groot

@iamcodingcat Most of the screen-shots are focus on search latency, it doens't help much. A better way is: **focus on a short time of search spike**, observe these metrics: -...

@iamcodingcat Could you use describe_collection to print the informations of the two collection: "user_embedding_2025_12_16_2025_12_17_03_43" and "user_embedding_2025_12_17_2025_12_18_03_40" ``` print(client.describe_collection(collection_name="user_embedding_2025_12_16_2025_12_17_03_43")) print(client.describe_collection(collection_name="user_embedding_2025_12_17_2025_12_18_03_40")) ```

@iamcodingcat Seems all the query requests ran into timeout after 2025/12/18 01:52:51. Could you show me how you invoked the client.query() interface after 2025/12/18 01:52:51? Did you set the embedding...

All the configurations are described here: https://milvus.io/docs/v2.5.x/configure_querynode.md#queryNodecachewarmup Actually, the description is mainly copied from the comments in the milvus.yaml. - "async", when you call load_collection(), load_collection() will wait untill all...

> > 是的，proto就够了。 > > https://github.com/milvus-io/milvus-proto 这是我们的API的repo > > 可以参考Java的repo > > 已经做好了，可以直接使用了， 你们可以review一下，可以作为 社区的scala版本使用，有bug 和疑问再找我。 但是好像比较重要的 bulkwriter 并没有在proto里面，这块怎么处理？ libraryDependencies += "io.github.mullerhai" % "storch-milvus_3" % "0.1.1-2.6.1" https://central.sonatype.com/artifact/io.github.mullerhai/storch-milvus_3 > > io.github.mullerhai...

可以参考这个目录下面的：https://github.com/milvus-io/milvus-sdk-java/tree/master/sdk-core/src/main/java/io/milvus/v2 目前java sdk里有两套封装，分别是V1和V2，看V2目录里的即可

另外，milvus-proto是通过git submodule的形式引入milvus-sdk-java项目的。在这个目录下面可以看到有"milvus-proto @ 16218b1"直接指向milvus-proto项目： https://github.com/milvus-io/milvus-sdk-java/tree/master/sdk-core/src/main 然后在本地的milvus-sdk-java项目路经下执行 git submodule update --init 命令会把milvus-proto下载下来。最后再由protobuf-maven-plugin插件将其生成接口原始代码: https://github.com/milvus-io/milvus-sdk-java/blob/4326fd0e654c270b90e2165fdf0456d29ef27d69/pom.xml#L230 我不知scalar项目里会如何做，不过路子应该是差不多的，不需要专门把milvus-proto发一个包到仓库。

/lgtm /approve

In entity_helper.py, the pack_field_value_to_field_data() method use "+=" to append bytes to `FieldData.vectors.int8_vector` https://github.com/milvus-io/pymilvus/blob/c6459680ea29da06da04aa5b1bd66d5ad1e95431/pymilvus/client/entity_helper.py#L439 `FieldData.vectors.int8_vector` is bytes. Normally, a bytes object can be appended without memory problem. But the `FieldData.vectors.int8_vector` is...

The similar problem with binary vector: ``` import time import random import numpy as np from pymilvus import ( connections, utility, FieldSchema, CollectionSchema, DataType, Collection, AnnSearchRequest, WeightedRanker, ) COLLECTION_NAME =...