milvus/internal/util/indexcgowrapper/codec_index_test.go
Jiquan Long 3f46c6d459
feat: support inverted index (#28783)
issue: https://github.com/milvus-io/milvus/issues/27704

Add inverted index for some data types in Milvus. This index type can
save a lot of memory compared to loading all data into RAM and speed up
the term query and range query.

Supported: `INT8`, `INT16`, `INT32`, `INT64`, `FLOAT`, `DOUBLE`, `BOOL`
and `VARCHAR`.

Not supported: `ARRAY` and `JSON`.

Note:
- The inverted index for `VARCHAR` is not designed to serve full-text
search now. We will treat every row as a whole keyword instead of
tokenizing it into multiple terms.
- The inverted index don't support retrieval well, so if you create
inverted index for field, those operations which depend on the raw data
will fallback to use chunk storage, which will bring some performance
loss. For example, comparisons between two columns and retrieval of
output fields.

The inverted index is very easy to be used.

Taking below collection as an example:

```python
fields = [
		FieldSchema(name="pk", dtype=DataType.VARCHAR, is_primary=True, auto_id=False, max_length=100),
		FieldSchema(name="int8", dtype=DataType.INT8),
		FieldSchema(name="int16", dtype=DataType.INT16),
		FieldSchema(name="int32", dtype=DataType.INT32),
		FieldSchema(name="int64", dtype=DataType.INT64),
		FieldSchema(name="float", dtype=DataType.FLOAT),
		FieldSchema(name="double", dtype=DataType.DOUBLE),
		FieldSchema(name="bool", dtype=DataType.BOOL),
		FieldSchema(name="varchar", dtype=DataType.VARCHAR, max_length=1000),
		FieldSchema(name="random", dtype=DataType.DOUBLE),
		FieldSchema(name="embeddings", dtype=DataType.FLOAT_VECTOR, dim=dim),
]
schema = CollectionSchema(fields)
collection = Collection("demo", schema)
```

Then we can simply create inverted index for field via:

```python
index_type = "INVERTED"
collection.create_index("int8", {"index_type": index_type})
collection.create_index("int16", {"index_type": index_type})
collection.create_index("int32", {"index_type": index_type})
collection.create_index("int64", {"index_type": index_type})
collection.create_index("float", {"index_type": index_type})
collection.create_index("double", {"index_type": index_type})
collection.create_index("bool", {"index_type": index_type})
collection.create_index("varchar", {"index_type": index_type})
```

Then, term query and range query on the field can be speed up
automatically by the inverted index:

```python
result = collection.query(expr='int64 in [1, 2, 3]', output_fields=["pk"])
result = collection.query(expr='int64 < 5', output_fields=["pk"])
result = collection.query(expr='int64 > 2997', output_fields=["pk"])
result = collection.query(expr='1 < int64 < 5', output_fields=["pk"])
```

---------

Signed-off-by: longjiquan <jiquan.long@zilliz.com>
2023-12-31 19:50:47 +08:00

331 lines
8.0 KiB
Go

package indexcgowrapper
import (
"math/rand"
"os"
"strconv"
"testing"
"github.com/stretchr/testify/assert"
"github.com/milvus-io/milvus-proto/go-api/v2/schemapb"
"github.com/milvus-io/milvus/internal/proto/indexpb"
"github.com/milvus-io/milvus/internal/storage"
"github.com/milvus-io/milvus/pkg/common"
"github.com/milvus-io/milvus/pkg/util/funcutil"
"github.com/milvus-io/milvus/pkg/util/metric"
"github.com/milvus-io/milvus/pkg/util/paramtable"
)
func TestMain(m *testing.M) {
paramtable.Init()
exitCode := m.Run()
os.Exit(exitCode)
}
type indexTestCase struct {
dtype schemapb.DataType
typeParams map[string]string
indexParams map[string]string
}
func generateBoolArray(numRows int) []bool {
ret := make([]bool, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, rand.Int()%2 == 0)
}
return ret
}
func generateInt8Array(numRows int) []int8 {
ret := make([]int8, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, int8(rand.Int()))
}
return ret
}
func generateInt16Array(numRows int) []int16 {
ret := make([]int16, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, int16(rand.Int()))
}
return ret
}
func generateInt32Array(numRows int) []int32 {
ret := make([]int32, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, int32(rand.Int()))
}
return ret
}
func generateInt64Array(numRows int) []int64 {
ret := make([]int64, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, int64(rand.Int()))
}
return ret
}
func generateFloat32Array(numRows int) []float32 {
ret := make([]float32, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, rand.Float32())
}
return ret
}
func generateFloat64Array(numRows int) []float64 {
ret := make([]float64, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, rand.Float64())
}
return ret
}
func generateStringArray(numRows int) []string {
ret := make([]string, 0, numRows)
for i := 0; i < numRows; i++ {
ret = append(ret, funcutil.GenRandomStr())
}
return ret
}
func generateFloatVectors(numRows, dim int) []float32 {
total := numRows * dim
ret := make([]float32, 0, total)
for i := 0; i < total; i++ {
ret = append(ret, rand.Float32())
}
return ret
}
func generateBinaryVectors(numRows, dim int) []byte {
total := (numRows * dim) / 8
ret := make([]byte, total)
_, err := rand.Read(ret)
if err != nil {
panic(err)
}
return ret
}
func genFieldData(dtype schemapb.DataType, numRows, dim int) storage.FieldData {
switch dtype {
case schemapb.DataType_Bool:
return &storage.BoolFieldData{
Data: generateBoolArray(numRows),
}
case schemapb.DataType_Int8:
return &storage.Int8FieldData{
Data: generateInt8Array(numRows),
}
case schemapb.DataType_Int16:
return &storage.Int16FieldData{
Data: generateInt16Array(numRows),
}
case schemapb.DataType_Int32:
return &storage.Int32FieldData{
Data: generateInt32Array(numRows),
}
case schemapb.DataType_Int64:
return &storage.Int64FieldData{
Data: generateInt64Array(numRows),
}
case schemapb.DataType_Float:
return &storage.FloatFieldData{
Data: generateFloat32Array(numRows),
}
case schemapb.DataType_Double:
return &storage.DoubleFieldData{
Data: generateFloat64Array(numRows),
}
case schemapb.DataType_String:
return &storage.StringFieldData{
Data: generateStringArray(numRows),
}
case schemapb.DataType_VarChar:
return &storage.StringFieldData{
Data: generateStringArray(numRows),
}
case schemapb.DataType_BinaryVector:
return &storage.BinaryVectorFieldData{
Dim: dim,
Data: generateBinaryVectors(numRows, dim),
}
case schemapb.DataType_FloatVector:
return &storage.FloatVectorFieldData{
Data: generateFloatVectors(numRows, dim),
Dim: dim,
}
default:
return nil
}
}
func genScalarIndexCases(dtype schemapb.DataType) []indexTestCase {
return []indexTestCase{
{
dtype: dtype,
typeParams: nil,
indexParams: map[string]string{
common.IndexTypeKey: "sort",
},
},
{
dtype: dtype,
typeParams: nil,
indexParams: map[string]string{
common.IndexTypeKey: "flat",
},
},
}
}
func genStringIndexCases(dtype schemapb.DataType) []indexTestCase {
return []indexTestCase{
{
dtype: dtype,
typeParams: nil,
indexParams: map[string]string{
common.IndexTypeKey: "sort",
},
},
{
dtype: dtype,
typeParams: nil,
indexParams: map[string]string{
common.IndexTypeKey: "marisa-trie",
},
},
}
}
func genFloatVecIndexCases(dtype schemapb.DataType) []indexTestCase {
return []indexTestCase{
{
dtype: dtype,
typeParams: nil,
indexParams: map[string]string{
common.IndexTypeKey: IndexFaissIVFPQ,
common.MetricTypeKey: metric.L2,
common.DimKey: strconv.Itoa(dim),
"nlist": strconv.Itoa(nlist),
"m": strconv.Itoa(m),
"nbits": strconv.Itoa(nbits),
},
},
{
dtype: dtype,
typeParams: nil,
indexParams: map[string]string{
common.IndexTypeKey: IndexFaissIVFFlat,
common.MetricTypeKey: metric.L2,
common.DimKey: strconv.Itoa(dim),
"nlist": strconv.Itoa(nlist),
},
},
}
}
func genBinaryVecIndexCases(dtype schemapb.DataType) []indexTestCase {
return []indexTestCase{
{
dtype: dtype,
typeParams: nil,
indexParams: map[string]string{
common.IndexTypeKey: IndexFaissBinIVFFlat,
common.MetricTypeKey: metric.JACCARD,
common.DimKey: strconv.Itoa(dim),
"nlist": strconv.Itoa(nlist),
"nbits": strconv.Itoa(nbits),
},
},
}
}
func genTypedIndexCase(dtype schemapb.DataType) []indexTestCase {
switch dtype {
case schemapb.DataType_Bool:
return genScalarIndexCases(dtype)
case schemapb.DataType_Int8:
return genScalarIndexCases(dtype)
case schemapb.DataType_Int16:
return genScalarIndexCases(dtype)
case schemapb.DataType_Int32:
return genScalarIndexCases(dtype)
case schemapb.DataType_Int64:
return genScalarIndexCases(dtype)
case schemapb.DataType_Float:
return genScalarIndexCases(dtype)
case schemapb.DataType_Double:
return genScalarIndexCases(dtype)
case schemapb.DataType_String:
return genScalarIndexCases(dtype)
case schemapb.DataType_VarChar:
return genStringIndexCases(dtype)
case schemapb.DataType_BinaryVector:
return genBinaryVecIndexCases(dtype)
case schemapb.DataType_FloatVector:
return genFloatVecIndexCases(dtype)
default:
return nil
}
}
func genIndexCase() []indexTestCase {
dtypes := []schemapb.DataType{
schemapb.DataType_Bool,
schemapb.DataType_Int8,
schemapb.DataType_Int16,
schemapb.DataType_Int32,
schemapb.DataType_Int64,
schemapb.DataType_Float,
schemapb.DataType_Double,
schemapb.DataType_String,
schemapb.DataType_VarChar,
schemapb.DataType_BinaryVector,
schemapb.DataType_FloatVector,
}
var ret []indexTestCase
for _, dtype := range dtypes {
ret = append(ret, genTypedIndexCase(dtype)...)
}
return ret
}
func genStorageConfig() *indexpb.StorageConfig {
params := paramtable.Get()
return &indexpb.StorageConfig{
Address: params.MinioCfg.Address.GetValue(),
AccessKeyID: params.MinioCfg.AccessKeyID.GetValue(),
SecretAccessKey: params.MinioCfg.SecretAccessKey.GetValue(),
BucketName: params.MinioCfg.BucketName.GetValue(),
RootPath: params.MinioCfg.RootPath.GetValue(),
IAMEndpoint: params.MinioCfg.IAMEndpoint.GetValue(),
UseSSL: params.MinioCfg.UseSSL.GetAsBool(),
UseIAM: params.MinioCfg.UseIAM.GetAsBool(),
}
}
func TestCgoIndex(t *testing.T) {
for _, testCase := range genIndexCase() {
index, err := NewCgoIndex(testCase.dtype, testCase.typeParams, testCase.indexParams)
assert.NoError(t, err, testCase)
dataset := GenDataset(genFieldData(testCase.dtype, nb, dim))
assert.NoError(t, index.Build(dataset), testCase)
blobs, err := index.Serialize()
assert.NoError(t, err, testCase)
copyIndex, err := NewCgoIndex(testCase.dtype, testCase.typeParams, testCase.indexParams)
assert.NoError(t, err, testCase)
assert.NoError(t, copyIndex.Load(blobs), testCase)
}
}