Format and and extra parameters for checking

Signed-off-by: FluorineDog <guilin.gou@zilliz.com>
pull/4973/head^2
FluorineDog 2020-11-16 15:41:56 +08:00 committed by yefu.chen
parent 9d3a21a9c9
commit 8c48cf30c0
45 changed files with 305 additions and 377 deletions

View File

@ -32,7 +32,7 @@ func main() {
etcdPort, _ := gparams.GParams.Load("etcd.port")
etcdAddr := etcdAddress + ":" + etcdPort
etcdRootPath, _ := gparams.GParams.Load("etcd.rootpath")
svr, err := master.CreateServer(ctx, etcdRootPath, etcdRootPath, []string{etcdAddr})
svr, err := master.CreateServer(ctx, etcdRootPath, etcdRootPath, etcdRootPath, []string{etcdAddr})
if err != nil {
log.Print("create server failed", zap.Error(err))
}

View File

@ -1,25 +1,15 @@
#pragma once
#include <vector>
#include <assert.h>
#include <stdexcept>
#include "utils/Types.h"
// #include "knowhere/index/Index.h"
#include "utils/Status.h"
#include "utils/EasyAssert.h"
namespace milvus::segcore {
#include <stdexcept>
namespace milvus {
using Timestamp = uint64_t; // TODO: use TiKV-like timestamp
using engine::DataType;
using engine::FieldElementType;
struct RowBasedRawData {
void* raw_data; // schema
int sizeof_per_row; // alignment
int64_t count;
};
inline int
field_sizeof(DataType data_type, int dim = 1) {
switch (data_type) {
@ -126,100 +116,4 @@ struct FieldMeta {
DataType type_ = DataType::NONE;
int dim_ = 1;
};
class Schema {
public:
void
AddField(std::string_view field_name, DataType data_type, int dim = 1) {
auto field_meta = FieldMeta(field_name, data_type, dim);
this->AddField(std::move(field_meta));
}
void
AddField(FieldMeta field_meta) {
auto offset = fields_.size();
fields_.emplace_back(field_meta);
offsets_.emplace(field_meta.get_name(), offset);
auto field_sizeof = field_meta.get_sizeof();
sizeof_infos_.push_back(field_sizeof);
total_sizeof_ += field_sizeof;
}
auto
begin() {
return fields_.begin();
}
auto
end() {
return fields_.end();
}
auto
begin() const {
return fields_.begin();
}
auto
end() const {
return fields_.end();
}
int
size() const {
return fields_.size();
}
const FieldMeta&
operator[](int field_index) const {
Assert(field_index >= 0);
Assert(field_index < fields_.size());
return fields_[field_index];
}
auto
get_total_sizeof() const {
return total_sizeof_;
}
const std::vector<int>&
get_sizeof_infos() {
return sizeof_infos_;
}
std::optional<int>
get_offset(const std::string& field_name) {
if (!offsets_.count(field_name)) {
return std::nullopt;
} else {
return offsets_[field_name];
}
}
const std::vector<FieldMeta>&
get_fields() {
return fields_;
}
const FieldMeta&
operator[](const std::string& field_name) const {
auto offset_iter = offsets_.find(field_name);
AssertInfo(offset_iter != offsets_.end(), "Cannot found field_name: " + field_name);
auto offset = offset_iter->second;
return (*this)[offset];
}
private:
// this is where data holds
std::vector<FieldMeta> fields_;
private:
// a mapping for random access
std::unordered_map<std::string, int> offsets_;
std::vector<int> sizeof_infos_;
int total_sizeof_ = 0;
};
using SchemaPtr = std::shared_ptr<Schema>;
using idx_t = int64_t;
} // namespace milvus::segcore
} // namespace milvus

View File

@ -0,0 +1,102 @@
#pragma once
#include "FieldMeta.h"
#include <vector>
namespace milvus {
class Schema {
public:
void
AddField(std::string_view field_name, DataType data_type, int dim = 1) {
auto field_meta = FieldMeta(field_name, data_type, dim);
this->AddField(std::move(field_meta));
}
void
AddField(FieldMeta field_meta) {
auto offset = fields_.size();
fields_.emplace_back(field_meta);
offsets_.emplace(field_meta.get_name(), offset);
auto field_sizeof = field_meta.get_sizeof();
sizeof_infos_.push_back(field_sizeof);
total_sizeof_ += field_sizeof;
}
auto
begin() {
return fields_.begin();
}
auto
end() {
return fields_.end();
}
auto
begin() const {
return fields_.begin();
}
auto
end() const {
return fields_.end();
}
int
size() const {
return fields_.size();
}
const FieldMeta&
operator[](int field_index) const {
Assert(field_index >= 0);
Assert(field_index < fields_.size());
return fields_[field_index];
}
auto
get_total_sizeof() const {
return total_sizeof_;
}
const std::vector<int>&
get_sizeof_infos() {
return sizeof_infos_;
}
std::optional<int>
get_offset(const std::string& field_name) {
if (!offsets_.count(field_name)) {
return std::nullopt;
} else {
return offsets_[field_name];
}
}
const std::vector<FieldMeta>&
get_fields() {
return fields_;
}
const FieldMeta&
operator[](const std::string& field_name) const {
auto offset_iter = offsets_.find(field_name);
AssertInfo(offset_iter != offsets_.end(), "Cannot found field_name: " + field_name);
auto offset = offset_iter->second;
return (*this)[offset];
}
private:
// this is where data holds
std::vector<FieldMeta> fields_;
private:
// a mapping for random access
std::unordered_map<std::string, int> offsets_;
std::vector<int> sizeof_infos_;
int total_sizeof_ = 0;
};
using SchemaPtr = std::shared_ptr<Schema>;
using idx_t = int64_t;
} // namespace milvus

View File

@ -16,7 +16,7 @@
namespace milvus {
namespace knowhere {
using Config = milvus::json;
using Config = milvus::Json;
} // namespace knowhere
} // namespace milvus

View File

@ -6,7 +6,6 @@ set(MILVUS_QUERY_SRCS
visitors/ShowPlanNodeVisitor.cpp
visitors/ExecPlanNodeVisitor.cpp
visitors/ShowExprVisitor.cpp
Parser.cpp
Plan.cpp
)
add_library(milvus_query ${MILVUS_QUERY_SRCS})

View File

@ -4,7 +4,7 @@
#include <any>
#include <string>
#include <optional>
#include "segcore/SegmentDefs.h"
#include "common/Schema.h"
namespace milvus::query {
class ExprVisitor;
@ -60,7 +60,7 @@ using FieldId = std::string;
struct TermExpr : Expr {
FieldId field_id_;
segcore::DataType data_type_;
DataType data_type_;
// std::vector<std::any> terms_;
protected:
@ -74,7 +74,7 @@ struct TermExpr : Expr {
struct RangeExpr : Expr {
FieldId field_id_;
segcore::DataType data_type_;
DataType data_type_;
// std::vector<std::tuple<OpType, std::any>> conditions_;
protected:
// prevent accidential instantiation

View File

@ -9,7 +9,7 @@
namespace milvus::query {
static std::unique_ptr<VectorPlanNode>
CreateVec(const std::string& field_name, const json& vec_info) {
CreateVecNode(const std::string& field_name, const Json& vec_info) {
// TODO add binary info
auto vec_node = std::make_unique<FloatVectorANNS>();
auto topK = vec_info["topk"];
@ -35,7 +35,7 @@ CreatePlanImplNaive(const std::string& dsl_str) {
auto iter = pack["vector"].begin();
auto key = iter.key();
auto& body = iter.value();
plan->plan_node_ = CreateVec(key, body);
plan->plan_node_ = CreateVecNode(key, body);
return plan;
}
}
@ -44,13 +44,18 @@ CreatePlanImplNaive(const std::string& dsl_str) {
auto iter = bool_dsl["vector"].begin();
auto key = iter.key();
auto& body = iter.value();
plan->plan_node_ = CreateVec(key, body);
plan->plan_node_ = CreateVecNode(key, body);
return plan;
} else {
PanicInfo("Unsupported DSL: vector node not detected");
}
}
static std::unique_ptr<Plan>
CreateRangeNode(const Json& range_group) {
return nullptr;
}
void
CheckNull(const Json& json) {
Assert(!json.is_null());
@ -82,13 +87,15 @@ class PlanParser {
};
std::unique_ptr<Plan>
CreatePlan(const std::string& dsl_str) {
CreatePlan(const Schema& schema, const std::string& dsl_str) {
(void)schema;
auto plan = CreatePlanImplNaive(dsl_str);
return plan;
}
std::unique_ptr<PlaceholderGroup>
ParsePlaceholderGroup(const std::string& blob) {
ParsePlaceholderGroup(const Plan* plan, const std::string& blob) {
(void)plan;
namespace ser = milvus::proto::service;
auto result = std::make_unique<PlaceholderGroup>();
ser::PlaceholderGroup ph_group;

View File

@ -1,6 +1,7 @@
#pragma once
#include <memory>
#include <string_view>
#include "common/Schema.h"
namespace milvus::query {
// NOTE: APIs for C wrapper
@ -10,10 +11,10 @@ struct Plan;
struct PlaceholderGroup;
std::unique_ptr<Plan>
CreatePlan(const std::string& dsl);
CreatePlan(const Schema& schema, const std::string& dsl);
std::unique_ptr<PlaceholderGroup>
ParsePlaceholderGroup(const std::string& placeholder_group_blob);
ParsePlaceholderGroup(const Plan* plan, const std::string& placeholder_group_blob);
int64_t
GetNumOfQueries(const PlaceholderGroup*);

View File

@ -49,7 +49,7 @@ struct QueryColumn {
};
struct TermQuery {
milvus::json json_obj;
milvus::Json json_obj;
// std::string field_name;
// std::vector<uint8_t> field_value;
// float boost;
@ -62,7 +62,7 @@ struct CompareExpr {
};
struct RangeQuery {
milvus::json json_obj;
milvus::Json json_obj;
// std::string field_name;
// std::vector<CompareExpr> compare_expr;
// float boost;
@ -77,7 +77,7 @@ struct VectorRecord {
struct VectorQuery {
std::string field_name;
milvus::json extra_params = {};
milvus::Json extra_params = {};
int64_t topk;
int64_t nq;
std::string metric_type = "";

View File

@ -5,7 +5,7 @@
namespace milvus::wtf {
using google::protobuf::RepeatedField;
using google::protobuf::RepeatedPtrField;
#if 0
#if 1
#if 0
void
CopyRowRecords(const RepeatedPtrField<proto::service::PlaceholderValue>& grpc_records,
@ -51,27 +51,27 @@ CopyRowRecords(const RepeatedPtrField<proto::service::PlaceholderValue>& grpc_re
#endif
Status
ProcessLeafQueryJson(const milvus::json& query_json, query_old::BooleanQueryPtr& query, std::string& field_name) {
#if 0
ProcessLeafQueryJson(const milvus::Json& query_json, query_old::BooleanQueryPtr& query, std::string& field_name) {
#if 1
if (query_json.contains("term")) {
auto leaf_query = std::make_shared<query_old::LeafQuery>();
auto term_query = std::make_shared<query_old::TermQuery>();
milvus::json json_obj = query_json["term"];
milvus::Json json_obj = query_json["term"];
JSON_NULL_CHECK(json_obj);
JSON_OBJECT_CHECK(json_obj);
term_query->json_obj = json_obj;
milvus::json::iterator json_it = json_obj.begin();
milvus::Json::iterator json_it = json_obj.begin();
field_name = json_it.key();
leaf_query->term_query = term_query;
query->AddLeafQuery(leaf_query);
} else if (query_json.contains("range")) {
auto leaf_query = std::make_shared<query_old::LeafQuery>();
auto range_query = std::make_shared<query_old::RangeQuery>();
milvus::json json_obj = query_json["range"];
milvus::Json json_obj = query_json["range"];
JSON_NULL_CHECK(json_obj);
JSON_OBJECT_CHECK(json_obj);
range_query->json_obj = json_obj;
milvus::json::iterator json_it = json_obj.begin();
milvus::Json::iterator json_it = json_obj.begin();
field_name = json_it.key();
leaf_query->range_query = range_query;
@ -91,7 +91,7 @@ ProcessLeafQueryJson(const milvus::json& query_json, query_old::BooleanQueryPtr&
}
Status
ProcessBooleanQueryJson(const milvus::json& query_json,
ProcessBooleanQueryJson(const milvus::Json& query_json,
query_old::BooleanQueryPtr& boolean_query,
query_old::QueryPtr& query_ptr) {
#if 1
@ -173,12 +173,12 @@ ProcessBooleanQueryJson(const milvus::json& query_json,
Status
DeserializeJsonToBoolQuery(const google::protobuf::RepeatedPtrField<::milvus::grpc::VectorParam>& vector_params,
const std::string& dsl_string,
query_old::BooleanQueryPtr& boolean_query,
query_old::QueryPtr& query_ptr) {
const std::string& dsl_string,
query_old::BooleanQueryPtr& boolean_query,
query_old::QueryPtr& query_ptr) {
#if 1
try {
milvus::json dsl_json = json::parse(dsl_string);
milvus::Json dsl_json = Json::parse(dsl_string);
if (dsl_json.empty()) {
return Status{SERVER_INVALID_ARGUMENT, "Query dsl is null"};
@ -189,16 +189,16 @@ DeserializeJsonToBoolQuery(const google::protobuf::RepeatedPtrField<::milvus::gr
}
for (const auto& vector_param : vector_params) {
const std::string& vector_string = vector_param.json();
milvus::json vector_json = json::parse(vector_string);
milvus::json::iterator it = vector_json.begin();
milvus::Json vector_json = Json::parse(vector_string);
milvus::Json::iterator it = vector_json.begin();
std::string placeholder = it.key();
auto vector_query = std::make_shared<query_old::VectorQuery>();
milvus::json::iterator vector_param_it = it.value().begin();
milvus::Json::iterator vector_param_it = it.value().begin();
if (vector_param_it != it.value().end()) {
const std::string& field_name = vector_param_it.key();
vector_query->field_name = field_name;
milvus::json param_json = vector_param_it.value();
milvus::Json param_json = vector_param_it.value();
int64_t topk = param_json["topk"];
// STATUS_CHECK(server::ValidateSearchTopk(topk));
vector_query->topk = topk;

View File

@ -17,9 +17,7 @@ class ExecPlanNodeVisitor : PlanNodeVisitor {
public:
using RetType = segcore::QueryResult;
ExecPlanNodeVisitor(segcore::SegmentBase& segment,
segcore::Timestamp timestamp,
const PlaceholderGroup& placeholder_group)
ExecPlanNodeVisitor(segcore::SegmentBase& segment, Timestamp timestamp, const PlaceholderGroup& placeholder_group)
: segment_(segment), timestamp_(timestamp), placeholder_group_(placeholder_group) {
}
// using RetType = nlohmann::json;
@ -37,7 +35,7 @@ class ExecPlanNodeVisitor : PlanNodeVisitor {
private:
// std::optional<RetType> ret_;
segcore::SegmentBase& segment_;
segcore::Timestamp timestamp_;
Timestamp timestamp_;
const PlaceholderGroup& placeholder_group_;
std::optional<RetType> ret_;

View File

@ -13,9 +13,7 @@ namespace impl {
class ExecPlanNodeVisitor : PlanNodeVisitor {
public:
using RetType = segcore::QueryResult;
ExecPlanNodeVisitor(segcore::SegmentBase& segment,
segcore::Timestamp timestamp,
const PlaceholderGroup& placeholder_group)
ExecPlanNodeVisitor(segcore::SegmentBase& segment, Timestamp timestamp, const PlaceholderGroup& placeholder_group)
: segment_(segment), timestamp_(timestamp), placeholder_group_(placeholder_group) {
}
// using RetType = nlohmann::json;
@ -33,7 +31,7 @@ class ExecPlanNodeVisitor : PlanNodeVisitor {
private:
// std::optional<RetType> ret_;
segcore::SegmentBase& segment_;
segcore::Timestamp timestamp_;
Timestamp timestamp_;
const PlaceholderGroup& placeholder_group_;
std::optional<RetType> ret_;

View File

@ -100,8 +100,7 @@ TermExtract(const TermExpr& expr_raw) {
void
ShowExprVisitor::visit(TermExpr& expr) {
Assert(!ret_.has_value());
Assert(segcore::field_is_vector(expr.data_type_) == false);
using segcore::DataType;
Assert(field_is_vector(expr.data_type_) == false);
auto terms = [&] {
switch (expr.data_type_) {
case DataType::INT8:
@ -125,7 +124,7 @@ ShowExprVisitor::visit(TermExpr& expr) {
Json res{{"expr_type", "Term"},
{"field_id", expr.field_id_},
{"data_type", segcore::datatype_name(expr.data_type_)},
{"data_type", datatype_name(expr.data_type_)},
{"terms", std::move(terms)}};
ret_ = res;
@ -142,8 +141,7 @@ CondtionExtract(const RangeExpr& expr_raw) {
void
ShowExprVisitor::visit(RangeExpr& expr) {
Assert(!ret_.has_value());
Assert(segcore::field_is_vector(expr.data_type_) == false);
using segcore::DataType;
Assert(field_is_vector(expr.data_type_) == false);
auto conditions = [&] {
switch (expr.data_type_) {
case DataType::BOOL:
@ -167,7 +165,7 @@ ShowExprVisitor::visit(RangeExpr& expr) {
Json res{{"expr_type", "Range"},
{"field_id", expr.field_id_},
{"data_type", segcore::datatype_name(expr.data_type_)},
{"data_type", datatype_name(expr.data_type_)},
{"conditions", std::move(conditions)}};
}
} // namespace milvus::query

View File

@ -1,6 +1,6 @@
#pragma once
#include "SegmentDefs.h"
#include "common/Schema.h"
#include "IndexMeta.h"
namespace milvus::segcore {

View File

@ -1,8 +1,8 @@
#pragma once
#include "AckResponder.h"
#include "SegmentDefs.h"
#include "knowhere//index/vector_index/IndexIVF.h"
#include "common/Schema.h"
#include "knowhere/index/vector_index/IndexIVF.h"
#include <memory>
namespace milvus::segcore {

View File

@ -2,9 +2,9 @@
//
//#include <shared_mutex>
//
//#include "SegmentDefs.h"
//#include "common/Schema.h"
// #include "segcore/SegmentBase.h"
#include "segcore/SegmentDefs.h"
#include "common/Schema.h"
#include "knowhere/index/IndexType.h"
#include "knowhere/common/Config.h"
#include <map>

View File

@ -1,7 +1,7 @@
#pragma once
#include "AckResponder.h"
#include <tbb/concurrent_vector.h>
#include "SegmentDefs.h"
#include "common/Schema.h"
#include <optional>
#include "InsertRecord.h"
#include <knowhere/index/vector_index/IndexIVF.h>

View File

@ -1,5 +1,5 @@
#pragma once
#include "SegmentDefs.h"
#include "common/Schema.h"
#include "ConcurrentVector.h"
#include "AckResponder.h"

View File

@ -3,9 +3,8 @@
#include "IndexMeta.h"
#include "utils/Types.h"
#include "segcore/SegmentDefs.h"
// #include "knowhere/index/Index.h"
// #include "knowhere/index/IndexType.h"
#include "common/Schema.h"
#include "query/deprecated/GeneralQuery.h"
#include "query/Plan.h"
@ -14,6 +13,11 @@ namespace segcore {
// using engine::DataChunk;
// using engine::DataChunkPtr;
using engine::QueryResult;
struct RowBasedRawData {
void* raw_data; // schema
int sizeof_per_row; // alignment
int64_t count;
};
int
TestABI();

View File

@ -15,7 +15,7 @@
namespace milvus {
using json = nlohmann::json;
using Json = nlohmann::json;
#define JSON_NULL_CHECK(json) \
do { \

View File

@ -1,11 +1,12 @@
#include <gtest/gtest.h>
#include "query/Parser.h"
#include "query/deprecated/Parser.h"
#include "query/Expr.h"
#include "query/PlanNode.h"
#include "query/generated/ExprVisitor.h"
#include "query/generated/PlanNodeVisitor.h"
#include "test_utils/DataGen.h"
#include "query/generated/ShowPlanNodeVisitor.h"
using namespace milvus;
TEST(Expr, Naive) {
SUCCEED();

View File

@ -1,5 +1,5 @@
#include <gtest/gtest.h>
#include "query/Parser.h"
#include "query/deprecated/Parser.h"
#include "query/Expr.h"
#include "query/PlanNode.h"
#include "query/generated/ExprVisitor.h"
@ -9,6 +9,9 @@
#include "query/generated/ExecPlanNodeVisitor.h"
#include "query/PlanImpl.h"
using namespace milvus;
using namespace milvus::query;
using namespace milvus::segcore;
TEST(Query, Naive) {
SUCCEED();
using namespace milvus::wtf;
@ -53,10 +56,11 @@ TEST(Query, Naive) {
TEST(Query, ShowExecutor) {
using namespace milvus::query;
using namespace milvus::segcore;
using namespace milvus;
auto node = std::make_unique<FloatVectorANNS>();
auto schema = std::make_shared<Schema>();
int64_t num_queries = 100L;
schema->AddField("fakevec", DataType::VECTOR_FLOAT, 16);
int64_t num_queries = 100L;
auto raw_data = DataGen(schema, num_queries);
auto& info = node->query_info_;
info.metric_type_ = "L2";
@ -94,7 +98,11 @@ TEST(Query, DSL) {
]
}
})";
auto plan = CreatePlan(dsl_string);
auto schema = std::make_shared<Schema>();
schema->AddField("fakevec", DataType::VECTOR_FLOAT, 16);
auto plan = CreatePlan(*schema, dsl_string);
auto res = shower.call_child(*plan->plan_node_);
std::cout << res.dump(4) << std::endl;
@ -113,16 +121,33 @@ TEST(Query, DSL) {
}
}
})";
auto plan2 = CreatePlan(dsl_string2);
auto plan2 = CreatePlan(*schema, dsl_string2);
auto res2 = shower.call_child(*plan2->plan_node_);
std::cout << res2.dump(4) << std::endl;
ASSERT_EQ(res, res2);
}
TEST(Query, ParsePlaceholderGroup) {
using namespace milvus::query;
using namespace milvus::segcore;
namespace ser = milvus::proto::service;
std::string dsl_string = R"(
{
"bool": {
"vector": {
"Vec": {
"metric_type": "L2",
"params": {
"nprobe": 10
},
"query": "$0",
"topk": 10
}
}
}
})";
auto schema = std::make_shared<Schema>();
schema->AddField("fakevec", DataType::VECTOR_FLOAT, 16);
auto plan = CreatePlan(*schema, dsl_string);
int num_queries = 10;
int dim = 16;
std::default_random_engine e;
@ -131,22 +156,21 @@ TEST(Query, ParsePlaceholderGroup) {
auto value = raw_group.add_placeholders();
value->set_tag("$0");
value->set_type(ser::PlaceholderType::VECTOR_FLOAT);
for(int i = 0; i < num_queries; ++i) {
for (int i = 0; i < num_queries; ++i) {
std::vector<float> vec;
for(int d = 0; d < dim; ++d) {
for (int d = 0; d < dim; ++d) {
vec.push_back(dis(e));
}
// std::string line((char*)vec.data(), (char*)vec.data() + vec.size() * sizeof(float));
value->add_values(vec.data(), vec.size() * sizeof(float));
}
auto blob = raw_group.SerializeAsString();
//ser::PlaceholderGroup new_group;
//new_group.ParseFromString()
auto fuck = ParsePlaceholderGroup(blob);
int x = 1+1;
// ser::PlaceholderGroup new_group;
// new_group.ParseFromString()
auto placeholder = ParsePlaceholderGroup(plan.get(), blob);
int x = 1 + 1;
}
TEST(Query, Exec) {
using namespace milvus::query;
using namespace milvus::segcore;

View File

@ -24,6 +24,7 @@
using std::cin;
using std::cout;
using std::endl;
using namespace milvus;
namespace {
auto

View File

@ -1,5 +1,5 @@
#pragma once
#include "segcore/SegmentDefs.h"
#include "common/Schema.h"
#include <random>
#include <memory>
#include <cstring>

View File

@ -36,7 +36,7 @@ func TestMaster_CollectionTask(t *testing.T) {
_, err = etcdCli.Delete(ctx, "/test/root", clientv3.WithPrefix())
assert.Nil(t, err)
svr, err := CreateServer(ctx, "/test/root/kv", "/test/root/meta", []string{etcdAddr})
svr, err := CreateServer(ctx, "/test/root/kv", "/test/root/meta", "/test/root/meta/tso", []string{etcdAddr})
assert.Nil(t, err)
err = svr.Run(10002)
assert.Nil(t, err)

View File

@ -34,7 +34,7 @@ func TestMaster_CreateCollection(t *testing.T) {
_, err = etcdCli.Delete(ctx, "/test/root", clientv3.WithPrefix())
assert.Nil(t, err)
svr, err := CreateServer(ctx, "/test/root/kv", "/test/root/meta", []string{etcdAddr})
svr, err := CreateServer(ctx, "/test/root/kv", "/test/root/meta", "/test/root/meta/tso", []string{etcdAddr})
assert.Nil(t, err)
err = svr.Run(10001)
assert.Nil(t, err)

View File

@ -16,8 +16,8 @@ type GlobalIDAllocator struct {
var allocator *GlobalIDAllocator
func Init(etcdAddr []string, rootPath string) {
InitGlobalIDAllocator("idTimestamp", tsoutil.NewTSOKVBase(etcdAddr, rootPath, "gid"))
func Init() {
InitGlobalIDAllocator("idTimestamp", tsoutil.NewTSOKVBase("gid"))
}
func InitGlobalIDAllocator(key string, base kvutil.Base) {

View File

@ -17,14 +17,7 @@ func TestMain(m *testing.M) {
if err != nil {
panic(err)
}
etcdPort, err := gparams.GParams.Load("etcd.port")
if err != nil {
panic(err)
}
etcdAddr := "127.0.0.1:" + etcdPort
GIdAllocator = NewGlobalIDAllocator("idTimestamp", tsoutil.NewTSOKVBase([]string{etcdAddr}, "/test/root/kv", "gid"))
GIdAllocator = NewGlobalIDAllocator("idTimestamp", tsoutil.NewTSOKVBase("gid"))
exitCode := m.Run()
os.Exit(exitCode)
}

View File

@ -72,15 +72,15 @@ func newKVBase(kvRoot string, etcdAddr []string) *kv.EtcdKV {
return kvBase
}
func Init(etcdAddr []string, rootPath string) {
func Init() {
rand.Seed(time.Now().UnixNano())
id.Init(etcdAddr, rootPath)
tso.Init(etcdAddr, rootPath)
id.Init()
tso.Init()
}
// CreateServer creates the UNINITIALIZED pd server with given configuration.
func CreateServer(ctx context.Context, kvRootPath, metaRootPath string, etcdAddr []string) (*Master, error) {
Init(etcdAddr, kvRootPath)
func CreateServer(ctx context.Context, kvRootPath string, metaRootPath, tsoRootPath string, etcdAddr []string) (*Master, error) {
Init()
etcdClient, err := clientv3.New(clientv3.Config{Endpoints: etcdAddr})
if err != nil {

View File

@ -38,7 +38,7 @@ func TestMaster_Partition(t *testing.T) {
assert.Nil(t, err)
port := 10000 + rand.Intn(1000)
svr, err := CreateServer(ctx, "/test/root/kv", "/test/root/meta", []string{etcdAddr})
svr, err := CreateServer(ctx, "/test/root/kv", "/test/root/meta", "/test/root/meta/tso", []string{etcdAddr})
assert.Nil(t, err)
err = svr.Run(int64(port))
assert.Nil(t, err)

View File

@ -1,8 +1,6 @@
package master
import (
"github.com/zilliztech/milvus-distributed/internal/master/id"
)
import "math/rand"
type ddRequestScheduler struct {
reqQueue chan task
@ -23,6 +21,7 @@ func (rs *ddRequestScheduler) Enqueue(task task) error {
return nil
}
//TODO, allocGlobalID
func allocGlobalID() (UniqueID, error) {
return id.AllocOne()
return rand.Int63(), nil
}

View File

@ -37,8 +37,8 @@ type GlobalTSOAllocator struct {
var allocator *GlobalTSOAllocator
func Init(etcdAddr []string, rootPath string) {
InitGlobalTsoAllocator("timestamp", tsoutil.NewTSOKVBase(etcdAddr, rootPath, "tso"))
func Init() {
InitGlobalTsoAllocator("timestamp", tsoutil.NewTSOKVBase("tso"))
}
func InitGlobalTsoAllocator(key string, base kvutil.Base) {

View File

@ -18,13 +18,7 @@ func TestMain(m *testing.M) {
if err != nil {
panic(err)
}
etcdPort, err := gparams.GParams.Load("etcd.port")
if err != nil {
panic(err)
}
etcdAddr := "127.0.0.1:" + etcdPort
GTsoAllocator = NewGlobalTSOAllocator("timestamp", tsoutil.NewTSOKVBase([]string{etcdAddr}, "/test/root/kv", "tso"))
GTsoAllocator = NewGlobalTSOAllocator("timestamp", tsoutil.NewTSOKVBase("tso"))
exitCode := m.Run()
os.Exit(exitCode)

View File

@ -106,16 +106,6 @@ func getTsMsg(msgType MsgType, reqID UniqueID, hashValue int32) *TsMsg {
TimeTickMsg: timeTickResult,
}
tsMsg = timeTickMsg
case internalPb.MsgType_kQueryNodeSegStats:
queryNodeSegStats := internalPb.QueryNodeSegStats{
MsgType: internalPb.MsgType_kQueryNodeSegStats,
PeerID: reqID,
}
queryNodeSegStatsMsg := &QueryNodeSegStatsMsg{
BaseMsg: baseMsg,
QueryNodeSegStats: queryNodeSegStats,
}
tsMsg = queryNodeSegStatsMsg
}
return &tsMsg
}
@ -462,11 +452,24 @@ func TestStream_PulsarMsgStream_DefaultRepackFunc(t *testing.T) {
consumerChannels := []string{"insert1", "insert2"}
consumerSubName := "subInsert"
baseMsg := BaseMsg{
BeginTimestamp: 0,
EndTimestamp: 0,
HashValues: []int32{1},
}
timeTickRequest := internalPb.TimeTickMsg{
MsgType: internalPb.MsgType_kTimeTick,
PeerID: int64(1),
Timestamp: uint64(1),
}
timeTick := &TimeTickMsg{
BaseMsg: baseMsg,
TimeTickMsg: timeTickRequest,
}
var tsMsg TsMsg = timeTick
msgPack := MsgPack{}
msgPack.Msgs = append(msgPack.Msgs, getTsMsg(internalPb.MsgType_kTimeTick, 1, 1))
msgPack.Msgs = append(msgPack.Msgs, getTsMsg(internalPb.MsgType_kSearch, 2, 2))
msgPack.Msgs = append(msgPack.Msgs, getTsMsg(internalPb.MsgType_kSearchResult, 3, 3))
msgPack.Msgs = append(msgPack.Msgs, getTsMsg(internalPb.MsgType_kQueryNodeSegStats, 4, 4))
msgPack.Msgs = append(msgPack.Msgs, &tsMsg)
inputStream := NewPulsarMsgStream(context.Background(), 100)
inputStream.SetPulsarCient(pulsarAddress)

View File

@ -57,24 +57,24 @@ func (it *InsertMsg) Marshal(input *TsMsg) ([]byte, error) {
func (it *InsertMsg) Unmarshal(input []byte) (*TsMsg, error) {
insertRequest := internalPb.InsertRequest{}
err := proto.Unmarshal(input, &insertRequest)
insertMsg := &InsertMsg{InsertRequest: insertRequest}
if err != nil {
return nil, err
}
insertMsg := &InsertMsg{InsertRequest: insertRequest}
for _, timestamp := range insertMsg.Timestamps {
insertMsg.BeginTimestamp = timestamp
insertMsg.EndTimestamp = timestamp
it.BeginTimestamp = timestamp
it.EndTimestamp = timestamp
break
}
for _, timestamp := range insertMsg.Timestamps {
if timestamp > insertMsg.EndTimestamp {
insertMsg.EndTimestamp = timestamp
if timestamp > it.EndTimestamp {
it.EndTimestamp = timestamp
}
if timestamp < insertMsg.BeginTimestamp {
insertMsg.BeginTimestamp = timestamp
if timestamp < it.BeginTimestamp {
it.BeginTimestamp = timestamp
}
}
var tsMsg TsMsg = insertMsg
return &tsMsg, nil
}
@ -102,24 +102,24 @@ func (dt *DeleteMsg) Marshal(input *TsMsg) ([]byte, error) {
func (dt *DeleteMsg) Unmarshal(input []byte) (*TsMsg, error) {
deleteRequest := internalPb.DeleteRequest{}
err := proto.Unmarshal(input, &deleteRequest)
deleteMsg := &DeleteMsg{DeleteRequest: deleteRequest}
if err != nil {
return nil, err
}
deleteMsg := &DeleteMsg{DeleteRequest: deleteRequest}
for _, timestamp := range deleteMsg.Timestamps {
deleteMsg.BeginTimestamp = timestamp
deleteMsg.EndTimestamp = timestamp
dt.BeginTimestamp = timestamp
dt.EndTimestamp = timestamp
break
}
for _, timestamp := range deleteMsg.Timestamps {
if timestamp > deleteMsg.EndTimestamp {
deleteMsg.EndTimestamp = timestamp
if timestamp > dt.EndTimestamp {
dt.EndTimestamp = timestamp
}
if timestamp < deleteMsg.BeginTimestamp {
deleteMsg.BeginTimestamp = timestamp
if timestamp < dt.BeginTimestamp {
dt.BeginTimestamp = timestamp
}
}
var tsMsg TsMsg = deleteMsg
return &tsMsg, nil
}
@ -147,13 +147,13 @@ func (st *SearchMsg) Marshal(input *TsMsg) ([]byte, error) {
func (st *SearchMsg) Unmarshal(input []byte) (*TsMsg, error) {
searchRequest := internalPb.SearchRequest{}
err := proto.Unmarshal(input, &searchRequest)
searchMsg := &SearchMsg{SearchRequest: searchRequest}
if err != nil {
return nil, err
}
searchMsg := &SearchMsg{SearchRequest: searchRequest}
searchMsg.BeginTimestamp = searchMsg.Timestamp
searchMsg.EndTimestamp = searchMsg.Timestamp
st.BeginTimestamp = searchMsg.Timestamp
st.EndTimestamp = searchMsg.Timestamp
var tsMsg TsMsg = searchMsg
return &tsMsg, nil
}
@ -181,13 +181,13 @@ func (srt *SearchResultMsg) Marshal(input *TsMsg) ([]byte, error) {
func (srt *SearchResultMsg) Unmarshal(input []byte) (*TsMsg, error) {
searchResultRequest := internalPb.SearchResult{}
err := proto.Unmarshal(input, &searchResultRequest)
searchResultMsg := &SearchResultMsg{SearchResult: searchResultRequest}
if err != nil {
return nil, err
}
searchResultMsg := &SearchResultMsg{SearchResult: searchResultRequest}
searchResultMsg.BeginTimestamp = searchResultMsg.Timestamp
searchResultMsg.EndTimestamp = searchResultMsg.Timestamp
srt.BeginTimestamp = searchResultMsg.Timestamp
srt.EndTimestamp = searchResultMsg.Timestamp
var tsMsg TsMsg = searchResultMsg
return &tsMsg, nil
}
@ -215,49 +215,17 @@ func (tst *TimeTickMsg) Marshal(input *TsMsg) ([]byte, error) {
func (tst *TimeTickMsg) Unmarshal(input []byte) (*TsMsg, error) {
timeTickMsg := internalPb.TimeTickMsg{}
err := proto.Unmarshal(input, &timeTickMsg)
if err != nil {
return nil, err
}
timeTick := &TimeTickMsg{TimeTickMsg: timeTickMsg}
timeTick.BeginTimestamp = timeTick.Timestamp
timeTick.EndTimestamp = timeTick.Timestamp
if err != nil {
return nil, err
}
tst.BeginTimestamp = timeTick.Timestamp
tst.EndTimestamp = timeTick.Timestamp
var tsMsg TsMsg = timeTick
return &tsMsg, nil
}
/////////////////////////////////////////QueryNodeSegStats//////////////////////////////////////////
type QueryNodeSegStatsMsg struct {
BaseMsg
internalPb.QueryNodeSegStats
}
func (qs *QueryNodeSegStatsMsg) Type() MsgType {
return qs.MsgType
}
func (qs *QueryNodeSegStatsMsg) Marshal(input *TsMsg) ([]byte, error) {
queryNodeSegStatsTask := (*input).(*QueryNodeSegStatsMsg)
queryNodeSegStats := &queryNodeSegStatsTask.QueryNodeSegStats
mb, err := proto.Marshal(queryNodeSegStats)
if err != nil {
return nil, err
}
return mb, nil
}
func (qs *QueryNodeSegStatsMsg) Unmarshal(input []byte) (*TsMsg, error) {
queryNodeSegStats := internalPb.QueryNodeSegStats{}
err := proto.Unmarshal(input, &queryNodeSegStats)
if err != nil {
return nil, err
}
queryNodeSegStatsMsg := &QueryNodeSegStatsMsg{QueryNodeSegStats: queryNodeSegStats}
var tsMsg TsMsg = queryNodeSegStatsMsg
return &tsMsg, nil
}
///////////////////////////////////////////Key2Seg//////////////////////////////////////////
//type Key2SegMsg struct {
// BaseMsg

View File

@ -30,14 +30,12 @@ func (dispatcher *UnmarshalDispatcher) addDefaultMsgTemplates() {
searchMsg := SearchMsg{}
searchResultMsg := SearchResultMsg{}
timeTickMsg := TimeTickMsg{}
queryNodeSegStatsMsg := QueryNodeSegStatsMsg{}
dispatcher.tempMap = make(map[internalPb.MsgType]UnmarshalFunc)
dispatcher.tempMap[internalPb.MsgType_kInsert] = insertMsg.Unmarshal
dispatcher.tempMap[internalPb.MsgType_kDelete] = deleteMsg.Unmarshal
dispatcher.tempMap[internalPb.MsgType_kSearch] = searchMsg.Unmarshal
dispatcher.tempMap[internalPb.MsgType_kSearchResult] = searchResultMsg.Unmarshal
dispatcher.tempMap[internalPb.MsgType_kTimeTick] = timeTickMsg.Unmarshal
dispatcher.tempMap[internalPb.MsgType_kQueryNodeSegStats] = queryNodeSegStatsMsg.Unmarshal
}
func NewUnmarshalDispatcher() *UnmarshalDispatcher {

View File

@ -41,8 +41,9 @@ func startMaster(ctx context.Context) {
rootPath := conf.Config.Etcd.Rootpath
kvRootPath := path.Join(rootPath, "kv")
metaRootPath := path.Join(rootPath, "meta")
tsoRootPath := path.Join(rootPath, "timestamp")
svr, err := master.CreateServer(ctx, kvRootPath, metaRootPath, []string{etcdAddr})
svr, err := master.CreateServer(ctx, kvRootPath, metaRootPath, tsoRootPath, []string{etcdAddr})
masterServer = svr
if err != nil {
log.Print("create server failed", zap.Error(err))

View File

@ -206,7 +206,6 @@ func (container *colSegContainer) getSegmentStatistics() *internalpb.QueryNodeSe
}
statisticData = append(statisticData, &stat)
segment.recentlyModified = false
}
return &internalpb.QueryNodeSegStats{

View File

@ -56,20 +56,6 @@ func (iNode *insertNode) Operate(in []*Msg) []*Msg {
insertData.insertIDs[task.SegmentID] = append(insertData.insertIDs[task.SegmentID], task.RowIDs...)
insertData.insertTimestamps[task.SegmentID] = append(insertData.insertTimestamps[task.SegmentID], task.Timestamps...)
insertData.insertRecords[task.SegmentID] = append(insertData.insertRecords[task.SegmentID], task.RowData...)
// check if segment exists, if not, create this segment
if !(*iNode.container).hasSegment(task.SegmentID) {
collection, err := (*iNode.container).getCollectionByName(task.CollectionName)
if err != nil {
log.Println(err)
continue
}
err = (*iNode.container).addSegment(task.SegmentID, task.PartitionTag, collection.ID())
if err != nil {
log.Println(err)
continue
}
}
}
// 2. do preInsert

View File

@ -28,7 +28,7 @@ func (stNode *serviceTimeNode) Operate(in []*Msg) []*Msg {
}
// update service time
stNode.node.tSafe.setTSafe(serviceTimeMsg.timeRange.timestampMax)
stNode.node.tSafe = serviceTimeMsg.timeRange.timestampMax
return nil
}

View File

@ -14,7 +14,6 @@ import "C"
import (
"context"
"sync"
)
type QueryNode struct {
@ -23,7 +22,7 @@ type QueryNode struct {
QueryNodeID uint64
pulsarURL string
tSafe tSafe
tSafe Timestamp
container *container
@ -33,16 +32,6 @@ type QueryNode struct {
statsService *statsService
}
type tSafe interface {
getTSafe() Timestamp
setTSafe(t Timestamp)
}
type serviceTime struct {
tSafeMu sync.Mutex
time Timestamp
}
func NewQueryNode(ctx context.Context, queryNodeID uint64, pulsarURL string) *QueryNode {
segmentsMap := make(map[int64]*Segment)
collections := make([]*Collection, 0)
@ -52,15 +41,13 @@ func NewQueryNode(ctx context.Context, queryNodeID uint64, pulsarURL string) *Qu
segments: segmentsMap,
}
var tSafe tSafe = &serviceTime{}
return &QueryNode{
ctx: ctx,
QueryNodeID: queryNodeID,
pulsarURL: pulsarURL,
tSafe: tSafe,
tSafe: 0,
container: &container,
@ -86,15 +73,3 @@ func (node *QueryNode) Start() {
func (node *QueryNode) Close() {
// TODO: close services
}
func (st *serviceTime) getTSafe() Timestamp {
st.tSafeMu.Lock()
defer st.tSafeMu.Unlock()
return st.time
}
func (st *serviceTime) setTSafe(t Timestamp) {
st.tSafeMu.Lock()
st.time = t
st.tSafeMu.Unlock()
}

View File

@ -151,7 +151,6 @@ func (s *Segment) segmentInsert(offset int64, entityIDs *[]UniqueID, timestamps
return errors.New("Insert failed, error code = " + strconv.Itoa(int(status)))
}
s.recentlyModified = true
return nil
}

View File

@ -3,7 +3,6 @@ package reader
import (
"context"
"fmt"
"log"
"strconv"
"time"
@ -14,55 +13,35 @@ import (
type statsService struct {
ctx context.Context
pulsarURL string
msgStream *msgstream.MsgStream
msgStream *msgstream.PulsarMsgStream
container *container
}
func newStatsService(ctx context.Context, container *container, pulsarURL string) *statsService {
func newStatsService(ctx context.Context, container *container, pulsarAddress string) *statsService {
// TODO: add pulsar message stream init
return &statsService{
ctx: ctx,
pulsarURL: pulsarURL,
msgStream: nil,
container: container,
}
}
func (sService *statsService) start() {
const (
receiveBufSize = 1024
sleepMillisecondTime = 1000
)
// start pulsar
producerChannels := []string{"statistic"}
statsStream := msgstream.NewPulsarMsgStream(sService.ctx, receiveBufSize)
statsStream.SetPulsarCient(sService.pulsarURL)
statsStream.CreatePulsarProducers(producerChannels)
var statsMsgStream msgstream.MsgStream = statsStream
sService.msgStream = &statsMsgStream
(*sService.msgStream).Start()
// start service
sleepMillisecondTime := 1000
fmt.Println("do segments statistic in ", strconv.Itoa(sleepMillisecondTime), "ms")
for {
select {
case <-sService.ctx.Done():
return
case <-time.After(sleepMillisecondTime * time.Millisecond):
default:
time.Sleep(time.Duration(sleepMillisecondTime) * time.Millisecond)
sService.sendSegmentStatistic()
}
}
}
func (sService *statsService) sendSegmentStatistic() {
statisticData := (*sService.container).getSegmentStatistics()
var statisticData = (*sService.container).getSegmentStatistics()
// fmt.Println("Publish segment statistic")
// fmt.Println(statisticData)
@ -70,15 +49,5 @@ func (sService *statsService) sendSegmentStatistic() {
}
func (sService *statsService) publicStatistic(statistic *internalpb.QueryNodeSegStats) {
var msg msgstream.TsMsg = &msgstream.QueryNodeSegStatsMsg{
QueryNodeSegStats: *statistic,
}
var msgPack = msgstream.MsgPack{
Msgs: []*msgstream.TsMsg{&msg},
}
err := (*sService.msgStream).Produce(&msgPack)
if err != nil {
log.Println(err)
}
// TODO: publish statistic
}

View File

@ -1,10 +1,12 @@
package tsoutil
import (
"fmt"
"path"
"time"
"github.com/zilliztech/milvus-distributed/internal/kv"
gparams "github.com/zilliztech/milvus-distributed/internal/util/paramtableutil"
"go.etcd.io/etcd/clientv3"
)
@ -25,10 +27,25 @@ func ParseTS(ts uint64) (time.Time, uint64) {
return physicalTime, logical
}
func NewTSOKVBase(etcdAddr []string, tsoRoot, subPath string) *kv.EtcdKV {
func NewTSOKVBase(subPath string) *kv.EtcdKV {
etcdAddr, err := gparams.GParams.Load("etcd.address")
if err != nil {
panic(err)
}
etcdPort, err := gparams.GParams.Load("etcd.port")
if err != nil {
panic(err)
}
etcdAddr = etcdAddr + ":" + etcdPort
fmt.Println("etcdAddr ::: ", etcdAddr)
client, _ := clientv3.New(clientv3.Config{
Endpoints: etcdAddr,
Endpoints: []string{etcdAddr},
DialTimeout: 5 * time.Second,
})
return kv.NewEtcdKV(client, path.Join(tsoRoot, subPath))
etcdRootPath, err := gparams.GParams.Load("etcd.rootpath")
if err != nil {
panic(err)
}
return kv.NewEtcdKV(client, path.Join(etcdRootPath, subPath))
}