Multi-endpoint support

[zhicwu]

Implements basic load-balancing and failover mentioned in #894. Key changes:

• the ability to connect to URI
• optional health-check and node-discovery(using system.clusters table)
• load-balancing policies: custom, firstAlive, random and roundRobin (server weight, shard num/weight, and replica are currently not considered)
• failover for connect errors(including connect timed out)

Configuration:

Option	Description	Default Value	Remark
auto_discovery	Whether to enable node discovery	false
node_discovery_interval	Node discovery interval in milliseconds, 0 for one-time discovery	0
node_discovery_limit	Maximum number of nodes can be discovered at a time	100
load_balancing_policy	Load balancing policy	''	use 'firstAlive', 'random', 'roundRobin', or full qualified class name implementing ClickHouseLoadBlanacingPolicy
load_balancing_tags	Load balancing tags for filtering nodes	''
health_check_method	Health check method	SELECT_ONE	or PING which might be faster but may not work with middleware like chproxy
health_check_interval	Health check interval in milliseconds, 0 for one-time health check	0
check_all_nodes	Whether all nodes should be checked or just faulty nodes	false
node_check_interval	Minimum interval in milliseconds required for checking node status	0
node_group_size	Maximum number of nodes can be used in operation at a time	50
failover	Maximum times failover allowed	0
retry	Maximum times retry allowed	0

Performance:

Usage:

• Java client

  try (ClickHouseClient client = ClickHouseClient.newInstance(ClickHouseProtocol.HTTP);
          ClickHouseResponse response = client.connect(
                  "http://(grpc://myoffice/mydb),localhost,{https://explorer@play.clickhouse.com:443}/system?failover=2")
                  .query("select 1").executeAndWait()) {
      Assert.assertEquals(response.firstRecord().getValue(0).asInteger(), 1);
  }

• JDBC driver

  try (Connection conn = DriverManager.getConnection(
          "jdbc:ch://127.0.0.1:8128,(tcp://myhome?docker_cli_path=/usr/local/bin/docker),grpc://myhome/my/test/dbx1?decompress=0&failover=2", "default", "")) {
      try (Statement stmt = conn.createStatement();
          ResultSet rs = stmt.executeQuery("select 1")) {
          Assert.assertTrue(rs.next());
          Assert.assertEquals(rs.getInt(1), 1);
      }
  }

[mzitnik]

@zhicwu what was changed that the build it falling?

[zhicwu]

@zhicwu what was changed that the build it falling?

The initial build failure was related to how we handle null parameter for insertion. It was impacted by fundamental change s being made to pass ClickHouseConfig among ClickHouseNode, ClickHouseClient, and ClickHouseRequest. This resulted in incorrect format(TabSepartedWithNamesAndTypes instead of TabSeprated) used in testInsertDefaultValue.

[zhicwu]

The build failure(session locked) happens occasionally in the past(see ClickHouse/ClickHouse#35948), which is irrelevant to this PR.

[Yuzer-Ly]

@zhicwu I am using the driver version 0.4.6, Why does each request use the same endpoint? and it changed when i restart my application. my conf like that:

spring:
  datasource:
     dynamic:
         datasource:
             clickhouse:
                url: jdbc:ch://ip1:8123,ip2:8123,ip3:8123?load_balancing_policy=random
                username: username
                password: password
                driver-class-name: com.clickhouse.jdbc.ClickHouseDriver
                hikari:
                   connection-timeout: 30000
                   idle-timeout: 60000
                   minimum-idle: 5
                   maximum-pool-size: 50