ClickHouse JDBC Bridge

JDBC bridge for ClickHouse®. It acts as a stateless proxy passing queries from ClickHouse to external datasources. With this extension, you can run distributed query on ClickHouse across multiple datasources in real time, which in a way simplifies the process of building data pipelines for data warehousing, monitoring and integrity check etc.

Overview

Known Issues / Limitation

• Connection issue like jdbc-bridge is not running or connect timed out - see performance section and this issue for details

• Complex data types like Array and Tuple are currently not supported - they're treated as String

• Pushdown is not supported and query may execute twice because of type inferring

• Mutation is not fully supported - only insertion in simple cases

• Scripting is experimental

Quick Start

• Docker Compose

  git clone https://github.com/ClickHouse/clickhouse-jdbc-bridge.git
  cd clickhouse-jdbc-bridge/misc/quick-start
  docker-compose up -d
  ...
  docker-compose ps
  
             Name                         Command               State              Ports
  --------------------------------------------------------------------------------------------------
  quick-start_ch-server_1         /entrypoint.sh                   Up      8123/tcp, 9000/tcp, 9009/tcp
  quick-start_db-mariadb10_1      docker-entrypoint.sh mysqld      Up      3306/tcp
  quick-start_db-mysql5_1         docker-entrypoint.sh mysqld      Up      3306/tcp, 33060/tcp
  quick-start_db-mysql8_1         docker-entrypoint.sh mysqld      Up      3306/tcp, 33060/tcp
  quick-start_db-postgres13_1     docker-entrypoint.sh postgres    Up      5432/tcp
  quick-start_jdbc-bridge_1       /sbin/my_init                    Up      9019/tcp
  
  # issue below query, and you'll see "ch-server        1" returned
  docker-compose run ch-server clickhouse-client --query="select * from jdbc('self?datasource_column', 'select 1')"

• Docker CLI

  It's easier to get started using all-in-one docker image:

  # build all-in-one docker image
  git clone https://github.com/ClickHouse/clickhouse-jdbc-bridge.git
  cd clickhouse-jdbc-bridge
  docker build -t my/clickhouse-all-in-one -f all-in-one.Dockerfile .
  
  # start container in background
  docker run --rm -d --name ch-and-jdbc-bridge my/clickhouse-all-in-one
  
  # enter container to add datasource and issue query
  docker exec -it ch-and-jdbc-bridge bash
  
  cp /etc/clickhouse-jdbc-bridge/config/datasources/datasource.json.example \
      /etc/clickhouse-jdbc-bridge/config/datasources/ch-server.json
  
  # you're supposed to see "ch-server        1" returned from ClickHouse
  clickhouse-client --query="select * from jdbc('self?datasource_column', 'select 1')"

  Alternatively, if you prefer the hard way ;)

  # create a network for ClickHouse and JDBC brigde, so that they can communicate with each other
  docker network create ch-net --attachable
  # start the two containers
  docker run --rm -d --network ch-net --name jdbc-bridge --hostname jdbc-bridge clickhouse/jdbc-bridge
  docker run --rm -d --network ch-net --name ch-server --hostname ch-server \
      --entrypoint /bin/bash clickhouse/clickhouse-server -c \
      "echo '<clickhouse><jdbc_bridge><host>jdbc-bridge</host><port>9019</port></jdbc_bridge></clickhouse>' \
          > /etc/clickhouse-server/config.d/jdbc_bridge_config.xml && /entrypoint.sh"
  # add named datasource and query
  docker exec -it jdbc-bridge cp /app/config/datasources/datasource.json.example \
      /app/config/datasources/ch-server.json
  docker exec -it jdbc-bridge cp /app/config/queries/query.json.example \
      /app/config/queries/show-query-logs.json
  # issue below query, and you'll see "ch-server        1" returned from ClickHouse
  docker exec -it ch-server clickhouse-client \
      --query="select * from jdbc('self?datasource_column', 'select 1')"

• Debian/RPM Package

  Besides docker, you can download and install released Debian/RPM package on existing Linux system.

  Debian/Ubuntu

  apt update && apt install -y procps wget
  wget https://github.com/ClickHouse/clickhouse-jdbc-bridge/releases/download/v2.1.0/clickhouse-jdbc-bridge_2.1.0-1_all.deb
  apt install --no-install-recommends -f ./clickhouse-jdbc-bridge_2.1.0-1_all.deb
  clickhouse-jdbc-bridge

  CentOS/RHEL

  yum install -y wget
  wget https://github.com/ClickHouse/clickhouse-jdbc-bridge/releases/download/v2.1.0/clickhouse-jdbc-bridge-2.1.0-1.noarch.rpm
  yum localinstall -y clickhouse-jdbc-bridge-2.1.0-1.noarch.rpm
  clickhouse-jdbc-bridge

• Java CLI

  wget https://github.com/ClickHouse/clickhouse-jdbc-bridge/releases/download/v2.1.0/clickhouse-jdbc-bridge-2.1.0-shaded.jar
  # add named datasource
  wget -P config/datasources https://raw.githubusercontent.com/ClickHouse/clickhouse-jdbc-bridge/master/misc/quick-start/jdbc-bridge/config/datasources/ch-server.json
  # start jdbc bridge, and then issue below query in ClickHouse for testing
  # select * from jdbc('ch-server', 'select 1')
  java -jar clickhouse-jdbc-bridge-2.1.0-shaded.jar

Usage

In most cases, you'll use jdbc table function to query against external datasources:

select * from jdbc('<datasource>', '<schema>', '<query>')

schema is optional but others are mandatory. Please be aware that the query is in native format of the given datasource. For example, if the query is select * from some_table limit 10, it may work in MariaDB but not in PostgreSQL, as the latter one does not understand limit.

Assuming you started a test environment using docker-compose, please refer to examples below to get familiar with JDBC bridge.

• Data Source

  -- show datasources and usage
  select * from jdbc('', 'show datasources')
  -- access named datasource
  select * from jdbc('ch-server', 'select 1')
  -- adhoc datasource is NOT recommended for security reason
  select *
  from jdbc('jdbc:clickhouse://localhost:8123/system?compress=false&ssl=false&user=default', 'select 1')

• Schema

  By default, any adhoc query passed to JDBC bridge will be executed twice. The first run is for type inferring, while the second for retrieving results. Although metadata will be cached(for up to 5 minutes by default), executing same query twice could be a problem - that's where schema comes into play.

  -- inline schema
  select * from jdbc('ch-server', 'num UInt8, str String', 'select 1 as num, ''2'' as str')
  select * from jdbc('ch-server', 'num Nullable(Decimal(10,0)), Nullable(str FixedString(1)) DEFAULT ''x''', 'select 1 as num, ''2'' as str')
  -- named schema
  select * from jdbc('ch-server', 'query-log', 'show-query-logs')

• Query

  -- adhoc query
  select * from jdbc('ch-server', 'system', 'select * from query_log where user != ''default''')
  select * from jdbc('ch-server', 'select * from query_log where user != ''default''')
  select * from jdbc('ch-server', 'select * from system.query_log where user != ''default''')
  
  -- table query
  select * from jdbc('ch-server', 'system', 'query_log')
  select * from jdbc('ch-server', 'query_log')
  
  -- saved query
  select * from jdbc('ch-server', 'scripts/show-query-logs.sql')
  
  -- named query
  select * from jdbc('ch-server', 'show-query-logs')
  
  -- scripting
  select * from jdbc('script', '[1,2,3]')
  select * from jdbc('script', 'js', '[1,2,3]')
  select * from jdbc('script', 'scripts/one-two-three.js')

• Query Parameters

  select *
  from jdbc('ch-server?datasource_column&max_rows=1&fetch_size=1&one=1&two=2',
      'select {{one}} union all select {{ two }}')

  Query result:

  ┌─datasource─┬─1─┐
  │ ch-server  │ 1 │
  └────────────┴───┘

• JDBC Table

  drop table if exists system.test;
  create table system.test (
      a String,
      b UInt8
  ) engine=JDBC('ch-server', '', 'select user as a, is_initial_query as b from system.processes');

• JDBC Dictionary

  drop dictionary if exists system.dict_test;
  create dictionary system.dict_test
  (
      b UInt64 DEFAULT 0,
      a String
  ) primary key b
  SOURCE(CLICKHOUSE(HOST 'localhost' PORT 9000 USER 'default' TABLE 'test' DB 'system'))
  LIFETIME(MIN 82800 MAX 86400)
  LAYOUT(FLAT());

• Mutation

  -- use query parameter
  select * from jdbc('ch-server?mutation', 'drop table if exists system.test_table');
  select * from jdbc('ch-server?mutation', 'create table system.test_table(a String, b UInt8) engine=Memory()');
  select * from jdbc('ch-server?mutation', 'insert into system.test_table values(''a'', 1)');
  select * from jdbc('ch-server?mutation', 'truncate table system.test_table');
  
  -- use JDBC table engine
  drop table if exists system.test_table;
  create table system.test_table (
      a String,
      b UInt8
  ) engine=Memory();
  
  drop table if exists system.jdbc_table;
  create table system.jdbc_table (
      a String,
      b UInt8
  ) engine=JDBC('ch-server?batch_size=1000', 'system', 'test_table');
  
  insert into system.jdbc_table(a, b) values('a', 1);
  
  select * from system.test_table;

  Query result:

  ┌─a─┬─b─┐
  │ a │ 1 │
  └───┴───┘

• Monitoring

  You can use Prometheus to monitor metrics exposed by JDBC bridge.

  curl -v http://jdbc-bridge:9019/metrics

Configuration

• JDBC Driver

  By default, all JDBC drivers should be placed under drivers directory. You can override that by customizing driverUrls in datasource configuration file. For example:

  {
      "testdb": {
          "driverUrls": [
              "drivers/mariadb",
              "D:\\drivers\\mariadb",
              "/mnt/d/drivers/mariadb",
              "https://repo1.maven.org/maven2/org/mariadb/jdbc/mariadb-java-client/2.7.4/mariadb-java-client-2.7.4.jar"
          ],
          "driverClassName": "org.mariadb.jdbc.Driver",
          ...
      }
  }

• Named Data Source

  By default, named datasource is defined in configuration file in JSON format under config/datasources directory. You may check examples at misc/quick-start/jdbc-bridge/config/datasources. If you use modern editors like VSCode, you may find it's helpful to use JSON schema for validation and smart autocomplete.

• Saved Query

  Saved queries and scripts are under scripts directory by default. For example: show-query-logs.sql.

• Named Query

  Similar as named datasource, named queries are JSON configuration files under config/queries. You may refer to examples at misc/quick-start/jdbc-bridge/config/queries.

• Logging

  You can customize logging configuration in logging.properties.

• Vert.x

  If you're familiar with Vert.x, you can customize its configuration by changing config/httpd.json and config/vertx.json.

• Query Parameters

  All supported query parameters can be found at here. datasource_column=true can be simplied as datasource_column, for example:

  select * from jdbc('ch-server?datasource_column=true', 'select 1')
  
  select * from jdbc('ch-server?datasource_column', 'select 1')

• Timeout

  Couple of timeout settings you should be aware of:

  1. datasource timeout, for example: max_execution_time in MariaDB
  2. JDBC driver timeout, for example: connectTimeout and socketTimeout in MariaDB Connector/J
  3. JDBC bridge timeout, for examples: queryTimeout in config/server.json, and maxWorkerExecuteTime in config/vertx.json
  4. ClickHouse timeout like max_execution_time, keep_alive_timeout and http_receive_timeout etc.
  5. Client timeout, for example: socketTimeout in ClickHouse JDBC driver

Migration

• Upgrade to 2.x

  2.x is a complete re-write not fully compatible with older version. You'll have to re-define your datasources and update your queries accordingly.

Build

You can use Maven to build ClickHouse JDBC bridge, for examples:

git clone https://github.com/ClickHouse/clickhouse-jdbc-bridge.git
cd clickhouse-jdbc-bridge
# compile and run unit tests
mvn -Prelease verify
# release shaded jar, rpm and debian packages
mvn -Prelease package

In order to build docker images:

git clone https://github.com/ClickHouse/clickhouse-jdbc-bridge.git
cd clickhouse-jdbc-bridge
docker build -t my/clickhouse-jdbc-bridge .
# or if you want to build the all-ine-one image
docker build --build-arg revision=20.9.3 -f all-in-one.Dockerfile -t my/clickhouse-all-in-one .

Develop

JDBC bridge is extensible. You may take ConfigDataSource and ScriptDataSource as examples to create your own extension.

An extension for JDBC bridge is basically a Java class with 3 optional parts:

1. Extension Name

   By default, extension class name will be treated as name for the extension. However, you can declare a static member in your extension class to override that, for instance:

   public static final String EXTENSION_NAME = "myExtension";

2. Initialize Method

   Initialize method will be called once and only once at the time when loading your extension, for example:

   public static void initialize(ExtensionManager manager) {
       ...
   }

3. Instantiation Method

   In order to create instance of your extension, in general you should define a static method like below so that JDBC bridge knows how(besides walking through all possible constructors):

   public static MyExtension newInstance(Object... args) {
       ...
   }

Assume your extension class is com.mycompany.MyExtension, you can load it into JDBC bridge by:

• put your extension package(e.g. my-extension.jar) and required dependencies under extensions directory

• update server.json by adding your extension, for example

    ...
    "extensions": [
        ...
        {
            "class": "com.mycompany.MyExtension"
        }
    ]
    ...

Note: order of the extension matters. The first NamedDataSource extension will be set as default for all named datasources.

Performance

Below is a rough performance comparison to help you understand overhead caused by JDBC bridge as well as its stability. MariaDB, ClickHouse, and JDBC bridge are running on separated KVMs. ApacheBench(ab) is used on another KVM to simulate 20 concurrent users to issue same query 100,000 times after warm-up. Please refer to this in order to setup test environment and run tests by yourself.

Note: clickhouse(patched) is a patched version of ClickHouse server by disabling XDBC bridge health check. jdbc-dual on the other hand means dual instances of JDBC bridge managed by docker swarm on same KVM(due to limited resources ;).