Clickhouse源码分析-Replicated Database创建流程

前置准备

DDL：

CREATE DATABASE my_replicated_db

ENGINE = Replicated('/clickhouse/databases/my_replicated_db', '{shard}', '{replica}');

这里需要你提前启动1个clickhouse，1个clickhouse-keeper。

源码分析

断点：

b Interpreters/DDLWorker.cpp:146

b Databases/DatabaseReplicatedWorker.cpp:135

前台线程调用栈：

后台线程调用栈：

可以看到前台线程执行成功后直接返回给客户端：

主要看一下initializeReplication：

cpp 复制代码

    String active_path = fs::path(database->replica_path) / "active";
    String active_id = toString(ServerUUID::get());

database->replica_path = /clickhouse/databases/my_replicated_db/replicas/s1|r1

active_path = /clickhouse/databases/my_replicated_db/replicas/s1|r1/active

cpp 复制代码

    String log_ptr_str = zookeeper->get(database->replica_path + "/log_ptr");
    UInt32 our_log_ptr = parse<UInt32>(log_ptr_str);
    UInt32 max_log_ptr = parse<UInt32>(zookeeper->get(database->zookeeper_path + "/max_log_ptr"));
    logs_to_keep = parse<UInt32>(zookeeper->get(database->zookeeper_path + "/logs_to_keep"));

log_ptr_str = 0

database->zookeeper_path = /clickhouse/databases/my_replicated_db

max_log_ptr = 21956

logs_to_keep = 1000

cpp 复制代码

    UInt64 digest;
    String digest_str;
    UInt64 local_digest;
    if (zookeeper->tryGet(database->replica_path + "/digest", digest_str))
    {
        digest = parse<UInt64>(digest_str);
        std::lock_guard lock{database->metadata_mutex};
        local_digest = database->tables_metadata_digest;
    }

digest = 0

local_digest = 0

再看一下recoverLostReplica：

这一部分为了构建依赖：

/// Create all needed tables in a proper order

TablesDependencyGraph tables_dependencies("DatabaseReplicated (" + getDatabaseName() + ")");

for (const auto & [table_name, create_table_query] : table_name_to_metadata)

{

/// Note that table_name could contain a dot inside (e.g. .inner.1234-1234-1234-1234)

/// And QualifiedTableName::parseFromString doesn't handle this.

auto qualified_name = QualifiedTableName{.database = getDatabaseName(), .table = table_name};

auto query_ast = parseQueryFromMetadataInZooKeeper(table_name, create_table_query);

tables_dependencies.addDependencies(qualified_name, getDependenciesFromCreateQuery(getContext()->getGlobalContext(), qualified_name, query_ast, getContext()->getCurrentDatabase()).dependencies);

}

执行创建表语句：

cpp 复制代码

    //外层遍历每一个依赖块, 有依赖关系的表放在一个依赖块
    for (const auto & tables_to_create : tables_to_create_by_level)
    {
        for (const auto & table_id : tables_to_create)
        {
                .....

                /// Check larger comment in DatabaseOnDisk::createTableFromAST
                /// TL;DR applySettingsFromQuery will move the settings from engine to query level
                /// making it possible to overcome a backward incompatible change.
                InterpreterSetQuery::applySettingsFromQuery(query_ast, create_query_context);
                LOG_INFO(log, "Executing {}", query_ast->formatForLogging());
                InterpreterCreateQuery(query_ast, create_query_context).execute();

            //内存执行有依赖关系表的创建
        }

        runner.waitForAllToFinishAndRethrowFirstError();
    }

未完待续......

奇怪问题

问题1：关于gdb

不显示源码的问题：

(gdb) l 133 in ./build/./src/Databases/DatabaseReplicatedWorker.cpp (gdb) 133 in ./build/./src/Databases/DatabaseReplicatedWorker.cpp (gdb) 133 in ./build/./src/Databases/DatabaseReplicatedWorker.cpp (gdb) 133 in ./build/./src/Databases/DatabaseReplicatedWorker.cpp

info source查看：

(gdb) set substitute-path ./build/./src /home/yanglw/work/ClickHouse/src

(gdb) info source

Current source file is ./build/./src/Databases/DatabaseReplicatedWorker.cpp

Compilation directory is ./build

Contains 533 lines.

Source language is c++.

Producer is Ubuntu clang version 19.1.7 (++20250114103320+cd708029e0b2-1~exp1~20250114103432.75).

Compiled with DWARF 5 debugging format.

Does not include preprocessor macro info.

解决：/home/user/clickhouse/src为调试的Clickhouse的源码路径

(gdb) set substitute-path ./build/./src /home/user/clickhouse/src

问题2：关于clickhouse

CREATE TABLE IF NOT EXISTS my_replicated_db.replicated_table

(

`event_date` Date,

`event_time` DateTime,

`user_id` UInt32,

`event_type` String,

`value` Float64

)

ENGINE = ReplicatedMergeTree('/clickhouse/databases/my_replicated_db/replicated_table', '{shard}', '{replica}')

PARTITION BY toYYYYMM(event_date)

ORDER BY (event_date, event_time, user_id)

SETTINGS index_granularity = 8192

Query id: 49e0846f-3642-4a1c-a0a5-9ee4faa469ce

Elapsed: 0.003 sec.

Received exception from server (version 25.6.1):

Code: 80. DB::Exception: Received from localhost:9000. DB::Exception: Explicit zookeeper_path and replica_name are specified in ReplicatedMergeTree arguments. If you really want to specify it explicitly, then you should use some macros to distinguish different shards and replicas. (INCORRECT_QUERY)

关键信息：

DB::Exception: Explicit zookeeper_path and replica_name are specified in ReplicatedMergeTree arguments.

可以将database_replicated_allow_replicated_engine_arguments设置为1，使session中能使用明确zookeeper路径创建复制表。

SET database_replicated_allow_replicated_engine_arguments = 1;