Current state: Accepted
ISSUE: https://github.com/milvus-io/milvus/issues/15604
PRs:
Keywords: bulk load, import
Released: with Milvus 2.1
Authors:
Summary
Import data by a shortcut to get better performance compared with insert().
Motivation
Typically, it cost several hours to insert one billion entities with 128-dimensional vectors. We need a new interface to do bulk load for the following purposes:
- import data from JSON format files. (first stage)
- import data from Numpy format files. (first stage)
- copy a collection within on Milvus 2.0 service. (second stage)
- copy a collection from one Milvus 2.0 service to another. (second stage)
- import data from Milvus 1.x to Milvus 2.0 (third stage)
- parquet/faiss files (TBD)
Some points to consider:
- JSON format is flexible, ideally, the import API ought to parse user's JSON files without asking user to reformat the files according to a strict rule.
- User can store scalar fields and vector fields in a JSON file, with row-based or column-based, the import API ought t support both of them.
A row-based example:
{
"table": {
"rows": [
{"id": 1, "year": 2021, "vector": [1.0, 1.1, 1.2]},
{"id": 2, "year": 2022, "vector": [2.0, 2.1, 2.2]},
{"id": 3, "year": 2023, "vector": [3.0, 3.1, 3.2]}
]
}
}
A column-based example:
{
"table": {
"columns": [
"id": [1, 2, 3],
"year": [2021, 2022, 2023],
"vector": [
[1.0, 1.1, 1.2],
[2.0, 2.1, 2.2],
[3.0, 3.1, 3.2]
]
]
}
}
- Numpy file is a binary format, we only treat it as vector data. Each Numpy file represents a vector field.
- Transferring a large file from client to server proxy to datanode is time-consume work and occupies too much network bandwidth, we will ask users to upload data files to MinIO/S3 where the datanode can access directly. Let the datanode read and parse files from MinIO/S3.
- Users may store scalar fields and vector fields in different format files. For example, store scalar fields in JSON files and store vector fields in Numpy files.
- The parameter of import API is easy to expand in future
SDK Interfaces
Based on the several points, we choose a JSON object as a parameter of python import() API, the API declaration will be like this:
def import(options)
The "options" is a JSON object which has the following format:
{
"data_source": { // required
"type": "minio", // required, currently only support "minio"/"s3"
"address": "localhost:9000", // optional, milvus server will use its minio setting if without this value
"accesskey_id": "minioadmin", // optional, milvus server will use its minio setting if without this value
"accesskey_secret": "minioadmin", // optional, milvus server will use its minio setting if without this value
"use_ssl": false, // optional, milvus server will use its minio setting if without this value
"bucket_name": "mybucket" // optional, milvus server will use its minio setting if without this value
},
"internal_data": { // optional, external_data or internal_data. (external files include json, npy, etc. internal files are exported by milvus)
"path": "xxx/xxx/xx", // relative path to the source storage where store the exported data
"collections_mapping": { // optional, give a new name to collection during importing.
"aaa": "bbb",
"ccc": "ddd"
}
},
"external_data": { // optional, external_data or internal_data. (external files include json, npy, etc. internal files are exported by milvus)
"target_collection": "xxx", // target collection name
"files": [ // required
{
"file": xxxx/xx.json, // required, relative path under the storage source defined by DataSource, currently support json/npy
"type": "row_based", // required, row_based or column_based
"fields_mapping": { // optional, specify the target fields which should be imported. Milvus will import all fields if this list is empty.
"table.rows.id": "uid",
"table.rows.year": "year",
"table.rows.vector": "vector",
}
},
{
"file": xxxx/xx.json, // required, relative path under the storage source defined by DataSource, currently support json/npy
"type": "column_based", // required, row_based or column_based
"fields_mapping": { // optional, specify the target fields which should be imported. Milvus will import all fields if this list is empty.
"table.columns.id": "uid",
"table.columns.year": "year",
"table.columns.vector": "vector",
}
},
{
"file": xxxx/xx.npy, // required, relative path under the storage source defined by DataSource, currently support json/npy
"type": "column_based", // required, row_based or column_based
"fields_mapping": { // optional, specify the target fields which should be imported. Milvus will import all fields if this list is empty.
"vector": "vector",
}
}
],
"default_fields": { // optional, use default value to fill some fields
"age": 0,
"weight": 0.0
},
}
}
Key fields of the JSON object:
- "data_source": contains the address and login methods of MinIO/S3. If the address and login methods are not provided, Milvus will use its MinIO/S3 configurations.
- "internal_data": reserved field for collection clone, not available in the first stage. It requires another API export().
- "external_data": for importing data from user's files. Let datanode where to read the data files and how to parse them.