Deploy Vald Agent on Docker

Vald is designed and implemented based on Cloud-Native architecture. However, there may be cases that want to use only Vald Agent without Kubernetes.

This article will show you how to deploy and run the Vald Agent on Docker. Fashion-mnist is used as an example dataset, same as Get Started.


  • docker: v19.0 ~
  • go: v1.14 ~
  • libhdf5 (only required for this tutorial.)

Hdf5 is required for this tutorial. If hdf5 is not installed, please install hdf5.

[Optional] Install hdf5
# yum
yum install -y hdf5-devel

# apt
apt-get install libhdf5-serial-dev

# homebrew
brew install hdf5


This chapter will show you how to deploy Vald Agent on docker.
This chapter will use NGT for the core engine of Vald Agent.

  1. Clone the vdaas/vald repository

    git clone
  2. Create config.yaml

    The configuration of Vald agent for docker is set using config.yaml
    You can also check the sample configuration.

    cat << EOF > config.yaml
    version: v0.0.0
    time_zone: JST
      logger: glg
      level: debug
      format: raw
        - name: agent-grpc
          port: 8081
          mode: GRPC
          probe_wait_time: "3s"
            shutdown_duration: "5s"
            handler_timeout: ""
            idle_timeout: ""
            read_header_timeout: ""
            read_timeout: ""
            write_timeout: ""
        - agent-grpc
        - agent-grpc
      full_shutdown_duration: 600s
        enabled: false
        # cert: /path/to/cert
        # key: /path/to/key
        # ca: /path/to/ca
      # vector dimension
      dimension: 784
      # bulk insert chunk size
      bulk_insert_chunk_size: 10
      # distance_type, which should be "l1", "l2" "angle", "hamming", "cosine", "normalizedangle", "normalizedcosine" or "jaccard"
      distance_type: l2
      # object_type, which should be "float" or "uint8"
      object_type: float
      # creation edge size
      creation_edge_size: 20
      # search edge size
      search_edge_size: 10
      # in-memory mode enabled
      enable_in_memory_mode: true
      # The limit duration of automatic indexing
      # auto_index_duration_limit should be 30m-6h for producation use. Below setting is a just example
      auto_index_duration_limit: 1m
      # Check duration of automatic indexing.
      # auto_index_check_duration be 10m-1h for producation use. Below setting is a just example
      auto_index_check_duration: 10s
      # The number of cache to trigger automatic indexing
      auto_index_length: 100
  3. Deploy Vald Agent on Docker

    To deploy Vald agent on docker with config.yaml, you can run below command.

    docker run -v $(pwd)/config.yaml:/etc/server/config.yaml -p 8081:8081 --rm -it vdaas/vald-agent-ngt
  4. Verify

    If the deployment success, you can confirm the output will be similar to below.

    2020-07-01 03:02:41	[INFO]:	maxprocs: Leaving GOMAXPROCS=4: CPU quota undefined
    2020-07-01 03:02:41	[INFO]:	service agent ngt v0.0.0 starting...
    2020-07-01 03:02:41	[INFO]:	daemon start
    2020-07-01 03:02:41	[INFO]:	server agent-grpc executing preStartFunc
    2020-07-01 12:02:41	[INFO]:	gRPC server agent-grpc starting on

Run using example code

  1. Download dataset

    In this tutorial. we use fashion-mnist as a dataset for indexing and search query.

    # move to working directory
    cd example/client/agent
    # download fashion-mnist testing dataset
  2. Running example

    Vald provides multiple language client libraries such as Go, Java, Node.js, Python, and so on.
    In this example, the fashion-mnist dataset will insert into the Vald and search using vald-client-go.

    We use example/client/agent/main.go to run the example. The example code is the same as running an example only Vald agent on Kubernetes. If you want to learn the detail of running an example, please refer to the tutorial of standalone Vald Agent on kubernetes.

    # run example
    go run main.go


    • We recommend you to run CreateIndex() after Insert() without waiting auto indexing.
  3. Clean Up

    Stop the Vald Agent docker container via Ctrl+C.