Merge remote-tracking branch 'upstream/branch-0.5.0' into branch-0.5.0-yk

Former-commit-id: dd16511856cac95f72d9dbb07301f5c11ffebeb9

ci/jenkinsfile/milvus_build.groovy

 container('milvus-build-env') {
-    timeout(time: 40, unit: 'MINUTES') {
+    timeout(time: 120, unit: 'MINUTES') {
         gitlabCommitStatus(name: 'Build Engine') {
             dir ("milvus_engine") {
                 try {

ci/jenkinsfile/milvus_build_no_ut.groovy

 container('milvus-build-env') {
-    timeout(time: 40, unit: 'MINUTES') {
+    timeout(time: 120, unit: 'MINUTES') {
         gitlabCommitStatus(name: 'Build Engine') {
             dir ("milvus_engine") {
                 try {

core/CHANGELOG.md

@@ -20,6 +20,7 @@ Please mark all change in change log and use the ticket from JIRA.
 - MS-647 - [monitor] grafana display average cpu-temp
 - MS-644 - Search crashed with index-type: flat
 - MS-624 - Search vectors failed if time ranges long enough
+- MS-652 - IVFSQH quantization double free
 
 ## Improvement
 - MS-552 - Add and change the easylogging library

core/cmake/ThirdPartyPackages.cmake

@@ -158,6 +158,10 @@ if(USE_JFROG_CACHE STREQUAL "ON")
     endif()
 
     set(THIRDPARTY_PACKAGE_CACHE "${THIRDPARTY_DIR}/cache")
+    if(NOT EXISTS ${THIRDPARTY_PACKAGE_CACHE})
+        message(STATUS "Will create cached directory: ${THIRDPARTY_PACKAGE_CACHE}")
+        file(MAKE_DIRECTORY ${THIRDPARTY_PACKAGE_CACHE})
+    endif()
 endif()
 
 macro(resolve_dependency DEPENDENCY_NAME)
@@ -324,8 +328,8 @@ if(DEFINED ENV{MILVUS_SQLITE_ORM_URL})
     set(SQLITE_ORM_SOURCE_URL "$ENV{MILVUS_SQLITE_ORM_URL}")
 else()
     set(SQLITE_ORM_SOURCE_URL
-            "http://192.168.1.105:6060/Test/sqlite_orm/-/archive/master/sqlite_orm-master.zip")
-#            "https://github.com/fnc12/sqlite_orm/archive/${SQLITE_ORM_VERSION}.zip")
+#            "http://192.168.1.105:6060/Test/sqlite_orm/-/archive/master/sqlite_orm-master.zip")
+            "https://github.com/fnc12/sqlite_orm/archive/${SQLITE_ORM_VERSION}.zip")
 endif()
 set(SQLITE_ORM_MD5 "ba9a405a8a1421c093aa8ce988ff8598")
 
@@ -370,9 +374,9 @@ if(DEFINED ENV{MILVUS_GRPC_URL})
     set(GRPC_SOURCE_URL "$ENV{MILVUS_GRPC_URL}")
 else()
     set(GRPC_SOURCE_URL
-            "http://git.zilliz.tech/kun.yu/grpc/-/archive/master/grpc-master.tar.gz")
+            "https://github.com/youny626/grpc-milvus/archive/${GRPC_VERSION}.zip")
 endif()
-set(GRPC_MD5 "7ec59ad54c85a12dcbbfede09bf413a9")
+set(GRPC_MD5 "0362ba219f59432c530070b5f5c3df73")
 
 
 # ----------------------------------------------------------------------

core/src/index/cmake/ThirdPartyPackagesCore.cmake

@@ -125,6 +125,10 @@ endif()
 if(USE_JFROG_CACHE STREQUAL "ON")
     set(JFROG_ARTFACTORY_CACHE_URL "${JFROG_ARTFACTORY_URL}/milvus/thirdparty/cache/${CMAKE_OS_NAME}/${KNOWHERE_BUILD_ARCH}/${BUILD_TYPE}")
     set(THIRDPARTY_PACKAGE_CACHE "${THIRDPARTY_DIR}/cache")
+    if(NOT EXISTS ${THIRDPARTY_PACKAGE_CACHE})
+        message(STATUS "Will create cached directory: ${THIRDPARTY_PACKAGE_CACHE}")
+        file(MAKE_DIRECTORY ${THIRDPARTY_PACKAGE_CACHE})
+    endif()
 endif()
 
 macro(resolve_dependency DEPENDENCY_NAME)

core/src/index/knowhere/knowhere/index/vector_index/IndexIVFSQHybrid.cpp

@@ -27,6 +27,7 @@
 namespace knowhere {
 
 #ifdef CUSTOMIZATION
+
 IndexModelPtr
 IVFSQHybrid::Train(const DatasetPtr& dataset, const Config& config) {
     auto build_cfg = std::dynamic_pointer_cast<IVFSQCfg>(config);
@@ -79,8 +80,24 @@ IVFSQHybrid::CopyGpuToCpu(const Config& config) {
 
 VectorIndexPtr
 IVFSQHybrid::CopyCpuToGpu(const int64_t& device_id, const Config& config) {
-    auto p = CopyCpuToGpuWithQuantizer(device_id, config);
-    return p.first;
+    if (auto res = FaissGpuResourceMgr::GetInstance().GetRes(device_id)) {
+        ResScope rs(res, device_id, false);
+        faiss::gpu::GpuClonerOptions option;
+        option.allInGpu = true;
+
+        faiss::IndexComposition index_composition;
+        index_composition.index = index_.get();
+        index_composition.quantizer = nullptr;
+        index_composition.mode = 0;  // copy all
+
+        auto gpu_index = faiss::gpu::index_cpu_to_gpu(res->faiss_res.get(), device_id, &index_composition, &option);
+
+        std::shared_ptr<faiss::Index> device_index = std::shared_ptr<faiss::Index>(gpu_index);
+        auto new_idx = std::make_shared<IVFSQHybrid>(device_index, device_id, res);
+        return new_idx;
+    } else {
+        KNOWHERE_THROW_MSG("CopyCpuToGpu Error, can't get gpu_resource");
+    }
 }
 
 void

core/src/scheduler/task/SearchTask.cpp

@@ -315,10 +315,10 @@ XSearchTask::MergeTopkArray(std::vector<int64_t>& tar_ids, std::vector<float>& t
         return;
     }
 
-    std::vector<int64_t> id_buf(nq * topk, -1);
-    std::vector<float> dist_buf(nq * topk, 0.0);
-
     uint64_t output_k = std::min(topk, tar_input_k + src_input_k);
+    std::vector<int64_t> id_buf(nq * output_k, -1);
+    std::vector<float> dist_buf(nq * output_k, 0.0);
+
     uint64_t buf_k, src_k, tar_k;
     uint64_t src_idx, tar_idx, buf_idx;
     uint64_t src_input_k_multi_i, tar_input_k_multi_i, buf_k_multi_i;
@@ -349,6 +349,7 @@ XSearchTask::MergeTopkArray(std::vector<int64_t>& tar_ids, std::vector<float>& t
             if (src_k < src_input_k) {
                 while (buf_k < output_k && src_k < src_input_k) {
                     src_idx = src_input_k_multi_i + src_k;
+                    buf_idx = buf_k_multi_i + buf_k;
                     id_buf[buf_idx] = src_ids[src_idx];
                     dist_buf[buf_idx] = src_distance[src_idx];
                     src_k++;
@@ -356,6 +357,8 @@ XSearchTask::MergeTopkArray(std::vector<int64_t>& tar_ids, std::vector<float>& t
                 }
             } else {
                 while (buf_k < output_k && tar_k < tar_input_k) {
+                    tar_idx = tar_input_k_multi_i + tar_k;
+                    buf_idx = buf_k_multi_i + buf_k;
                     id_buf[buf_idx] = tar_ids[tar_idx];
                     dist_buf[buf_idx] = tar_distance[tar_idx];
                     tar_k++;

core/unittest/db/test_search.cpp

@@ -110,87 +110,102 @@ CheckTopkResult(const std::vector<int64_t>& input_ids_1,
 
 } // namespace
 
-TEST(DBSearchTest, TOPK_TEST) {
-    uint64_t NQ = 15;
-    uint64_t TOP_K = 64;
-    bool ascending;
+void MergeTopkToResultSetTest(uint64_t topk_1, uint64_t topk_2, uint64_t nq, uint64_t topk, bool ascending) {
     std::vector<int64_t> ids1, ids2;
     std::vector<float> dist1, dist2;
     ms::ResultSet result;
+    BuildResult(ids1, dist1, topk_1, nq, ascending);
+    BuildResult(ids2, dist2, topk_2, nq, ascending);
+    ms::XSearchTask::MergeTopkToResultSet(ids1, dist1, topk_1, nq, topk, ascending, result);
+    ms::XSearchTask::MergeTopkToResultSet(ids2, dist2, topk_2, nq, topk, ascending, result);
+    CheckTopkResult(ids1, dist1, ids2, dist2, topk, nq, ascending, result);
+}
+
+TEST(DBSearchTest, MERGE_RESULT_SET_TEST) {
+    uint64_t NQ = 15;
+    uint64_t TOP_K = 64;
 
     /* test1, id1/dist1 valid, id2/dist2 empty */
-    ascending = true;
-    BuildResult(ids1, dist1, TOP_K, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids1, dist1, TOP_K, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
+    MergeTopkToResultSetTest(TOP_K, 0, NQ, TOP_K, true);
+    MergeTopkToResultSetTest(TOP_K, 0, NQ, TOP_K, false);
 
     /* test2, id1/dist1 valid, id2/dist2 valid */
-    BuildResult(ids2, dist2, TOP_K, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids2, dist2, TOP_K, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
+    MergeTopkToResultSetTest(TOP_K, TOP_K, NQ, TOP_K, true);
+    MergeTopkToResultSetTest(TOP_K, TOP_K, NQ, TOP_K, false);
 
     /* test3, id1/dist1 small topk */
-    ids1.clear();
-    dist1.clear();
-    result.clear();
-    BuildResult(ids1, dist1, TOP_K/2, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids1, dist1, TOP_K/2, NQ, TOP_K, ascending, result);
-    ms::XSearchTask::MergeTopkToResultSet(ids2, dist2, TOP_K, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
+    MergeTopkToResultSetTest(TOP_K/2, TOP_K, NQ, TOP_K, true);
+    MergeTopkToResultSetTest(TOP_K/2, TOP_K, NQ, TOP_K, false);
 
     /* test4, id1/dist1 small topk, id2/dist2 small topk */
-    ids2.clear();
-    dist2.clear();
-    result.clear();
-    BuildResult(ids2, dist2, TOP_K/3, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids1, dist1, TOP_K/2, NQ, TOP_K, ascending, result);
-    ms::XSearchTask::MergeTopkToResultSet(ids2, dist2, TOP_K/3, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
-
-/////////////////////////////////////////////////////////////////////////////////////////
-    ascending = false;
-    ids1.clear();
-    dist1.clear();
-    ids2.clear();
-    dist2.clear();
-    result.clear();
+    MergeTopkToResultSetTest(TOP_K/2, TOP_K/3, NQ, TOP_K, true);
+    MergeTopkToResultSetTest(TOP_K/2, TOP_K/3, NQ, TOP_K, false);
+}
+
+void MergeTopkArrayTest(uint64_t topk_1, uint64_t topk_2, uint64_t nq, uint64_t topk, bool ascending) {
+    std::vector<int64_t> ids1, ids2;
+    std::vector<float> dist1, dist2;
+    ms::ResultSet result;
+    BuildResult(ids1, dist1, topk_1, nq, ascending);
+    BuildResult(ids2, dist2, topk_2, nq, ascending);
+    uint64_t result_topk = std::min(topk, topk_1 + topk_2);
+    ms::XSearchTask::MergeTopkArray(ids1, dist1, topk_1, ids2, dist2, topk_2, nq, topk, ascending);
+    if (ids1.size() != result_topk * nq) {
+        std::cout << ids1.size() << " " << result_topk * nq << std::endl;
+    }
+    ASSERT_TRUE(ids1.size() == result_topk * nq);
+    ASSERT_TRUE(dist1.size() == result_topk * nq);
+    for (uint64_t i = 0; i < nq; i++) {
+        for (uint64_t k = 1; k < result_topk; k++) {
+            if (ascending) {
+                if (dist1[i * result_topk + k] < dist1[i * result_topk + k - 1]) {
+                    std::cout << dist1[i * result_topk + k - 1] << " " << dist1[i * result_topk + k] << std::endl;
+                }
+                ASSERT_TRUE(dist1[i * result_topk + k] >= dist1[i * result_topk + k - 1]);
+            } else {
+                if (dist1[i * result_topk + k] > dist1[i * result_topk + k - 1]) {
+                    std::cout << dist1[i * result_topk + k - 1] << " " << dist1[i * result_topk + k] << std::endl;
+                }
+                ASSERT_TRUE(dist1[i * result_topk + k] <= dist1[i * result_topk + k - 1]);
+            }
+        }
+    }
+}
+
+TEST(DBSearchTest, MERGE_ARRAY_TEST) {
+    uint64_t NQ = 15;
+    uint64_t TOP_K = 64;
 
     /* test1, id1/dist1 valid, id2/dist2 empty */
-    BuildResult(ids1, dist1, TOP_K, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids1, dist1, TOP_K, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
+    MergeTopkArrayTest(TOP_K, 0, NQ, TOP_K, true);
+    MergeTopkArrayTest(TOP_K, 0, NQ, TOP_K, false);
+    MergeTopkArrayTest(0, TOP_K, NQ, TOP_K, true);
+    MergeTopkArrayTest(0, TOP_K, NQ, TOP_K, false);
 
     /* test2, id1/dist1 valid, id2/dist2 valid */
-    BuildResult(ids2, dist2, TOP_K, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids2, dist2, TOP_K, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
+    MergeTopkArrayTest(TOP_K, TOP_K, NQ, TOP_K, true);
+    MergeTopkArrayTest(TOP_K, TOP_K, NQ, TOP_K, false);
 
     /* test3, id1/dist1 small topk */
-    ids1.clear();
-    dist1.clear();
-    result.clear();
-    BuildResult(ids1, dist1, TOP_K/2, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids1, dist1, TOP_K/2, NQ, TOP_K, ascending, result);
-    ms::XSearchTask::MergeTopkToResultSet(ids2, dist2, TOP_K, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
+    MergeTopkArrayTest(TOP_K/2, TOP_K, NQ, TOP_K, true);
+    MergeTopkArrayTest(TOP_K/2, TOP_K, NQ, TOP_K, false);
+    MergeTopkArrayTest(TOP_K, TOP_K/2, NQ, TOP_K, true);
+    MergeTopkArrayTest(TOP_K, TOP_K/2, NQ, TOP_K, false);
 
     /* test4, id1/dist1 small topk, id2/dist2 small topk */
-    ids2.clear();
-    dist2.clear();
-    result.clear();
-    BuildResult(ids2, dist2, TOP_K/3, NQ, ascending);
-    ms::XSearchTask::MergeTopkToResultSet(ids1, dist1, TOP_K/2, NQ, TOP_K, ascending, result);
-    ms::XSearchTask::MergeTopkToResultSet(ids2, dist2, TOP_K/3, NQ, TOP_K, ascending, result);
-    CheckTopkResult(ids1, dist1, ids2, dist2, TOP_K, NQ, ascending, result);
+    MergeTopkArrayTest(TOP_K/2, TOP_K/3, NQ, TOP_K, true);
+    MergeTopkArrayTest(TOP_K/2, TOP_K/3, NQ, TOP_K, false);
+    MergeTopkArrayTest(TOP_K/3, TOP_K/2, NQ, TOP_K, true);
+    MergeTopkArrayTest(TOP_K/3, TOP_K/2, NQ, TOP_K, false);
 }
 
 TEST(DBSearchTest, REDUCE_PERF_TEST) {
     int32_t index_file_num = 478;   /* sift1B dataset, index files num */
     bool ascending = true;
 
-    std::vector<int32_t> thread_vec = {4, 8, 11};
-    std::vector<int32_t> nq_vec = {1, 10, 100, 1000};
-    std::vector<int32_t> topk_vec = {1, 4, 16, 64, 256, 1024};
+    std::vector<int32_t> thread_vec = {4, 8};
+    std::vector<int32_t> nq_vec = {1, 10, 100};
+    std::vector<int32_t> topk_vec = {1, 4, 16, 64};
     int32_t NQ = nq_vec[nq_vec.size()-1];
     int32_t TOPK = topk_vec[topk_vec.size()-1];