upsert有数据但报错num_rows should be greater than 0

[Waddleswy]

报错信息里包含rows信息，但报的错误是num_rows = 0，collection更新失败

错误原因
UpsertRequest collectionName:xxxxxx failed, error code: 2, reason: num_rows should be greater than 0: invalid parameter[expected=invalid num_rows][actual=0]

错误信息
update error, rows:[{xxxxxxx}]"|"io.milvus.v2.exception.MilvusClientException: num_rows should be greater than 0: invalid parameter[expected=invalid num_rows][actual=0] at io.milvus.v2.utils.RpcUtils.handleResponse(RpcUtils.java:63) at io.milvus.v2.service.vector.VectorService.upsert(VectorService.java:172) at io.milvus.v2.client.MilvusClientV2.lambda$upsert$28(MilvusClientV2.java:478) at io.milvus.v2.utils.RpcUtils.retry(RpcUtils.java:100) at io.milvus.v2.client.MilvusClientV2.upsert(MilvusClientV2.java:478)

sdk版本
<dependency> <groupId>io.milvus</groupId> <artifactId>milvus-sdk-java</artifactId> <version>2.4.3</version>

[yhmo]

“num_rows should be greater than 0” 是服务端返回的报错，意思是请求里数据行数为0

这里是一个用java sdk 2.4.3 写的 upsert示例，不管是和milvus server v2.4版本还是v2.6版本都运行正常：

package io.milvus.v2;

import com.google.gson.*;
import io.milvus.v2.client.*;
import io.milvus.v2.common.ConsistencyLevel;
import io.milvus.v2.service.collection.request.CreateCollectionReq;
import io.milvus.v2.service.collection.request.DropCollectionReq;
import io.milvus.v2.service.vector.request.*;
import io.milvus.v2.service.vector.response.*;

import java.util.*;

public class SimpleExample {
    public static void main(String[] args) {

        ConnectConfig config = ConnectConfig.builder()
                .uri("http://localhost:19530")
                .build();
        MilvusClientV2 client = new MilvusClientV2(config);

        String collectionName = "java_sdk_example_simple_v2";
        // drop collection if exists
        client.dropCollection(DropCollectionReq.builder()
                .collectionName(collectionName)
                .build());

        // quickly create a collection with "id" field and "vector" field
        client.createCollection(CreateCollectionReq.builder()
                .collectionName(collectionName)
                .dimension(4)
                .build());
        System.out.printf("Collection '%s' created\n", collectionName);

        // insert some data
        List<JsonObject> rows = new ArrayList<>();
        Gson gson = new Gson();
        for (int i = 0; i < 100; i++) {
            JsonObject row = new JsonObject();
            row.addProperty("id", i);
            row.add("vector", gson.toJsonTree(new float[]{1.0f, 1.0f, 1.0f, 1.0f}));
            rows.add(row);
        }
        InsertResp insertR = client.insert(InsertReq.builder()
                .collectionName(collectionName)
                .data(rows)
                .build());
        System.out.printf("%d rows inserted\n", insertR.getInsertCnt());

        // get row count
        QueryResp countR = client.query(QueryReq.builder()
                .collectionName(collectionName)
                .filter("")
                .outputFields(Collections.singletonList("count(*)"))
                .consistencyLevel(ConsistencyLevel.STRONG)
                .build());
        System.out.printf("%d rows persisted\n", (long)countR.getQueryResults().get(0).getEntity().get("count(*)"));

        rows.clear();
        JsonObject updateRow = new JsonObject();
        updateRow.addProperty("id", 2);
        updateRow.add("vector", gson.toJsonTree(new float[]{999.0f, 999.0f, 999.0f, 999.0f}));
        rows.add(updateRow);
        UpsertResp upsertR = client.upsert(UpsertReq.builder()
                .collectionName(collectionName)
                .data(rows)
                .build());
        System.out.printf("%d rows upserted\n", upsertR.getUpsertCnt());
    }
}

运行之后可以用Attu看到id=2的那行数据成功被更新为[999, 999, 999, 999]

[Waddleswy]

我们这边的失败是偶发的，也不是每次都会出现，且报错信息里这个rows是有信息的，只是涉及我们内部数据我替换成了xxx。报的错是row num = 0，想问一下如果是row信息为空产生的错误那报错里为什么会有我们正确的rows呢？
"update error, rows:[{xxxxxxx}]"|"io.milvus.v2.exception.MilvusClientException: num_rows should be greater than 0: invalid parameter[expected=invalid num_rows][actual=0]
at io.milvus.v2.utils.RpcUtils.handleResponse(RpcUtils.java:63)

[yhmo]

从sdk的代码逻辑看，只有输入一个空的List时，才会返回num_rows should be greater than 0的报错

List<JsonObject> rows = new ArrayList<>();
UpsertResp upsertR = client.upsert(UpsertReq.builder()
        .collectionName(collectionName)
        .data(rows)
        .build());

其他方式，无论是输入一个空的row：

List<JsonObject> rows = new ArrayList<>();
JsonObject updateRow = new JsonObject();
rows.add(updateRow);
UpsertResp upsertR = client.upsert(UpsertReq.builder()
        .collectionName(collectionName)
        .data(rows)
        .build());

还是输入null值value：

List<JsonObject> rows = new ArrayList<>();
JsonObject updateRow = new JsonObject();
updateRow.add("id", null);
updateRow.add("vector", null);
rows.add(updateRow);
UpsertResp upsertR = client.upsert(UpsertReq.builder()
        .collectionName(collectionName)
        .data(rows)
        .build());

都会被sdk检查出来，报的错也不是num_rows should be greater than 0，而是诸如 "The field: id is not provided"，或者"Type mismatch for field 'id'"这种信息。

光看sdk代码逻辑，我没有看到有输入一个空的List给server端的可能性。如果你有能复现这个问题的精简代码的话，就比较容易定位问题原因。