Cluster Discovery

Overview

ClickHouse's Cluster Discovery feature simplifies cluster configuration by allowing nodes to automatically discover and register themselves without the need for explicit definition in the configuration files. This is especially beneficial in cases where the manual definition of each node becomes cumbersome.

note

Cluster Discovery is an experimental feature and can be changed or removed in future versions. To enable it include the allow_experimental_cluster_discovery setting in your configuration file:

<clickhouse>
    <!-- ... -->
    <allow_experimental_cluster_discovery>1</allow_experimental_cluster_discovery>
    <!-- ... -->
</clickhouse>

Remote Servers Configuration

Traditional Manual Configuration

Traditionally, in ClickHouse, each shard and replica in the cluster needed to be manually specified in the configuration:

<remote_servers>
    <cluster_name>
        <shard>
            <replica>
                <host>node1</host>
                <port>9000</port>
            </replica>
            <replica>
                <host>node2</host>
                <port>9000</port>
            </replica>
        </shard>
        <shard>
            <replica>
                <host>node3</host>
                <port>9000</port>
            </replica>
            <replica>
                <host>node4</host>
                <port>9000</port>
            </replica>
        </shard>
    </cluster_name>
</remote_servers>

Using Cluster Discovery

With Cluster Discovery, rather than defining each node explicitly, you simply specify a path in ZooKeeper. All nodes that register under this path in ZooKeeper will be automatically discovered and added to the cluster.

<remote_servers>
    <cluster_name>
        <discovery>
            <path>/clickhouse/discovery/cluster_name</path>

            <!-- # Optional configuration parameters: -->

            <!-- ## Authentication credentials to access all other nodes in cluster: -->
            <!-- <user>user1</user> -->
            <!-- <password>pass123</password> -->
            <!-- ### Alternatively to password, interserver secret may be used: -->
            <!-- <secret>secret123</secret> -->

            <!-- ## Shard for current node (see below): -->
            <!-- <shard>1</shard> -->

            <!-- ## Observer mode (see below): -->
            <!-- <observer/> -->
        </discovery>
    </cluster_name>
</remote_servers>

If you want to specify a shard number for a particular node, you can include the <shard> tag within the <discovery> section:

for node1 and node2:

<discovery>
    <path>/clickhouse/discovery/cluster_name</path>
    <shard>1</shard>
</discovery>

for node3 and node4:

<discovery>
    <path>/clickhouse/discovery/cluster_name</path>
    <shard>2</shard>
</discovery>

Observer mode

Nodes configured in observer mode will not register themselves as replicas. They will solely observe and discover other active replicas in the cluster without actively participating. To enable observer mode, include the <observer/> tag within the <discovery> section:

<discovery>
    <path>/clickhouse/discovery/cluster_name</path>
    <observer/>
</discovery>

Use-Cases and Limitations

As nodes are added or removed from the specified ZooKeeper path, they are automatically discovered or removed from the cluster without the need for configuration changes or server restarts.

However, changes affect only cluster configuration, not the data or existing databases and tables.

Consider the following example with a cluster of 3 nodes:

<remote_servers>
    <default>
        <discovery>
            <path>/clickhouse/discovery/default_cluster</path>
        </discovery>
    </default>
</remote_servers>

SELECT * EXCEPT (default_database, errors_count, slowdowns_count, estimated_recovery_time, database_shard_name, database_replica_name)
FROM system.clusters WHERE cluster = 'default';

┌─cluster─┬─shard_num─┬─shard_weight─┬─replica_num─┬─host_name────┬─host_address─┬─port─┬─is_local─┬─user─┬─is_active─┐
│ default │         1 │            1 │           1 │ 92d3c04025e8 │ 172.26.0.5   │ 9000 │        0 │      │      ᴺᵁᴸᴸ │
│ default │         1 │            1 │           2 │ a6a68731c21b │ 172.26.0.4   │ 9000 │        1 │      │      ᴺᵁᴸᴸ │
│ default │         1 │            1 │           3 │ 8e62b9cb17a1 │ 172.26.0.2   │ 9000 │        0 │      │      ᴺᵁᴸᴸ │
└─────────┴───────────┴──────────────┴─────────────┴──────────────┴──────────────┴──────┴──────────┴──────┴───────────┘

CREATE TABLE event_table ON CLUSTER default (event_time DateTime, value String)
ENGINE = ReplicatedMergeTree('/clickhouse/tables/event_table', '{replica}')
ORDER BY event_time PARTITION BY toYYYYMM(event_time);

INSERT INTO event_table ...

Then, we add a new node to the cluster, starting a new node with the same entry in the remote_servers section in a configuration file:

┌─cluster─┬─shard_num─┬─shard_weight─┬─replica_num─┬─host_name────┬─host_address─┬─port─┬─is_local─┬─user─┬─is_active─┐
│ default │         1 │            1 │           1 │ 92d3c04025e8 │ 172.26.0.5   │ 9000 │        0 │      │      ᴺᵁᴸᴸ │
│ default │         1 │            1 │           2 │ a6a68731c21b │ 172.26.0.4   │ 9000 │        1 │      │      ᴺᵁᴸᴸ │
│ default │         1 │            1 │           3 │ 8e62b9cb17a1 │ 172.26.0.2   │ 9000 │        0 │      │      ᴺᵁᴸᴸ │
│ default │         1 │            1 │           4 │ b0df3669b81f │ 172.26.0.6   │ 9000 │        0 │      │      ᴺᵁᴸᴸ │
└─────────┴───────────┴──────────────┴─────────────┴──────────────┴──────────────┴──────┴──────────┴──────┴───────────┘

The fourth node is participating in the cluster, but table event_table still exists only on the first three nodes:

SELECT hostname(), database, table FROM clusterAllReplicas(default, system.tables) WHERE table = 'event_table' FORMAT PrettyCompactMonoBlock

┌─hostname()───┬─database─┬─table───────┐
│ a6a68731c21b │ default  │ event_table │
│ 92d3c04025e8 │ default  │ event_table │
│ 8e62b9cb17a1 │ default  │ event_table │
└──────────────┴──────────┴─────────────┘

If you need to have tables replicated on all the nodes, you may use the Replicated database engine in alternative to cluster discovery.