跳到主要内容

Deploy a Databend Meta Service Cluster

tip

Expected deployment time: 5 minutes ⏱

databend-meta is a global service for the metadata(such as user, table schema etc.). It can be a single process service for testing, or a cluster of 3 or 5 nodes for production use.

databend-meta has to be deployed before booting up a databend-query cluster.

                               .~~~~~~~~~~~~~~~~.
                               !   MySQL        !
                               !   ClickHouse   !
                               !   REST API     !
 .------.                      '~~~~~~~~~~~~~~~~'
 | meta |                               ^
 '------'                               |
    ^                                   v
    |                          .----------------.
 .------.                      |                |
 | meta |<-------------------->| databend-query |
 '------'                      |                |
    |                          '----------------'
    v                                   ^
 .------.                               !
 | meta |                               v
 '------'                      .~~~~~~~~~~~~~~~~.
                               !                !
                               !     AWS S3     !
                               !                !
                               '~~~~~~~~~~~~~~~~'

1. Download​

You can find the latest binaries on the github release page or build from source.

mkdir databend && cd databend
curl -LJO https://github.com/datafuselabs/databend/releases/download/v0.7.32-nightly/databend-v0.7.32-nightly-x86_64-unknown-linux-musl.tar.gz
tar xzvf databend-v0.7.32-nightly-x86_64-unknown-linux-musl.tar.gz

You can find two executable files:

ls
# databend-meta databend-query

2. Deploy databend-meta​

tip

Standalone mode should NEVER be used in production. But do not worry. Anytime when needed, a standalone databand-meta can be extended to a cluster of 3 or 5 nodes with Cluster management API.

2.1 Standalone mode​

2.1.1 Create databend-meta.toml​

databend-meta.toml
log_dir = "metadata/_logs"
admin_api_address = "127.0.0.1:28101"
grpc_api_address = "127.0.0.1:9101"

[raft_config]
id = 1
single = true
raft_dir = "metadata/datas"

2.1.2 Start the databend-meta​

./databend-meta -c ./databend-meta.toml > meta.log 2>&1 &

2.1.3 Check databend-meta​

curl -I  http://127.0.0.1:28101/v1/health

Check the response is HTTP/1.1 200 OK.

2.2 Cluster mode​

In this chapter we will deploy a databend-meta cluster with 3 nodes.

First we are going to start up the first node in single mode, to form a single node cluster. Then join the other two nodes to the cluster to finally form a 3-nodes cluster.

tip

One of the 3 nodes will be elected to be a leader which serves write and read API. Other nodes are follower, and will redirect requests to the leader. Thus a databend-query can be configured to connect to any of the nodes in a cluster.

2.2.1 Create databend-meta-1.toml​

databend-meta-1.toml
log_dir            = "metadata/_logs1"
metric_api_address = "0.0.0.0:28100"
admin_api_address  = "0.0.0.0:28101"
grpc_api_address   = "0.0.0.0:9191"

[raft_config]
id                  = 1
raft_dir            = "metadata/datas1"
raft_api_port       = 28103
raft_listen_host    = "127.0.0.1"
raft_advertise_host = "localhost"

  • metric_api_address is the service for metrics collection.

  • admin_api_address is the service for retrieving cluster status.

  • grpc_api_address is the service for applications to write or read metadata.

  • raft_config.id is the globally unique id for this node; it is a u64.

  • raft_config.raft_dir is the local dir to store metadata, including raft log and state machine etc.

  • raft_config.raft_api_port,raft_config.raft_listen_host and raft_config.raft_advertise_host defines the service for internal raft communication. Application should never touch this port.

    raft_listen_host is the host the internal raft server listens on. raft_advertise_host is the host the internal raft client to connect to.

  • single tells the node to initialize a single node cluster if it is not initialized. Otherwise, this arg is just ignored.

For more information about config, see Configuration

2.2.2 Start the databend-meta node-1​

./databend-meta -c ./databend-meta-1.toml > meta1.log 2>&1 &

2.2.3 Check databend-meta node-1​

curl -I  http://127.0.0.1:28101/v1/health

Check the response is HTTP/1.1 200 OK.

2.2.4 Create config files for other 2 nodes:​

The config for other nodes are similar, except the single should be replaced with join, and the id has to be different.

databend-meta-2.toml
log_dir            = "metadata/_logs2"
metric_api_address = "0.0.0.0:28200"
admin_api_address  = "0.0.0.0:28201"
grpc_api_address   = "0.0.0.0:28202"

[raft_config]
id                  = 2
raft_dir            = "metadata/datas2"
raft_api_port       = 28203
raft_listen_host    = "127.0.0.1"
raft_advertise_host = "localhost"
join                = ["localhost:28103"]
databend-meta-3.toml
log_dir            = "metadata/_logs3"
metric_api_address = "0.0.0.0:28300"
admin_api_address  = "0.0.0.0:28301"
grpc_api_address   = "0.0.0.0:28302"

[raft_config]
id                  = 3
raft_dir            = "metadata/datas3"
raft_api_port       = 28303
raft_listen_host    = "127.0.0.1"
raft_advertise_host = "localhost"
join                = ["localhost:28103"]

The arg join specifies a list of raft addresses(<raft_advertise_host>:<raft_api_port>) of nodes in the existing cluster it wants to be joined to.

Start other nodes.​

./databend-meta -c ./databend-meta-2.toml > meta2.log 2>&1 &
./databend-meta -c ./databend-meta-3.toml > meta3.log 2>&1 &