Back up and Restore Data Using APIs

Milvus Backup provides data backup and restoration features to ensure the security of your Milvus data.

Obtain Milvus Backup

You can either download the compiled binary or build from the source.

To download the compiled binary, go to the release page, where you can find all official releases. Remember, always use the binaries in the release marked as Latest.

To compile from the source, do as follows:

git clone git@github.com:zilliztech/milvus-backup.git
go get
go build

Prepare configuration file

Download the example configuration file and tailor it to fit your needs.

Then create a folder alongside the downloaded or built Milvus Backup binary, name the folder configs, and place the configuration file inside the configs folder.

Your folder structure should be similar to the following:

workspace
├── milvus-backup
└── configs
     └── backup.yaml

Because Milvus Backup cannot back up your data to a local path, ensure that Minio settings are correct when tailoring the configuration file.

Start up the API server

Then you can start the API server as follows:

./milvus-backup server

The API server listens on port 8080 by default. You can change it by running it with the -p flag. To start the API server listening on port 443, do as follows:

./milvus-backup server -p 443

You can access the Swagger UI using http://localhost:/api/v1/docs/index.html.

Prepare data

If you run an empty local Milvus instance listening on the default port 19530, use the example Python scripts to generate some data in your instance. Feel free to make necessary changes to the scripts to fit your needs.

Obtain the scripts. Then run the scripts to generate the data. Ensure that PyMilvus, the official Milvus Python SDK, has been installed.

python example/prepare_data.py

This step is optional. If you skip this, ensure that you already have some data in your Milvus instance.

Back up data

Note that running Milvus Backup against a Milvus instance will not normally affect the running of the instance. Your Milvus instance is fully functional during backup or restore.

Run the following command to create a backup. Change collection_names and backup_name if necessary.

curl --location --request POST 'http://localhost:8080/api/v1/create' \
--header 'Content-Type: application/json' \
--data-raw '{
  "async": true,
  "backup_name": "my_backup",
  "collection_names": [
    "hello_milvus"
  ]
}'

Once the command is executed, you can list the backups in the bucket specified in the Minio settings as follows:

curl --location --request GET 'http://localhost:8080/api/v1/list' \
--header 'Content-Type: application/json'

And download the backup files as follows:

curl --location --request GET 'http://localhost:8080/api/v1/get_backup?backup_id=<test_backup_id>&backup_name=my_backup' \
--header 'Content-Type: application/json'

While running the above command, change backup_id and backup_name to the one returned by the list API.

Now, you can save the backup files to a safe place for restoration in the future, or upload them to Zilliz Cloud to create a managed vector database with your data. For details, refer to Migrate from Milvus to Zilliz Cloud.

Restore data

You can call the restore API command with a collection_suffix option to create a new collection by restoring the data from the backup. Change collection_names and backup_name if necessary.

curl --location --request POST 'http://localhost:8080/api/v1/restore' \
--header 'Content-Type: application/json' \
--data-raw '{
    "async": true,
    "collection_names": [
    "hello_milvus"
  ],
    "collection_suffix": "_recover",
    "backup_name":"my_backup"
}'

The collection_suffix option allows you to set a suffix for the new collection to be created. The above command will create a new collection called hello_milvus_recover in your Milvus instance.

If you prefer to restore the backed-up collection without changing its name, drop the collection before restoring it from the backup. You can now clean the data generated in Prepare data by running the following command.

python example/clean_data.py

Then run the following command to restore the data from the backup.

curl --location --request POST 'http://localhost:8080/api/v1/restore' \
--header 'Content-Type: application/json' \
--data-raw '{
    "async": true,
    "collection_names": [
    "hello_milvus"
  ],
    "collection_suffix": "",
    "backup_name":"my_backup"
}'

The restore process can be time-consuming depending on the size of the data to be restored. Therefore, all restore tasks are running asynchronously. You can check the status of a restore task by running:

curl --location --request GET 'http://localhost:8080/api/v1/get_restore?id=<test_restore_id>' \
--header 'Content-Type: application/json'

Remember to change test_restore_id to the one restored by the restore API.

Verify restored data

Once the restore completes, you can verify the restored data by indexing the restored collection as follows:

python example/verify_data.py

Note that the above script assumes that you have run the restore command with the -s flag and the suffix is set to -recover. Feel free to make necessary changes to the script to fit your need.