
OUTPUT_VARIABLE ${GIT_BRANCH_NAME})
|
||||
if(NOT GIT_BRANCH_NAME MATCHES "${GIT_BRANCH_NAME_REGEX}")
|
||||
execute_process(COMMAND "git" rev-parse --abbrev-ref HEAD OUTPUT_VARIABLE ${GIT_BRANCH_NAME})
|
||||
endif ()
|
||||
if(NOT GIT_BRANCH_NAME MATCHES "${GIT_BRANCH_NAME_REGEX}")
|
||||
execute_process(COMMAND "git" symbolic-ref --short -q HEAD HEAD OUTPUT_VARIABLE ${GIT_BRANCH_NAME})
|
||||
endif ()
|
||||
ENDMACRO(GET_GIT_BRANCH_NAME)
|
||||
|
||||
GET_GIT_BRANCH_NAME(GIT_BRANCH_NAME)
|
||||
@ -43,7 +55,7 @@ if (NOT GIT_BRANCH_NAME STREQUAL "")
|
||||
endif ()
|
||||
|
||||
set(MILVUS_VERSION "${GIT_BRANCH_NAME}")
|
||||
string(REGEX MATCH "[0-9]+\\.[0-9]+\\.[0-9]" MILVUS_VERSION "${MILVUS_VERSION}")
|
||||
string(REGEX MATCH "${GIT_BRANCH_NAME_REGEX}" MILVUS_VERSION "${MILVUS_VERSION}")
|
||||
|
||||
if (CMAKE_BUILD_TYPE STREQUAL "Release")
|
||||
set(BUILD_TYPE "Release")
|
||||
|
||||
@ -56,7 +56,7 @@ while getopts "p:d:t:f:ulrcgjhxzme" arg; do
|
||||
USE_JFROG_CACHE="ON"
|
||||
;;
|
||||
x)
|
||||
CUSTOMIZATION="OFF" # force use ori faiss
|
||||
CUSTOMIZATION="ON"
|
||||
;;
|
||||
g)
|
||||
GPU_VERSION="ON"
|
||||
|
||||
2
core/src/cache/GpuCacheMgr.cpp
vendored
2
core/src/cache/GpuCacheMgr.cpp
vendored
@ -25,6 +25,7 @@
|
||||
namespace milvus {
|
||||
namespace cache {
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
std::mutex GpuCacheMgr::mutex_;
|
||||
std::unordered_map<uint64_t, GpuCacheMgrPtr> GpuCacheMgr::instance_;
|
||||
|
||||
@ -76,6 +77,7 @@ GpuCacheMgr::GetIndex(const std::string& key) {
|
||||
DataObjPtr obj = GetItem(key);
|
||||
return obj;
|
||||
}
|
||||
#endif
|
||||
|
||||
} // namespace cache
|
||||
} // namespace milvus
|
||||
|
||||
2
core/src/cache/GpuCacheMgr.h
vendored
2
core/src/cache/GpuCacheMgr.h
vendored
@ -25,6 +25,7 @@
|
||||
namespace milvus {
|
||||
namespace cache {
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
class GpuCacheMgr;
|
||||
using GpuCacheMgrPtr = std::shared_ptr<GpuCacheMgr>;
|
||||
|
||||
@ -42,6 +43,7 @@ class GpuCacheMgr : public CacheMgr<DataObjPtr> {
|
||||
static std::mutex mutex_;
|
||||
static std::unordered_map<uint64_t, GpuCacheMgrPtr> instance_;
|
||||
};
|
||||
#endif
|
||||
|
||||
} // namespace cache
|
||||
} // namespace milvus
|
||||
|
||||
@ -105,7 +105,8 @@ DBImpl::Stop() {
|
||||
shutting_down_.store(true, std::memory_order_release);
|
||||
|
||||
// makesure all memory data serialized
|
||||
MemSerialize();
|
||||
std::set<std::string> sync_table_ids;
|
||||
SyncMemData(sync_table_ids);
|
||||
|
||||
// wait compaction/buildindex finish
|
||||
bg_timer_thread_.join();
|
||||
@ -329,7 +330,10 @@ DBImpl::CreateIndex(const std::string& table_id, const TableIndex& index) {
|
||||
return SHUTDOWN_ERROR;
|
||||
}
|
||||
|
||||
Status status;
|
||||
// serialize memory data
|
||||
std::set<std::string> sync_table_ids;
|
||||
auto status = SyncMemData(sync_table_ids);
|
||||
|
||||
{
|
||||
std::unique_lock<std::mutex> lock(build_index_mutex_);
|
||||
|
||||
@ -588,12 +592,12 @@ DBImpl::StartMetricTask() {
|
||||
}
|
||||
|
||||
Status
|
||||
DBImpl::MemSerialize() {
|
||||
DBImpl::SyncMemData(std::set<std::string>& sync_table_ids) {
|
||||
std::lock_guard<std::mutex> lck(mem_serialize_mutex_);
|
||||
std::set<std::string> temp_table_ids;
|
||||
mem_mgr_->Serialize(temp_table_ids);
|
||||
for (auto& id : temp_table_ids) {
|
||||
compact_table_ids_.insert(id);
|
||||
sync_table_ids.insert(id);
|
||||
}
|
||||
|
||||
if (!temp_table_ids.empty()) {
|
||||
@ -612,7 +616,7 @@ DBImpl::StartCompactionTask() {
|
||||
}
|
||||
|
||||
// serialize memory data
|
||||
MemSerialize();
|
||||
SyncMemData(compact_table_ids_);
|
||||
|
||||
// compactiong has been finished?
|
||||
{
|
||||
|
||||
@ -150,7 +150,7 @@ class DBImpl : public DB {
|
||||
BackgroundBuildIndex();
|
||||
|
||||
Status
|
||||
MemSerialize();
|
||||
SyncMemData(std::set<std::string>& sync_table_ids);
|
||||
|
||||
Status
|
||||
GetFilesToBuildIndex(const std::string& table_id, const std::vector<int>& file_types,
|
||||
|
||||
@ -151,6 +151,7 @@ ExecutionEngineImpl::HybridLoad() const {
|
||||
return;
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
const std::string key = location_ + ".quantizer";
|
||||
|
||||
server::Config& config = server::Config::GetInstance();
|
||||
@ -205,6 +206,7 @@ ExecutionEngineImpl::HybridLoad() const {
|
||||
auto cache_quantizer = std::make_shared<CachedQuantizer>(quantizer);
|
||||
cache::GpuCacheMgr::GetInstance(best_device_id)->InsertItem(key, cache_quantizer);
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
void
|
||||
@ -342,6 +344,7 @@ ExecutionEngineImpl::CopyToGpu(uint64_t device_id, bool hybrid) {
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
auto index = std::static_pointer_cast<VecIndex>(cache::GpuCacheMgr::GetInstance(device_id)->GetIndex(location_));
|
||||
bool already_in_cache = (index != nullptr);
|
||||
if (already_in_cache) {
|
||||
@ -364,16 +367,19 @@ ExecutionEngineImpl::CopyToGpu(uint64_t device_id, bool hybrid) {
|
||||
if (!already_in_cache) {
|
||||
GpuCache(device_id);
|
||||
}
|
||||
#endif
|
||||
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
Status
|
||||
ExecutionEngineImpl::CopyToIndexFileToGpu(uint64_t device_id) {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
gpu_num_ = device_id;
|
||||
auto to_index_data = std::make_shared<ToIndexData>(PhysicalSize());
|
||||
cache::DataObjPtr obj = std::static_pointer_cast<cache::DataObj>(to_index_data);
|
||||
milvus::cache::GpuCacheMgr::GetInstance(device_id)->InsertItem(location_, obj);
|
||||
#endif
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
@ -584,15 +590,17 @@ ExecutionEngineImpl::Cache() {
|
||||
|
||||
Status
|
||||
ExecutionEngineImpl::GpuCache(uint64_t gpu_id) {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
cache::DataObjPtr obj = std::static_pointer_cast<cache::DataObj>(index_);
|
||||
milvus::cache::GpuCacheMgr::GetInstance(gpu_id)->InsertItem(location_, obj);
|
||||
|
||||
#endif
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
// TODO(linxj): remove.
|
||||
Status
|
||||
ExecutionEngineImpl::Init() {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
server::Config& config = server::Config::GetInstance();
|
||||
std::vector<int64_t> gpu_ids;
|
||||
Status s = config.GetGpuResourceConfigBuildIndexResources(gpu_ids);
|
||||
@ -604,6 +612,9 @@ ExecutionEngineImpl::Init() {
|
||||
|
||||
std::string msg = "Invalid gpu_num";
|
||||
return Status(SERVER_INVALID_ARGUMENT, msg);
|
||||
#else
|
||||
return Status::OK();
|
||||
#endif
|
||||
}
|
||||
|
||||
} // namespace engine
|
||||
|
||||
@ -74,7 +74,7 @@ function(ExternalProject_Use_Cache project_name package_file install_path)
|
||||
${CMAKE_COMMAND} -E echo
|
||||
"Extracting ${package_file} to ${install_path}"
|
||||
COMMAND
|
||||
${CMAKE_COMMAND} -E tar xzvf ${package_file} ${install_path}
|
||||
${CMAKE_COMMAND} -E tar xzf ${package_file} ${install_path}
|
||||
WORKING_DIRECTORY ${INDEX_BINARY_DIR}
|
||||
)
|
||||
|
||||
|
||||
@ -17,6 +17,7 @@
|
||||
|
||||
#include <iostream> // TODO(linxj): using Log instead
|
||||
|
||||
#include "knowhere/common/Log.h"
|
||||
#include "knowhere/common/Timer.h"
|
||||
|
||||
namespace knowhere {
|
||||
@ -51,30 +52,18 @@ TimeRecorder::PrintTimeRecord(const std::string& msg, double span) {
|
||||
std::cout << str_log << std::endl;
|
||||
break;
|
||||
}
|
||||
// case 1: {
|
||||
// SERVER_LOG_DEBUG << str_log;
|
||||
// break;
|
||||
//}
|
||||
case 1: {
|
||||
KNOWHERE_LOG_DEBUG << str_log;
|
||||
break;
|
||||
}
|
||||
// case 2: {
|
||||
// SERVER_LOG_INFO << str_log;
|
||||
// break;
|
||||
//}
|
||||
// KNOWHERE_LOG_TRACE << str_log;
|
||||
// break;
|
||||
// }
|
||||
// case 3: {
|
||||
// SERVER_LOG_WARNING << str_log;
|
||||
// break;
|
||||
//}
|
||||
// case 4: {
|
||||
// SERVER_LOG_ERROR << str_log;
|
||||
// break;
|
||||
//}
|
||||
// case 5: {
|
||||
// SERVER_LOG_FATAL << str_log;
|
||||
// break;
|
||||
//}
|
||||
// default: {
|
||||
// SERVER_LOG_INFO << str_log;
|
||||
// break;
|
||||
//}
|
||||
// KNOWHERE_LOG_WARNING << str_log;
|
||||
// break;
|
||||
// }
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -718,29 +718,38 @@ NsgIndex::Search(const float* query, const unsigned& nq, const unsigned& dim, co
|
||||
int64_t* ids, SearchParams& params) {
|
||||
std::vector<std::vector<Neighbor>> resset(nq);
|
||||
|
||||
TimeRecorder rc("search");
|
||||
if (nq == 1) {
|
||||
params.search_length = k;
|
||||
TimeRecorder rc("NsgIndex::search", 1);
|
||||
// TODO(linxj): when to use openmp
|
||||
if (nq <= 4) {
|
||||
GetNeighbors(query, resset[0], nsg, ¶ms);
|
||||
} else {
|
||||
//#pragma omp parallel for schedule(dynamic, 50)
|
||||
#pragma omp parallel for
|
||||
for (unsigned int i = 0; i < nq; ++i) {
|
||||
// TODO(linxj): when to use openmp
|
||||
auto single_query = query + i * dim;
|
||||
GetNeighbors(single_query, resset[i], nsg, ¶ms);
|
||||
}
|
||||
}
|
||||
rc.ElapseFromBegin("cost");
|
||||
|
||||
rc.RecordSection("search");
|
||||
for (unsigned int i = 0; i < nq; ++i) {
|
||||
for (unsigned int j = 0; j < k; ++j) {
|
||||
// ids[i * k + j] = resset[i][j].id;
|
||||
|
||||
// Fix(linxj): bug, reset[i][j] out of range
|
||||
ids[i * k + j] = ids_[resset[i][j].id];
|
||||
dist[i * k + j] = resset[i][j].distance;
|
||||
int64_t var = resset[i].size() - k;
|
||||
if (var >= 0) {
|
||||
for (unsigned int j = 0; j < k; ++j) {
|
||||
ids[i * k + j] = ids_[resset[i][j].id];
|
||||
dist[i * k + j] = resset[i][j].distance;
|
||||
}
|
||||
} else {
|
||||
for (unsigned int j = 0; j < resset[i].size(); ++j) {
|
||||
ids[i * k + j] = ids_[resset[i][j].id];
|
||||
dist[i * k + j] = resset[i][j].distance;
|
||||
}
|
||||
for (unsigned int j = resset[i].size(); j < k; ++j) {
|
||||
ids[i * k + j] = -1;
|
||||
dist[i * k + j] = -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
rc.RecordSection("merge");
|
||||
|
||||
//>> Debug: test single insert
|
||||
// int x_0 = resset[0].size();
|
||||
|
||||
@ -15,6 +15,8 @@
|
||||
// specific language governing permissions and limitations
|
||||
// under the License.
|
||||
|
||||
#define USE_FAISS_V0_2_1 0
|
||||
|
||||
#include <gtest/gtest.h>
|
||||
|
||||
#include <hdf5.h>
|
||||
@ -26,21 +28,6 @@
|
||||
#include <cstdio>
|
||||
#include <vector>
|
||||
|
||||
#define USE_FAISS_V1_5_3 0
|
||||
|
||||
#if USE_FAISS_V1_5_3
|
||||
#include <faiss/gpu/GpuAutoTune.h>
|
||||
#include <faiss/utils.h>
|
||||
#include <sys/stat.h>
|
||||
#include <cstdlib>
|
||||
#include <cstring>
|
||||
|
||||
#else
|
||||
#include <faiss/gpu/GpuCloner.h>
|
||||
#include <faiss/index_factory.h>
|
||||
#include <faiss/utils/distances.h>
|
||||
#endif
|
||||
|
||||
#include <faiss/AutoTune.h>
|
||||
#include <faiss/Index.h>
|
||||
#include <faiss/IndexIVF.h>
|
||||
@ -48,8 +35,22 @@
|
||||
#include <faiss/gpu/StandardGpuResources.h>
|
||||
#include <faiss/index_io.h>
|
||||
|
||||
#if USE_FAISS_V0_2_1
|
||||
#include <faiss/gpu/GpuAutoTune.h>
|
||||
#include <faiss/utils.h>
|
||||
#include <sys/stat.h>
|
||||
#include <cstdlib>
|
||||
#include <cstring>
|
||||
#else
|
||||
#include <faiss/gpu/GpuCloner.h>
|
||||
#include <faiss/index_factory.h>
|
||||
#include <faiss/utils/distances.h>
|
||||
#endif
|
||||
|
||||
#ifdef CUSTOMIZATION
|
||||
#include <faiss/gpu/GpuIndexIVFSQHybrid.h>
|
||||
#else
|
||||
#include <faiss/gpu/GpuIndexIVF.h>
|
||||
#endif
|
||||
|
||||
/*****************************************************
|
||||
@ -295,10 +296,12 @@ load_base_data(faiss::Index*& index, const std::string& ann_test_name, const std
|
||||
cpu_index = faiss::gpu::index_gpu_to_cpu(gpu_index);
|
||||
delete gpu_index;
|
||||
|
||||
#ifdef CUSTOMIZATION
|
||||
faiss::IndexIVF* cpu_ivf_index = dynamic_cast<faiss::IndexIVF*>(cpu_index);
|
||||
if (cpu_ivf_index != nullptr) {
|
||||
cpu_ivf_index->to_readonly();
|
||||
}
|
||||
#endif
|
||||
|
||||
printf("[%.3f s] Writing index file: %s\n", elapsed() - t0, index_file_name.c_str());
|
||||
faiss::write_index(cpu_index, index_file_name.c_str());
|
||||
@ -374,13 +377,15 @@ test_with_nprobes(const std::string& ann_test_name, const std::string& index_key
|
||||
faiss::Index *gpu_index, *index;
|
||||
if (query_mode != MODE_CPU) {
|
||||
faiss::gpu::GpuClonerOptions option;
|
||||
#ifdef CUSTOMIZATION
|
||||
option.allInGpu = true;
|
||||
|
||||
faiss::IndexComposition index_composition;
|
||||
index_composition.index = cpu_index;
|
||||
index_composition.quantizer = nullptr;
|
||||
|
||||
#endif
|
||||
switch (query_mode) {
|
||||
#ifdef CUSTOMIZATION
|
||||
case MODE_MIX: {
|
||||
index_composition.mode = 1; // 0: all data, 1: copy quantizer, 2: copy data
|
||||
|
||||
@ -403,7 +408,9 @@ test_with_nprobes(const std::string& ann_test_name, const std::string& index_key
|
||||
index = cpu_index;
|
||||
break;
|
||||
}
|
||||
#endif
|
||||
case MODE_GPU:
|
||||
#ifdef CUSTOMIZATION
|
||||
index_composition.mode = 0; // 0: all data, 1: copy quantizer, 2: copy data
|
||||
|
||||
// warm up the transmission
|
||||
@ -412,6 +419,14 @@ test_with_nprobes(const std::string& ann_test_name, const std::string& index_key
|
||||
|
||||
copy_time = elapsed();
|
||||
gpu_index = faiss::gpu::index_cpu_to_gpu(&res, GPU_DEVICE_IDX, &index_composition, &option);
|
||||
#else
|
||||
// warm up the transmission
|
||||
gpu_index = faiss::gpu::index_cpu_to_gpu(&res, GPU_DEVICE_IDX, cpu_index, &option);
|
||||
delete gpu_index;
|
||||
|
||||
copy_time = elapsed();
|
||||
gpu_index = faiss::gpu::index_cpu_to_gpu(&res, GPU_DEVICE_IDX, cpu_index, &option);
|
||||
#endif
|
||||
copy_time = elapsed() - copy_time;
|
||||
printf("[%.3f s] Copy data completed, cost %f s\n", elapsed() - t0, copy_time);
|
||||
|
||||
|
||||
@ -59,9 +59,9 @@ print_banner() {
|
||||
#endif
|
||||
<< " library." << std::endl;
|
||||
#ifdef MILVUS_CPU_VERSION
|
||||
std::cout << "You are using Milvus CPU version" << std::endl;
|
||||
std::cout << "You are using Milvus CPU edition" << std::endl;
|
||||
#else
|
||||
std::cout << "You are using Milvus GPU version" << std::endl;
|
||||
std::cout << "You are using Milvus GPU edition" << std::endl;
|
||||
#endif
|
||||
std::cout << std::endl;
|
||||
}
|
||||
|
||||
@ -25,11 +25,13 @@ namespace scheduler {
|
||||
|
||||
void
|
||||
BuildIndexPass::Init() {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
server::Config& config = server::Config::GetInstance();
|
||||
Status s = config.GetGpuResourceConfigBuildIndexResources(build_gpu_ids_);
|
||||
if (!s.ok()) {
|
||||
throw;
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
bool
|
||||
|
||||
@ -29,6 +29,7 @@ namespace scheduler {
|
||||
|
||||
void
|
||||
FaissFlatPass::Init() {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
server::Config& config = server::Config::GetInstance();
|
||||
Status s = config.GetEngineConfigGpuSearchThreshold(threshold_);
|
||||
if (!s.ok()) {
|
||||
@ -38,6 +39,7 @@ FaissFlatPass::Init() {
|
||||
if (!s.ok()) {
|
||||
throw;
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
bool
|
||||
|
||||
@ -29,6 +29,7 @@ namespace scheduler {
|
||||
|
||||
void
|
||||
FaissIVFFlatPass::Init() {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
server::Config& config = server::Config::GetInstance();
|
||||
Status s = config.GetEngineConfigGpuSearchThreshold(threshold_);
|
||||
if (!s.ok()) {
|
||||
@ -38,6 +39,7 @@ FaissIVFFlatPass::Init() {
|
||||
if (!s.ok()) {
|
||||
throw;
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
bool
|
||||
|
||||
@ -29,12 +29,14 @@ namespace scheduler {
|
||||
|
||||
void
|
||||
FaissIVFSQ8HPass::Init() {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
server::Config& config = server::Config::GetInstance();
|
||||
Status s = config.GetEngineConfigGpuSearchThreshold(threshold_);
|
||||
if (!s.ok()) {
|
||||
threshold_ = std::numeric_limits<int64_t>::max();
|
||||
}
|
||||
s = config.GetGpuResourceConfigSearchResources(gpus);
|
||||
#endif
|
||||
}
|
||||
|
||||
bool
|
||||
|
||||
@ -29,6 +29,7 @@ namespace scheduler {
|
||||
|
||||
void
|
||||
FaissIVFSQ8Pass::Init() {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
server::Config& config = server::Config::GetInstance();
|
||||
Status s = config.GetEngineConfigGpuSearchThreshold(threshold_);
|
||||
if (!s.ok()) {
|
||||
@ -38,6 +39,7 @@ FaissIVFSQ8Pass::Init() {
|
||||
if (!s.ok()) {
|
||||
throw;
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
bool
|
||||
|
||||
@ -99,6 +99,12 @@ Utils::IndexTypeName(const milvus::IndexType& index_type) {
|
||||
return "NSG";
|
||||
case milvus::IndexType::IVFSQ8H:
|
||||
return "IVFSQ8H";
|
||||
case milvus::IndexType::IVFPQ:
|
||||
return "IVFPQ";
|
||||
case milvus::IndexType::SPTAGKDT:
|
||||
return "SPTAGKDT";
|
||||
case milvus::IndexType::SPTAGBKT:
|
||||
return "SPTAGBKT";
|
||||
default:
|
||||
return "Unknown index type";
|
||||
}
|
||||
|
||||
@ -37,6 +37,9 @@ enum class IndexType {
|
||||
IVFSQ8 = 3,
|
||||
NSG = 4,
|
||||
IVFSQ8H = 5,
|
||||
IVFPQ = 6,
|
||||
SPTAGKDT = 7,
|
||||
SPTAGBKT = 8,
|
||||
};
|
||||
|
||||
enum class MetricType {
|
||||
|
||||
@ -182,6 +182,7 @@ Config::ValidateConfig() {
|
||||
return s;
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
int64_t engine_gpu_search_threshold;
|
||||
s = GetEngineConfigGpuSearchThreshold(engine_gpu_search_threshold);
|
||||
if (!s.ok()) {
|
||||
@ -195,7 +196,6 @@ Config::ValidateConfig() {
|
||||
return s;
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
if (gpu_resource_enable) {
|
||||
int64_t resource_cache_capacity;
|
||||
s = GetGpuResourceConfigCacheCapacity(resource_cache_capacity);
|
||||
@ -325,13 +325,13 @@ Config::ResetDefaultConfig() {
|
||||
return s;
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
/* gpu resource config */
|
||||
s = SetEngineConfigGpuSearchThreshold(CONFIG_ENGINE_GPU_SEARCH_THRESHOLD_DEFAULT);
|
||||
if (!s.ok()) {
|
||||
return s;
|
||||
}
|
||||
|
||||
/* gpu resource config */
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
s = SetGpuResourceConfigEnable(CONFIG_GPU_RESOURCE_ENABLE_DEFAULT);
|
||||
if (!s.ok()) {
|
||||
return s;
|
||||
@ -632,6 +632,7 @@ Config::CheckEngineConfigOmpThreadNum(const std::string& value) {
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
Status
|
||||
Config::CheckEngineConfigGpuSearchThreshold(const std::string& value) {
|
||||
if (!ValidationUtil::ValidateStringIsNumber(value).ok()) {
|
||||
@ -761,6 +762,7 @@ Config::CheckGpuResourceConfigBuildIndexResources(const std::vector<std::string>
|
||||
|
||||
return Status::OK();
|
||||
}
|
||||
#endif
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
ConfigNode&
|
||||
@ -981,6 +983,7 @@ Config::GetEngineConfigOmpThreadNum(int64_t& value) {
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
Status
|
||||
Config::GetEngineConfigGpuSearchThreshold(int64_t& value) {
|
||||
std::string str =
|
||||
@ -1097,6 +1100,7 @@ Config::GetGpuResourceConfigBuildIndexResources(std::vector<int64_t>& value) {
|
||||
}
|
||||
return Status::OK();
|
||||
}
|
||||
#endif
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
/* server config */
|
||||
@ -1284,6 +1288,8 @@ Config::SetEngineConfigOmpThreadNum(const std::string& value) {
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
/* gpu resource config */
|
||||
Status
|
||||
Config::SetEngineConfigGpuSearchThreshold(const std::string& value) {
|
||||
Status s = CheckEngineConfigGpuSearchThreshold(value);
|
||||
@ -1294,7 +1300,6 @@ Config::SetEngineConfigGpuSearchThreshold(const std::string& value) {
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
/* gpu resource config */
|
||||
Status
|
||||
Config::SetGpuResourceConfigEnable(const std::string& value) {
|
||||
Status s = CheckGpuResourceConfigEnable(value);
|
||||
@ -1348,6 +1353,7 @@ Config::SetGpuResourceConfigBuildIndexResources(const std::string& value) {
|
||||
SetConfigValueInMem(CONFIG_GPU_RESOURCE, CONFIG_GPU_RESOURCE_BUILD_INDEX_RESOURCES, value);
|
||||
return Status::OK();
|
||||
} // namespace server
|
||||
#endif
|
||||
|
||||
} // namespace server
|
||||
} // namespace milvus
|
||||
|
||||
@ -170,6 +170,8 @@ class Config {
|
||||
CheckEngineConfigUseBlasThreshold(const std::string& value);
|
||||
Status
|
||||
CheckEngineConfigOmpThreadNum(const std::string& value);
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
Status
|
||||
CheckEngineConfigGpuSearchThreshold(const std::string& value);
|
||||
|
||||
@ -184,6 +186,7 @@ class Config {
|
||||
CheckGpuResourceConfigSearchResources(const std::vector<std::string>& value);
|
||||
Status
|
||||
CheckGpuResourceConfigBuildIndexResources(const std::vector<std::string>& value);
|
||||
#endif
|
||||
|
||||
std::string
|
||||
GetConfigStr(const std::string& parent_key, const std::string& child_key, const std::string& default_value = "");
|
||||
@ -239,6 +242,8 @@ class Config {
|
||||
GetEngineConfigUseBlasThreshold(int64_t& value);
|
||||
Status
|
||||
GetEngineConfigOmpThreadNum(int64_t& value);
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
Status
|
||||
GetEngineConfigGpuSearchThreshold(int64_t& value);
|
||||
|
||||
@ -253,6 +258,7 @@ class Config {
|
||||
GetGpuResourceConfigSearchResources(std::vector<int64_t>& value);
|
||||
Status
|
||||
GetGpuResourceConfigBuildIndexResources(std::vector<int64_t>& value);
|
||||
#endif
|
||||
|
||||
public:
|
||||
/* server config */
|
||||
@ -300,6 +306,8 @@ class Config {
|
||||
SetEngineConfigUseBlasThreshold(const std::string& value);
|
||||
Status
|
||||
SetEngineConfigOmpThreadNum(const std::string& value);
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
Status
|
||||
SetEngineConfigGpuSearchThreshold(const std::string& value);
|
||||
|
||||
@ -314,6 +322,7 @@ class Config {
|
||||
SetGpuResourceConfigSearchResources(const std::string& value);
|
||||
Status
|
||||
SetGpuResourceConfigBuildIndexResources(const std::string& value);
|
||||
#endif
|
||||
|
||||
private:
|
||||
std::unordered_map<std::string, std::unordered_map<std::string, std::string>> config_map_;
|
||||
|
||||
@ -183,7 +183,11 @@ Server::Start() {
|
||||
|
||||
// print version information
|
||||
SERVER_LOG_INFO << "Milvus " << BUILD_TYPE << " version: v" << MILVUS_VERSION << ", built at " << BUILD_TIME;
|
||||
|
||||
#ifdef MILVUS_CPU_VERSION
|
||||
SERVER_LOG_INFO << "CPU edition";
|
||||
#else
|
||||
SERVER_LOG_INFO << "GPU edition";
|
||||
#endif
|
||||
server::Metrics::GetInstance().Init();
|
||||
server::SystemInfo::GetInstance().Init();
|
||||
|
||||
|
||||
@ -30,9 +30,13 @@ class StringHelpFunctions {
|
||||
StringHelpFunctions() = default;
|
||||
|
||||
public:
|
||||
// trim blanks from begin and end
|
||||
// " a b c " => "a b c"
|
||||
static void
|
||||
TrimStringBlank(std::string& string);
|
||||
|
||||
// trim quotes from begin and end
|
||||
// "'abc'" => "abc"
|
||||
static void
|
||||
TrimStringQuote(std::string& string, const std::string& qoute);
|
||||
|
||||
@ -46,6 +50,8 @@ class StringHelpFunctions {
|
||||
static void
|
||||
SplitStringByDelimeter(const std::string& str, const std::string& delimeter, std::vector<std::string>& result);
|
||||
|
||||
// merge strings with delimeter
|
||||
// "a", "b", "c" => "a,b,c"
|
||||
static void
|
||||
MergeStringWithDelimeter(const std::vector<std::string>& strs, const std::string& delimeter, std::string& result);
|
||||
|
||||
|
||||
@ -218,10 +218,9 @@ ValidationUtil::ValidateGpuIndex(int32_t gpu_index) {
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
Status
|
||||
ValidationUtil::GetGpuMemory(int32_t gpu_index, size_t& memory) {
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
|
||||
cudaDeviceProp deviceProp;
|
||||
auto cuda_err = cudaGetDeviceProperties(&deviceProp, gpu_index);
|
||||
if (cuda_err) {
|
||||
@ -232,10 +231,9 @@ ValidationUtil::GetGpuMemory(int32_t gpu_index, size_t& memory) {
|
||||
}
|
||||
|
||||
memory = deviceProp.totalGlobalMem;
|
||||
#endif
|
||||
|
||||
return Status::OK();
|
||||
}
|
||||
#endif
|
||||
|
||||
Status
|
||||
ValidationUtil::ValidateIpAddress(const std::string& ip_address) {
|
||||
|
||||
@ -64,8 +64,10 @@ class ValidationUtil {
|
||||
static Status
|
||||
ValidateGpuIndex(int32_t gpu_index);
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
static Status
|
||||
GetGpuMemory(int32_t gpu_index, size_t& memory);
|
||||
#endif
|
||||
|
||||
static Status
|
||||
ValidateIpAddress(const std::string& ip_address);
|
||||
|
||||
@ -22,6 +22,7 @@
|
||||
|
||||
#include <cmath>
|
||||
#include <memory>
|
||||
#include <vector>
|
||||
|
||||
// TODO(lxj): add conf checker
|
||||
|
||||
@ -129,17 +130,35 @@ IVFPQConfAdapter::Match(const TempMetaConf& metaconf) {
|
||||
conf->metric_type = metaconf.metric_type;
|
||||
conf->gpu_id = metaconf.gpu_id;
|
||||
conf->nbits = 8;
|
||||
|
||||
if (!(conf->d % 4))
|
||||
conf->m = conf->d / 4; // compression radio = 16
|
||||
else if (!(conf->d % 2))
|
||||
conf->m = conf->d / 2; // compression radio = 8
|
||||
else if (!(conf->d % 3))
|
||||
conf->m = conf->d / 3; // compression radio = 12
|
||||
else
|
||||
conf->m = conf->d; // same as SQ8, compression radio = 4
|
||||
|
||||
MatchBase(conf);
|
||||
|
||||
/*
|
||||
* Faiss 1.6
|
||||
* Only 1, 2, 3, 4, 6, 8, 10, 12, 16, 20, 24, 28, 32 dims per sub-quantizer are currently supporte with
|
||||
* no precomputed codes. Precomputed codes supports any number of dimensions, but will involve memory overheads.
|
||||
*/
|
||||
static std::vector<int64_t> support_dim_per_subquantizer{32, 28, 24, 20, 16, 12, 10, 8, 6, 4, 3, 2, 1};
|
||||
static std::vector<int64_t> support_subquantizer{96, 64, 56, 48, 40, 32, 28, 24, 20, 16, 12, 8, 4, 3, 2, 1};
|
||||
std::vector<int64_t> resset;
|
||||
for (const auto& dimperquantizer : support_dim_per_subquantizer) {
|
||||
if (!(conf->d % dimperquantizer)) {
|
||||
auto subquantzier_num = conf->d / dimperquantizer;
|
||||
auto finder = std::find(support_subquantizer.begin(), support_subquantizer.end(), subquantzier_num);
|
||||
if (finder != support_subquantizer.end()) {
|
||||
resset.push_back(subquantzier_num);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (resset.empty()) {
|
||||
// todo(linxj): throw exception here.
|
||||
return nullptr;
|
||||
}
|
||||
static int64_t compression_level = 1; // 1:low, 2:high
|
||||
if (compression_level == 1) {
|
||||
conf->m = resset[int(resset.size() / 2)];
|
||||
WRAPPER_LOG_DEBUG << "PQ m = " << conf->m << ", compression radio = " << conf->d / conf->m * 4;
|
||||
}
|
||||
return conf;
|
||||
}
|
||||
|
||||
|
||||
@ -132,8 +132,8 @@ BaseTest::SetUp() {
|
||||
void
|
||||
BaseTest::TearDown() {
|
||||
milvus::cache::CpuCacheMgr::GetInstance()->ClearCache();
|
||||
milvus::cache::GpuCacheMgr::GetInstance(0)->ClearCache();
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
milvus::cache::GpuCacheMgr::GetInstance(0)->ClearCache();
|
||||
knowhere::FaissGpuResourceMgr::GetInstance().Free();
|
||||
#endif
|
||||
}
|
||||
|
||||
@ -98,24 +98,25 @@ class SchedulerTest : public testing::Test {
|
||||
protected:
|
||||
void
|
||||
SetUp() override {
|
||||
res_mgr_ = std::make_shared<ResourceMgr>();
|
||||
ResourcePtr disk = ResourceFactory::Create("disk", "DISK", 0, true, false);
|
||||
ResourcePtr cpu = ResourceFactory::Create("cpu", "CPU", 0, true, false);
|
||||
disk_resource_ = res_mgr_->Add(std::move(disk));
|
||||
cpu_resource_ = res_mgr_->Add(std::move(cpu));
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
constexpr int64_t cache_cap = 1024 * 1024 * 1024;
|
||||
cache::GpuCacheMgr::GetInstance(0)->SetCapacity(cache_cap);
|
||||
cache::GpuCacheMgr::GetInstance(1)->SetCapacity(cache_cap);
|
||||
|
||||
ResourcePtr disk = ResourceFactory::Create("disk", "DISK", 0, true, false);
|
||||
ResourcePtr cpu = ResourceFactory::Create("cpu", "CPU", 0, true, false);
|
||||
ResourcePtr gpu_0 = ResourceFactory::Create("gpu0", "GPU", 0);
|
||||
ResourcePtr gpu_1 = ResourceFactory::Create("gpu1", "GPU", 1);
|
||||
|
||||
res_mgr_ = std::make_shared<ResourceMgr>();
|
||||
disk_resource_ = res_mgr_->Add(std::move(disk));
|
||||
cpu_resource_ = res_mgr_->Add(std::move(cpu));
|
||||
gpu_resource_0_ = res_mgr_->Add(std::move(gpu_0));
|
||||
gpu_resource_1_ = res_mgr_->Add(std::move(gpu_1));
|
||||
|
||||
auto PCIE = Connection("IO", 11000.0);
|
||||
res_mgr_->Connect("cpu", "gpu0", PCIE);
|
||||
res_mgr_->Connect("cpu", "gpu1", PCIE);
|
||||
#endif
|
||||
|
||||
scheduler_ = std::make_shared<Scheduler>(res_mgr_);
|
||||
|
||||
@ -138,17 +139,6 @@ class SchedulerTest : public testing::Test {
|
||||
std::shared_ptr<Scheduler> scheduler_;
|
||||
};
|
||||
|
||||
void
|
||||
insert_dummy_index_into_gpu_cache(uint64_t device_id) {
|
||||
MockVecIndex* mock_index = new MockVecIndex();
|
||||
mock_index->ntotal_ = 1000;
|
||||
engine::VecIndexPtr index(mock_index);
|
||||
|
||||
cache::DataObjPtr obj = std::static_pointer_cast<cache::DataObj>(index);
|
||||
|
||||
cache::GpuCacheMgr::GetInstance(device_id)->InsertItem("location", obj);
|
||||
}
|
||||
|
||||
class SchedulerTest2 : public testing::Test {
|
||||
protected:
|
||||
void
|
||||
@ -157,16 +147,13 @@ class SchedulerTest2 : public testing::Test {
|
||||
ResourcePtr cpu0 = ResourceFactory::Create("cpu0", "CPU", 0, true, false);
|
||||
ResourcePtr cpu1 = ResourceFactory::Create("cpu1", "CPU", 1, true, false);
|
||||
ResourcePtr cpu2 = ResourceFactory::Create("cpu2", "CPU", 2, true, false);
|
||||
ResourcePtr gpu0 = ResourceFactory::Create("gpu0", "GPU", 0, true, true);
|
||||
ResourcePtr gpu1 = ResourceFactory::Create("gpu1", "GPU", 1, true, true);
|
||||
|
||||
res_mgr_ = std::make_shared<ResourceMgr>();
|
||||
disk_ = res_mgr_->Add(std::move(disk));
|
||||
cpu_0_ = res_mgr_->Add(std::move(cpu0));
|
||||
cpu_1_ = res_mgr_->Add(std::move(cpu1));
|
||||
cpu_2_ = res_mgr_->Add(std::move(cpu2));
|
||||
gpu_0_ = res_mgr_->Add(std::move(gpu0));
|
||||
gpu_1_ = res_mgr_->Add(std::move(gpu1));
|
||||
|
||||
auto IO = Connection("IO", 5.0);
|
||||
auto PCIE1 = Connection("PCIE", 11.0);
|
||||
auto PCIE2 = Connection("PCIE", 20.0);
|
||||
@ -174,8 +161,15 @@ class SchedulerTest2 : public testing::Test {
|
||||
res_mgr_->Connect("cpu0", "cpu1", IO);
|
||||
res_mgr_->Connect("cpu1", "cpu2", IO);
|
||||
res_mgr_->Connect("cpu0", "cpu2", IO);
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
ResourcePtr gpu0 = ResourceFactory::Create("gpu0", "GPU", 0, true, true);
|
||||
ResourcePtr gpu1 = ResourceFactory::Create("gpu1", "GPU", 1, true, true);
|
||||
gpu_0_ = res_mgr_->Add(std::move(gpu0));
|
||||
gpu_1_ = res_mgr_->Add(std::move(gpu1));
|
||||
res_mgr_->Connect("cpu1", "gpu0", PCIE1);
|
||||
res_mgr_->Connect("cpu2", "gpu1", PCIE2);
|
||||
#endif
|
||||
|
||||
scheduler_ = std::make_shared<Scheduler>(res_mgr_);
|
||||
|
||||
|
||||
@ -175,6 +175,7 @@ TEST(CacheTest, CPU_CACHE_TEST) {
|
||||
cpu_mgr->PrintInfo();
|
||||
}
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
TEST(CacheTest, GPU_CACHE_TEST) {
|
||||
auto gpu_mgr = milvus::cache::GpuCacheMgr::GetInstance(0);
|
||||
|
||||
@ -202,6 +203,7 @@ TEST(CacheTest, GPU_CACHE_TEST) {
|
||||
gpu_mgr->ClearCache();
|
||||
ASSERT_EQ(gpu_mgr->ItemCount(), 0);
|
||||
}
|
||||
#endif
|
||||
|
||||
TEST(CacheTest, INVALID_TEST) {
|
||||
{
|
||||
|
||||
@ -25,6 +25,8 @@
|
||||
#include "utils/StringHelpFunctions.h"
|
||||
#include "utils/ValidationUtil.h"
|
||||
|
||||
#include <limits>
|
||||
|
||||
namespace {
|
||||
|
||||
static constexpr uint64_t KB = 1024;
|
||||
@ -63,9 +65,21 @@ TEST_F(ConfigTest, CONFIG_TEST) {
|
||||
int64_t port = server_config.GetInt64Value("port");
|
||||
ASSERT_NE(port, 0);
|
||||
|
||||
server_config.SetValue("test", "2.5");
|
||||
double test = server_config.GetDoubleValue("test");
|
||||
ASSERT_EQ(test, 2.5);
|
||||
server_config.SetValue("float_test", "2.5");
|
||||
double dbl = server_config.GetDoubleValue("float_test");
|
||||
ASSERT_LE(abs(dbl - 2.5), std::numeric_limits<double>::epsilon());
|
||||
float flt = server_config.GetFloatValue("float_test");
|
||||
ASSERT_LE(abs(flt - 2.5), std::numeric_limits<float>::epsilon());
|
||||
|
||||
server_config.SetValue("bool_test", "true");
|
||||
bool blt = server_config.GetBoolValue("bool_test");
|
||||
ASSERT_TRUE(blt);
|
||||
|
||||
server_config.SetValue("int_test", "34");
|
||||
int32_t it32 = server_config.GetInt32Value("int_test");
|
||||
ASSERT_EQ(it32, 34);
|
||||
int64_t it64 = server_config.GetInt64Value("int_test");
|
||||
ASSERT_EQ(it64, 34);
|
||||
|
||||
milvus::server::ConfigNode fake;
|
||||
server_config.AddChild("fake", fake);
|
||||
@ -236,6 +250,7 @@ TEST_F(ConfigTest, SERVER_CONFIG_VALID_TEST) {
|
||||
ASSERT_TRUE(s.ok());
|
||||
ASSERT_TRUE(int64_val == engine_omp_thread_num);
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
int64_t engine_gpu_search_threshold = 800;
|
||||
s = config.SetEngineConfigGpuSearchThreshold(std::to_string(engine_gpu_search_threshold));
|
||||
ASSERT_TRUE(s.ok());
|
||||
@ -251,7 +266,6 @@ TEST_F(ConfigTest, SERVER_CONFIG_VALID_TEST) {
|
||||
ASSERT_TRUE(s.ok());
|
||||
ASSERT_TRUE(bool_val == resource_enable_gpu);
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
int64_t gpu_cache_capacity = 1;
|
||||
s = config.SetGpuResourceConfigCacheCapacity(std::to_string(gpu_cache_capacity));
|
||||
ASSERT_TRUE(s.ok());
|
||||
@ -389,6 +403,7 @@ TEST_F(ConfigTest, SERVER_CONFIG_INVALID_TEST) {
|
||||
s = config.SetEngineConfigOmpThreadNum("10000");
|
||||
ASSERT_FALSE(s.ok());
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
s = config.SetEngineConfigGpuSearchThreshold("-1");
|
||||
ASSERT_FALSE(s.ok());
|
||||
|
||||
@ -396,7 +411,6 @@ TEST_F(ConfigTest, SERVER_CONFIG_INVALID_TEST) {
|
||||
s = config.SetGpuResourceConfigEnable("ok");
|
||||
ASSERT_FALSE(s.ok());
|
||||
|
||||
#ifdef MILVUS_GPU_VERSION
|
||||
s = config.SetGpuResourceConfigCacheCapacity("a");
|
||||
ASSERT_FALSE(s.ok());
|
||||
s = config.SetGpuResourceConfigCacheCapacity("128");
|
||||
|
||||
@ -313,6 +313,9 @@ TEST_F(RpcHandlerTest, TABLES_TEST) {
|
||||
std::vector<std::vector<float>> record_array;
|
||||
BuildVectors(0, VECTOR_COUNT, record_array);
|
||||
::milvus::grpc::VectorIds vector_ids;
|
||||
for (int64_t i = 0; i < VECTOR_COUNT; i++) {
|
||||
vector_ids.add_vector_id_array(i);
|
||||
}
|
||||
// Insert vectors
|
||||
// test invalid table name
|
||||
handler->Insert(&context, &request, &vector_ids);
|
||||
|
||||
@ -120,7 +120,13 @@ TEST(UtilTest, STRINGFUNCTIONS_TEST) {
|
||||
milvus::server::StringHelpFunctions::SplitStringByDelimeter(str, ",", result);
|
||||
ASSERT_EQ(result.size(), 3UL);
|
||||
|
||||
std::string merge_str;
|
||||
milvus::server::StringHelpFunctions::MergeStringWithDelimeter(result, ",", merge_str);
|
||||
ASSERT_EQ(merge_str, "a,b,c");
|
||||
result.clear();
|
||||
milvus::server::StringHelpFunctions::MergeStringWithDelimeter(result, ",", merge_str);
|
||||
ASSERT_TRUE(merge_str.empty());
|
||||
|
||||
auto status = milvus::server::StringHelpFunctions::SplitStringByQuote(str, ",", "\"", result);
|
||||
ASSERT_TRUE(status.ok());
|
||||
ASSERT_EQ(result.size(), 3UL);
|
||||
@ -211,6 +217,11 @@ TEST(UtilTest, STATUS_TEST) {
|
||||
str = status.ToString();
|
||||
ASSERT_FALSE(str.empty());
|
||||
|
||||
status = milvus::Status(milvus::DB_INVALID_PATH, "mistake");
|
||||
ASSERT_EQ(status.code(), milvus::DB_INVALID_PATH);
|
||||
str = status.ToString();
|
||||
ASSERT_FALSE(str.empty());
|
||||
|
||||
status = milvus::Status(milvus::DB_META_TRANSACTION_FAILED, "mistake");
|
||||
ASSERT_EQ(status.code(), milvus::DB_META_TRANSACTION_FAILED);
|
||||
str = status.ToString();
|
||||
@ -261,6 +272,10 @@ TEST(ValidationUtilTest, VALIDATE_TABLENAME_TEST) {
|
||||
table_name = std::string(10000, 'a');
|
||||
status = milvus::server::ValidationUtil::ValidateTableName(table_name);
|
||||
ASSERT_EQ(status.code(), milvus::SERVER_INVALID_TABLE_NAME);
|
||||
|
||||
table_name = "";
|
||||
status = milvus::server::ValidationUtil::ValidatePartitionName(table_name);
|
||||
ASSERT_EQ(status.code(), milvus::SERVER_INVALID_TABLE_NAME);
|
||||
}
|
||||
|
||||
TEST(ValidationUtilTest, VALIDATE_DIMENSION_TEST) {
|
||||
|
||||
@ -16,25 +16,25 @@
|
||||
|
||||
### 软硬件环境
|
||||
|
||||
操作系统: CentOS Linux release 7.6.1810 (Core)
|
||||
操作系统:CentOS Linux release 7.6.1810 (Core)
|
||||
|
||||
CPU: Intel(R) Xeon(R) CPU E5-2678 v3 @ 2.50GHz
|
||||
CPU:Intel(R) Xeon(R) CPU E5-2678 v3 @ 2.50GHz
|
||||
|
||||
GPU0: GeForce GTX 1080
|
||||
GPU0:GeForce GTX 1080
|
||||
|
||||
GPU1: GeForce GTX 1080
|
||||
GPU1:GeForce GTX 1080
|
||||
|
||||
内存: 503GB
|
||||
内存:503GB
|
||||
|
||||
Docker版本: 18.09
|
||||
Docker版本:18.09
|
||||
|
||||
NVIDIA Driver版本: 430.34
|
||||
NVIDIA Driver版本:430.34
|
||||
|
||||
Milvus版本: 0.5.3
|
||||
Milvus版本:0.5.3
|
||||
|
||||
SDK接口: Python 3.6.8
|
||||
SDK接口:Python 3.6.8
|
||||
|
||||
pymilvus版本: 0.2.5
|
||||
pymilvus版本:0.2.5
|
||||
|
||||
|
||||
|
||||
@ -51,7 +51,7 @@ pymilvus版本: 0.2.5
|
||||
|
||||
### 测试指标
|
||||
|
||||
- Query Elapsed Time: 数据库查询所有向量的时间(以秒计)。影响Query Elapsed Time的变量:
|
||||
- Query Elapsed Time:数据库查询所有向量的时间(以秒计)。影响Query Elapsed Time的变量:
|
||||
|
||||
- nq (被查询向量的数量)
|
||||
|
||||
@ -59,7 +59,7 @@ pymilvus版本: 0.2.5
|
||||
>
|
||||
> 被查询向量的数量nq将按照 [1, 5, 10, 200, 400, 600, 800, 1000]的数量分组。
|
||||
|
||||
- Recall: 实际返回的正确结果占总数之比 . 影响Recall的变量:
|
||||
- Recall:实际返回的正确结果占总数之比。影响Recall的变量:
|
||||
|
||||
- nq (被查询向量的数量)
|
||||
- topk (单条查询中最相似的K个结果)
|
||||
@ -76,7 +76,7 @@ pymilvus版本: 0.2.5
|
||||
|
||||
### 测试环境
|
||||
|
||||
数据集: sift1b-1,000,000,000向量, 128维
|
||||
数据集:sift1b-1,000,000,000向量,128维
|
||||
|
||||
表格属性:
|
||||
|
||||
@ -143,7 +143,7 @@ search_resources: cpu, gpu0
|
||||
| nq=800 | 23.24 |
|
||||
| nq=1000 | 27.41 |
|
||||
|
||||
当nq为1000时,在GPU模式下查询一条128维向量需要耗时约27毫秒。
|
||||
当nq为1000时,在CPU模式下查询一条128维向量需要耗时约27毫秒。
|
||||
|
||||
|
||||
|
||||
|
||||
@ -139,7 +139,7 @@ topk = 100
|
||||
|
||||
**总结**
|
||||
|
||||
当nq小于1200时,查询耗时随nq的增长快速增大;当nq大于1200时,查询耗时的增大则缓慢许多。这是因为gpu_search_threshold这一参数的值被设为1200,当nq<1200时,选择CPU进行操作,否则选择GPU进行操作。与CPU。
|
||||
当nq小于1200时,查询耗时随nq的增长快速增大;当nq大于1200时,查询耗时的增大则缓慢许多。这是因为gpu_search_threshold这一参数的值被设为1200,当nq小于1200时,选择CPU进行操作,否则选择GPU进行操作。
|
||||
|
||||
在GPU模式下的查询耗时由两部分组成:(1)索引从CPU到GPU的拷贝时间;(2)所有分桶的查询时间。当nq小于500时,索引从CPU到GPU 的拷贝时间无法被有效均摊,此时CPU模式时一个更优的选择;当nq大于500时,选择GPU模式更合理。和CPU相比,GPU具有更多的核数和更强的算力。当nq较大时,GPU在计算上的优势能被更好地被体现。
|
||||
|
||||
|
||||
@ -1208,9 +1208,9 @@ class TestCreateIndexParamsInvalid(object):
|
||||
nlist = index_params["nlist"]
|
||||
logging.getLogger().info(index_params)
|
||||
status, ids = connect.add_vectors(table, vectors)
|
||||
# if not isinstance(index_type, int) or not isinstance(nlist, int):
|
||||
with pytest.raises(Exception) as e:
|
||||
if (not index_type) or (not nlist) or (not isinstance(index_type, IndexType)) or (not isinstance(nlist, int)):
|
||||
with pytest.raises(Exception) as e:
|
||||
status = connect.create_index(table, index_params)
|
||||
else:
|
||||
status = connect.create_index(table, index_params)
|
||||
# else:
|
||||
# status = connect.create_index(table, index_params)
|
||||
# assert not status.OK()
|
||||
assert not status.OK()
|
||||
|
||||
@ -54,12 +54,12 @@ def gen_long_str(num):
|
||||
|
||||
def gen_invalid_ips():
|
||||
ips = [
|
||||
"255.0.0.0",
|
||||
"255.255.0.0",
|
||||
"255.255.255.0",
|
||||
"255.255.255.255",
|
||||
# "255.0.0.0",
|
||||
# "255.255.0.0",
|
||||
# "255.255.255.0",
|
||||
# "255.255.255.255",
|
||||
"127.0.0",
|
||||
"123.0.0.2",
|
||||
# "123.0.0.2",
|
||||
"12-s",
|
||||
" ",
|
||||
"12 s",
|
||||
@ -114,12 +114,12 @@ def gen_invalid_uris():
|
||||
|
||||
# invalid ip
|
||||
"tcp:// :%s" % port,
|
||||
"tcp://123.0.0.1:%s" % port,
|
||||
# "tcp://123.0.0.1:%s" % port,
|
||||
"tcp://127.0.0:%s" % port,
|
||||
"tcp://255.0.0.0:%s" % port,
|
||||
"tcp://255.255.0.0:%s" % port,
|
||||
"tcp://255.255.255.0:%s" % port,
|
||||
"tcp://255.255.255.255:%s" % port,
|
||||
# "tcp://255.255.0.0:%s" % port,
|
||||
# "tcp://255.255.255.0:%s" % port,
|
||||
# "tcp://255.255.255.255:%s" % port,
|
||||
"tcp://\n:%s" % port,
|
||||
|
||||
]
|
||||
@ -263,9 +263,7 @@ def gen_invalid_index_types():
|
||||
|
||||
def gen_invalid_nlists():
|
||||
nlists = [
|
||||
0,
|
||||
-1,
|
||||
1000000000000001,
|
||||
# None,
|
||||
[1,2,3],
|
||||
(1,2),
|
||||
@ -549,4 +547,4 @@ if __name__ == "__main__":
|
||||
p.join()
|
||||
time.sleep(3)
|
||||
status, count = milvus.get_table_row_count(table)
|
||||
assert count == process_num * loop_num
|
||||
assert count == process_num * loop_num
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user