fix: Fix multiple vector fields import (#33723) (#33724)

1. Fix dim mismatch with multi-vector fields and JSON import
2. Enhance: do not display file ID in GetImportResponse.

issue: https://github.com/milvus-io/milvus/issues/33681,
https://github.com/milvus-io/milvus/issues/33682

pr: https://github.com/milvus-io/milvus/pull/33723

---------

Signed-off-by: bigsheeper <yihao.dai@zilliz.com>
This commit is contained in:
yihao.dai 2024-06-10 21:55:55 +08:00 committed by GitHub
parent b71a404776
commit 396f8608dd
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 254 additions and 20 deletions

View File

@ -360,9 +360,9 @@ func (c *importChecker) checkGC(job ImportJob) {
job.GetState() != internalpb.ImportJobState_Failed {
return
}
GCRetention := Params.DataCoordCfg.ImportTaskRetention.GetAsDuration(time.Second)
cleanupTime := tsoutil.PhysicalTime(job.GetCleanupTs())
if time.Since(cleanupTime) >= GCRetention {
if time.Now().After(cleanupTime) {
GCRetention := Params.DataCoordCfg.ImportTaskRetention.GetAsDuration(time.Second)
log.Info("job has reached the GC retention", zap.Int64("jobID", job.GetJobID()),
zap.Time("cleanupTime", cleanupTime), zap.Duration("GCRetention", GCRetention))
tasks := c.imeta.GetTaskBy(WithJob(job.GetJobID()))

View File

@ -20,11 +20,13 @@ import (
"time"
"github.com/golang/protobuf/proto"
"go.uber.org/zap"
"github.com/milvus-io/milvus-proto/go-api/v2/commonpb"
"github.com/milvus-io/milvus-proto/go-api/v2/schemapb"
"github.com/milvus-io/milvus/internal/proto/datapb"
"github.com/milvus-io/milvus/internal/proto/internalpb"
"github.com/milvus-io/milvus/pkg/log"
"github.com/milvus-io/milvus/pkg/util/tsoutil"
)
@ -46,8 +48,11 @@ func UpdateJobState(state internalpb.ImportJobState) UpdateJobAction {
job.(*importJob).ImportJob.RequestedDiskSize = 0
// set cleanup ts
dur := Params.DataCoordCfg.ImportTaskRetention.GetAsDuration(time.Second)
cleanupTs := tsoutil.ComposeTSByTime(time.Now().Add(dur), 0)
cleanupTime := time.Now().Add(dur)
cleanupTs := tsoutil.ComposeTSByTime(cleanupTime, 0)
job.(*importJob).ImportJob.CleanupTs = cleanupTs
log.Info("set import job cleanup ts", zap.Int64("jobID", job.GetJobID()),
zap.Time("cleanupTime", cleanupTime), zap.Uint64("cleanupTs", cleanupTs))
}
}
}

View File

@ -428,7 +428,7 @@ func GetTaskProgresses(jobID int64, imeta ImportMeta, meta *meta) []*internalpb.
}
for _, fileStat := range task.GetFileStats() {
progresses = append(progresses, &internalpb.ImportTaskProgress{
FileName: fileStat.GetImportFile().String(),
FileName: fmt.Sprintf("%v", fileStat.GetImportFile().GetPaths()),
FileSize: fileStat.GetFileSize(),
Reason: task.GetReason(),
Progress: progress,

View File

@ -34,7 +34,7 @@ type RowParser interface {
}
type rowParser struct {
dim int
id2Dim map[int64]int
id2Field map[int64]*schemapb.FieldSchema
name2FieldID map[string]int64
pkField *schemapb.FieldSchema
@ -45,17 +45,18 @@ func NewRowParser(schema *schemapb.CollectionSchema) (RowParser, error) {
id2Field := lo.KeyBy(schema.GetFields(), func(field *schemapb.FieldSchema) int64 {
return field.GetFieldID()
})
vecField, err := typeutil.GetVectorFieldSchema(schema)
if err != nil {
return nil, err
}
dim := int64(0)
if typeutil.IsVectorType(vecField.DataType) && !typeutil.IsSparseFloatVectorType(vecField.DataType) {
dim, err = typeutil.GetDim(vecField)
if err != nil {
return nil, err
id2Dim := make(map[int64]int)
for id, field := range id2Field {
if typeutil.IsVectorType(field.GetDataType()) && !typeutil.IsSparseFloatVectorType(field.GetDataType()) {
dim, err := typeutil.GetDim(field)
if err != nil {
return nil, err
}
id2Dim[id] = int(dim)
}
}
pkField, err := typeutil.GetPrimaryFieldSchema(schema)
if err != nil {
return nil, err
@ -75,7 +76,7 @@ func NewRowParser(schema *schemapb.CollectionSchema) (RowParser, error) {
delete(name2FieldID, dynamicField.GetName())
}
return &rowParser{
dim: int(dim),
id2Dim: id2Dim,
id2Field: id2Field,
name2FieldID: name2FieldID,
pkField: pkField,
@ -92,7 +93,7 @@ func (r *rowParser) wrapTypeError(v any, fieldID int64) error {
func (r *rowParser) wrapDimError(actualDim int, fieldID int64) error {
field := r.id2Field[fieldID]
return merr.WrapErrImportFailed(fmt.Sprintf("expected dim '%d' for field '%s' with type '%s', got dim '%d'",
r.dim, field.GetName(), field.GetDataType().String(), actualDim))
r.id2Dim[fieldID], field.GetName(), field.GetDataType().String(), actualDim))
}
func (r *rowParser) wrapArrayValueTypeError(v any, eleType schemapb.DataType) error {
@ -268,7 +269,7 @@ func (r *rowParser) parseEntity(fieldID int64, obj any) (any, error) {
if !ok {
return nil, r.wrapTypeError(obj, fieldID)
}
if len(arr) != r.dim/8 {
if len(arr) != r.id2Dim[fieldID]/8 {
return nil, r.wrapDimError(len(arr)*8, fieldID)
}
vec := make([]byte, len(arr))
@ -289,7 +290,7 @@ func (r *rowParser) parseEntity(fieldID int64, obj any) (any, error) {
if !ok {
return nil, r.wrapTypeError(obj, fieldID)
}
if len(arr) != r.dim {
if len(arr) != r.id2Dim[fieldID] {
return nil, r.wrapDimError(len(arr), fieldID)
}
vec := make([]float32, len(arr))
@ -311,7 +312,7 @@ func (r *rowParser) parseEntity(fieldID int64, obj any) (any, error) {
if !ok {
return nil, r.wrapTypeError(obj, fieldID)
}
if len(arr) != r.dim {
if len(arr) != r.id2Dim[fieldID] {
return nil, r.wrapDimError(len(arr), fieldID)
}
vec := make([]byte, len(arr)*2)
@ -333,7 +334,7 @@ func (r *rowParser) parseEntity(fieldID int64, obj any) (any, error) {
if !ok {
return nil, r.wrapTypeError(obj, fieldID)
}
if len(arr) != r.dim {
if len(arr) != r.id2Dim[fieldID] {
return nil, r.wrapDimError(len(arr), fieldID)
}
vec := make([]byte, len(arr)*2)

View File

@ -0,0 +1,228 @@
// Licensed to the LF AI & Data foundation under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package importv2
import (
"context"
"fmt"
"math/rand"
"os"
"strings"
"time"
"github.com/golang/protobuf/proto"
"github.com/samber/lo"
"go.uber.org/zap"
"github.com/milvus-io/milvus-proto/go-api/v2/commonpb"
"github.com/milvus-io/milvus-proto/go-api/v2/milvuspb"
"github.com/milvus-io/milvus-proto/go-api/v2/schemapb"
"github.com/milvus-io/milvus/internal/proto/internalpb"
"github.com/milvus-io/milvus/internal/util/importutilv2"
"github.com/milvus-io/milvus/pkg/common"
"github.com/milvus-io/milvus/pkg/log"
"github.com/milvus-io/milvus/pkg/util/funcutil"
"github.com/milvus-io/milvus/pkg/util/merr"
"github.com/milvus-io/milvus/pkg/util/metric"
"github.com/milvus-io/milvus/tests/integration"
)
func (s *BulkInsertSuite) testMultipleVectorFields() {
const (
rowCount = 10000
dim1 = 64
dim2 = 32
)
c := s.Cluster
ctx, cancel := context.WithTimeout(c.GetContext(), 600*time.Second)
defer cancel()
collectionName := "TestBulkInsert_MultipleVectorFields_" + funcutil.GenRandomStr()
schema := integration.ConstructSchema(collectionName, 0, true, &schemapb.FieldSchema{
FieldID: 100,
Name: integration.Int64Field,
IsPrimaryKey: true,
DataType: schemapb.DataType_Int64,
AutoID: true,
}, &schemapb.FieldSchema{
FieldID: 101,
Name: integration.FloatVecField,
DataType: schemapb.DataType_FloatVector,
TypeParams: []*commonpb.KeyValuePair{
{
Key: common.DimKey,
Value: fmt.Sprintf("%d", dim1),
},
},
}, &schemapb.FieldSchema{
FieldID: 102,
Name: integration.BFloat16VecField,
DataType: schemapb.DataType_BFloat16Vector,
TypeParams: []*commonpb.KeyValuePair{
{
Key: common.DimKey,
Value: fmt.Sprintf("%d", dim2),
},
},
})
schema.EnableDynamicField = true
marshaledSchema, err := proto.Marshal(schema)
s.NoError(err)
createCollectionStatus, err := c.Proxy.CreateCollection(ctx, &milvuspb.CreateCollectionRequest{
DbName: "",
CollectionName: collectionName,
Schema: marshaledSchema,
ShardsNum: common.DefaultShardsNum,
})
s.NoError(err)
s.Equal(int32(0), createCollectionStatus.GetCode())
// create index 1
createIndexStatus, err := c.Proxy.CreateIndex(ctx, &milvuspb.CreateIndexRequest{
CollectionName: collectionName,
FieldName: integration.FloatVecField,
IndexName: "_default_1",
ExtraParams: integration.ConstructIndexParam(dim1, integration.IndexFaissIvfFlat, metric.L2),
})
s.NoError(err)
s.Equal(int32(0), createIndexStatus.GetCode())
s.WaitForIndexBuilt(ctx, collectionName, integration.FloatVecField)
// create index 2
createIndexStatus, err = c.Proxy.CreateIndex(ctx, &milvuspb.CreateIndexRequest{
CollectionName: collectionName,
FieldName: integration.BFloat16VecField,
IndexName: "_default_2",
ExtraParams: integration.ConstructIndexParam(dim2, integration.IndexFaissIvfFlat, metric.L2),
})
s.NoError(err)
s.Equal(int32(0), createIndexStatus.GetCode())
s.WaitForIndexBuilt(ctx, collectionName, integration.BFloat16VecField)
// import
var files []*internalpb.ImportFile
err = os.MkdirAll(c.ChunkManager.RootPath(), os.ModePerm)
s.NoError(err)
switch s.fileType {
case importutilv2.Numpy:
importFile, err := GenerateNumpyFiles(c.ChunkManager, schema, rowCount)
s.NoError(err)
importFile.Paths = lo.Filter(importFile.Paths, func(path string, _ int) bool {
return !strings.Contains(path, "$meta")
})
files = []*internalpb.ImportFile{importFile}
case importutilv2.JSON:
rowBasedFile := c.ChunkManager.RootPath() + "/" + "test.json"
GenerateJSONFile(s.T(), rowBasedFile, schema, rowCount)
defer os.Remove(rowBasedFile)
files = []*internalpb.ImportFile{
{
Paths: []string{
rowBasedFile,
},
},
}
case importutilv2.Parquet:
filePath := fmt.Sprintf("/tmp/test_%d.parquet", rand.Int())
err = GenerateParquetFile(filePath, schema, rowCount)
s.NoError(err)
defer os.Remove(filePath)
files = []*internalpb.ImportFile{
{
Paths: []string{
filePath,
},
},
}
}
importResp, err := c.Proxy.ImportV2(ctx, &internalpb.ImportRequest{
CollectionName: collectionName,
Files: files,
})
s.NoError(err)
s.Equal(int32(0), importResp.GetStatus().GetCode())
log.Info("Import result", zap.Any("importResp", importResp))
jobID := importResp.GetJobID()
err = WaitForImportDone(ctx, c, jobID)
s.NoError(err)
// load
loadStatus, err := c.Proxy.LoadCollection(ctx, &milvuspb.LoadCollectionRequest{
CollectionName: collectionName,
})
s.NoError(err)
s.Equal(commonpb.ErrorCode_Success, loadStatus.GetErrorCode())
s.WaitForLoad(ctx, collectionName)
segments, err := c.MetaWatcher.ShowSegments()
s.NoError(err)
s.NotEmpty(segments)
log.Info("Show segments", zap.Any("segments", segments))
// load refresh
loadStatus, err = c.Proxy.LoadCollection(ctx, &milvuspb.LoadCollectionRequest{
CollectionName: collectionName,
Refresh: true,
})
s.NoError(err)
s.Equal(commonpb.ErrorCode_Success, loadStatus.GetErrorCode())
s.WaitForLoadRefresh(ctx, "", collectionName)
// search vec 1
expr := fmt.Sprintf("%s > 0", integration.Int64Field)
nq := 10
topk := 10
roundDecimal := -1
params := integration.GetSearchParams(integration.IndexFaissIvfFlat, metric.L2)
searchReq := integration.ConstructSearchRequest("", collectionName, expr,
integration.FloatVecField, schemapb.DataType_FloatVector, nil, metric.L2, params, nq, dim1, topk, roundDecimal)
searchResult, err := c.Proxy.Search(ctx, searchReq)
err = merr.CheckRPCCall(searchResult, err)
s.NoError(err)
s.Equal(nq*topk, len(searchResult.GetResults().GetScores()))
// search vec 2
searchReq = integration.ConstructSearchRequest("", collectionName, expr,
integration.BFloat16VecField, schemapb.DataType_BFloat16Vector, nil, metric.L2, params, nq, dim2, topk, roundDecimal)
searchResult, err = c.Proxy.Search(ctx, searchReq)
err = merr.CheckRPCCall(searchResult, err)
s.NoError(err)
// s.Equal(nq*topk, len(searchResult.GetResults().GetScores())) // TODO: fix bf16vector search
}
func (s *BulkInsertSuite) TestMultipleVectorFields_JSON() {
s.fileType = importutilv2.JSON
s.testMultipleVectorFields()
}
func (s *BulkInsertSuite) TestMultipleVectorFields_Parquet() {
s.fileType = importutilv2.Parquet
s.testMultipleVectorFields()
}