ClickHouse源码阅读笔记（一）之主要流程
入口main函数在dbms/programs/main.cpp
int main(int argc_, char ** argv_)
{
...
/// Print a basic help if nothing was matched
MainFunc main_func = printHelp;//这里根据启动时传入的参数来确定后面执行哪个func，对于server来说，对应的函数为mainEntryClickHouseServer
for (auto & application : clickhouse_applications)
{
if (isClickhouseApp(application.first, argv))
{
main_func = application.second;
break;
}
}
return main_func(static_cast<int>(argv.size()), argv.data());//对于server，这里调用mainEntryClickHouseServer后，转到dbms/programs/server/server.cpp
}
在dbms/programs/server/server.cpp中，提供三类接口，按照源码的描述，说明如下：
/** Server provides three interfaces:
* 1. HTTP - simple interface for any applications.
适用于任何应用程序的HTTP接口。
* 2. TCP - interface for native clickhouse-client and for server to server internal communications.
用于本地client和server之间通信的TCP接口。
* More rich and efficient, but less compatible
丰富高效，但兼容性不好
* - data is transferred by columns;
数据按列传输
* - data is transferred compressed;
数据压缩后传输
* Allows to get more information in response.
允许在响应消息中获取更多信息
* 3. Interserver HTTP - for replication.
用于复制的内部HTTP。
*/
dbms/programs/server/server.cpp中的main函数会解析参数配置，初始化server，启动服务监听端口。
int mainEntryClickHouseServer(int argc, char ** argv)
{
DB::Server app;
try
{
return app.run(argc, argv);//这里调用run。
}
catch (...)
{
std::cerr << DB::getCurrentExceptionMessage(true) << "\n";
auto code = DB::getCurrentExceptionCode();
return code ? code : 1;
}
}
clickhouse使用poco这个网络库来处理网络请求，每个client连接的处理逻辑在dbms/programs/server//TCPHandler.cpp的run()方法中。
void TCPHandler::run()
{
try
{
runImpl();//这里调用 runImpl函数。
LOG_INFO(log, "Done processing connection.");
}
catch (Poco::Exception & e)
{
/// Timeout - not an error.
if (!strcmp(e.what(), "Timeout"))
{
LOG_DEBUG(log, "Poco::Exception. Code: " << ErrorCodes::POCO_EXCEPTION << ", e.code() = " << e.code()
<< ", e.displayText() = " << e.displayText() << ", e.what() = " << e.what());
}
else
throw;
}
}
在TCPHandler::runImpl()函数中，除去握手，初始化上下文，异常处理等代码，主要逻辑如下：
void TCPHandler::runImpl()
{
receivePacket();//接收请求
executeQuery(state.query, *query_context, false, state.stage, may_have_embedded_data);//处理请求
/// Does the request require receive data from client?
if (state.need_receive_data_for_insert)
processInsertQuery(connection_settings);//负责将结果返回给客户端
else if (state.need_receive_data_for_input)
{
/// It is special case for input(), all works for reading data from client will be done in callbacks.
/// state.io.in is NullAndDoCopyBlockInputStream so read it once.
state.io.in->read();
state.io.onFinish();
}
else if (state.io.pipeline.initialized())
processOrdinaryQueryWithProcessors(query_context->getSettingsRef().max_threads);//负责将结果返回给客户端
else
processOrdinaryQuery();//负责将结果返回给客户端
}
接下来，我们继续看executeQuery处理请求的逻辑，在dbms/src/Interpreters/executeQuery.cpp中，主要逻辑如下：
BlockIO executeQuery(
const String & query,
Context & context,
bool internal,
QueryProcessingStage::Enum stage,
bool may_have_embedded_data,
bool allow_processors)
{
std::tie(ast, streams) = executeQueryImpl(query.data(), query.data() + query.size(), context,
internal, stage, !may_have_embedded_data, nullptr, allow_processors);//这里调用executeQueryImpl
}
接下来再看executeQueryImpl的主要处理逻辑：
static std::tuple<ASTPtr, BlockIO> executeQueryImpl(
const char * begin,
const char * end,
Context & context,
bool internal,
QueryProcessingStage::Enum stage,
bool has_query_tail,
ReadBuffer * istr,
bool allow_processors)
{
ast = parseQuery(parser, begin, end, "", max_query_size, settings.max_parser_depth);//解析查询语句
if (use_processors)//使用pipeline
pipeline = interpreter->executeWithProcessors();
else//不使用pipiline
res = interpreter->execute();//根据interpreter的类型来调用对应类型的execute函数执行
}
下一篇文章会介绍interpreter，未完待续。。。