前言

上周出现了几次连接超时、连接池满还有dbc连接事务模板失败的问题。所以有必要深入了解下MySQL的连接过程。

正好，上周研究了怎么用Clion调试MySQL源码，接下来通过调试来研究一下吧。

服务端

启动

sql/main.cc

extern int mysqld_main(int argc, char **argv);

int main(int argc, char **argv) { return mysqld_main(argc, argv); }

• main：入口文件，仅调用了mysqld_main函数

sql/mysqld.cc

int mysqld_main(int argc, char **argv)
#endif
{
if (my_init())  // init my_sys library & pthreads
  {
    ...
  }
  ...
  if (load_defaults(MYSQL_CONFIG_NAME, load_default_groups, &argc, &argv,
                    &argv_alloc)) {
    ...
  }
  mysqld_socket_acceptor->connection_event_loop();

  mysqld_exit(signal_hand_thr_exit_code);
}

• mysql_main：MySQL服务端启动逻辑的主要处理函数
• my_init：系统库和线程初始化
• load_defaults：加载my.cnf各参数
• connection_event_loop：循环监听套接字。

sql/conn_handler/connection_acceptor.h

  /**
    Connection acceptor loop to accept connections from clients.
  */
  void connection_event_loop() {
    Connection_handler_manager *mgr =
        Connection_handler_manager::get_instance();
    while (!connection_events_loop_aborted()) {
      Channel_info *channel_info = m_listener->listen_for_connection_event();
      if (channel_info != NULL) mgr->process_new_connection(channel_info);
    }
  }

• connection_event_loop：通过socket_connection.cc::listen_for_connection_event循环监听，直到有新的连接，开始connection_handler_manager.cc::process_new_connection新连接的处理过程。

新连接

服务端一直处于监听状态，当有新连接请求时，调用process_new_connection处理新连接。

sql/conn_handler/connection_handler_manager.cc

void Connection_handler_manager::process_new_connection(
    Channel_info *channel_info) {
  if (connection_events_loop_aborted() ||
      !check_and_incr_conn_count(channel_info->is_admin_connection())) {
    channel_info->send_error_and_close_channel(ER_CON_COUNT_ERROR, 0, true);
    delete channel_info;
    return;
  }

  if (m_connection_handler->add_connection(channel_info)) {
    inc_aborted_connects();
    delete channel_info;
  }
}

• connection_events_loop_aborted：先判断是否已取消监听
• check_and_incr_conn_count：再判断（会加锁）是否现有连接数是否大于连接最大值（连接池满），未满，则将线程数加一，满了则拒绝连接。（注意，这里的判断逻辑使MySQL的实际最大连接数是max_connections + 1）
• add_connection：调用add_connection添加连接

sql/conn_handler/connection_handler_pre_thread.cc

bool Per_thread_connection_handler::add_connection(Channel_info *channel_info) {
  int error = 0;
  my_thread_handle id;

  DBUG_ENTER("Per_thread_connection_handler::add_connection");

  // Simulate thread creation for test case before we check thread cache
  DBUG_EXECUTE_IF("fail_thread_create", error = 1; goto handle_error;);

  if (!check_idle_thread_and_enqueue_connection(channel_info))
    DBUG_RETURN(false);

  /*
    There are no idle threads avaliable to take up the new
    connection. Create a new thread to handle the connection
  */
  channel_info->set_prior_thr_create_utime();
  error =
      mysql_thread_create(key_thread_one_connection, &id, &connection_attrib,
                          handle_connection, (void *)channel_info);
#ifndef DBUG_OFF
handle_error:
#endif  // !DBUG_OFF

  if (error) {
    ...
    //错误处理，略
  }

  Global_THD_manager::get_instance()->inc_thread_created();
  DBUG_PRINT("info", ("Thread created"));
  DBUG_RETURN(false);
}

• 调用check_idle_thread_and_enqueue_connection查看是否有空闲的线程，有则将本次连接信息加入等待队列，并给空闲线程发送唤醒信号；否则新建线程处理本次连接
• 在新线程中，调用handle_connection函数开始进行逻辑处理。

static void *handle_connection(void *arg) {
  Global_THD_manager *thd_manager = Global_THD_manager::get_instance();
  Connection_handler_manager *handler_manager =
      Connection_handler_manager::get_instance();
  Channel_info *channel_info = static_cast<Channel_info *>(arg);
  bool pthread_reused MY_ATTRIBUTE((unused)) = false;

  if (my_thread_init()) {
    ...
    //错误处理，略
  }

  for (;;) {
    THD *thd = init_new_thd(channel_info);
    if (thd == NULL) {
      ...
      //错误处理，略
    }

#ifdef HAVE_PSI_THREAD_INTERFACE
    if (pthread_reused) {
      ...
      //错误处理，略
    }
#endif

#ifdef HAVE_PSI_THREAD_INTERFACE
    /* Find the instrumented thread */
    PSI_thread *psi = PSI_THREAD_CALL(get_thread)();
    /* Save it within THD, so it can be inspected */
    thd->set_psi(psi);
#endif /* HAVE_PSI_THREAD_INTERFACE */
    mysql_thread_set_psi_id(thd->thread_id());
    mysql_thread_set_psi_THD(thd);
    mysql_socket_set_thread_owner(
        thd->get_protocol_classic()->get_vio()->mysql_socket);

    thd_manager->add_thd(thd);

    if (thd_prepare_connection(thd))
      handler_manager->inc_aborted_connects();
    else {
      while (thd_connection_alive(thd)) {
        if (do_command(thd)) break;
      }
      end_connection(thd);
    }
    close_connection(thd, 0, false, false);

    thd->get_stmt_da()->reset_diagnostics_area();
    thd->release_resources();

    // Clean up errors now, before possibly waiting for a new connection.
#ifndef HAVE_WOLFSSL
#if OPENSSL_VERSION_NUMBER < 0x10100000L
    ERR_remove_thread_state(0);
#endif /* OPENSSL_VERSION_NUMBER < 0x10100000L */
#endif
    thd_manager->remove_thd(thd);
    Connection_handler_manager::dec_connection_count();

#ifdef HAVE_PSI_THREAD_INTERFACE
    /*
      Delete the instrumentation for the job that just completed.
    */
    thd->set_psi(NULL);
    PSI_THREAD_CALL(delete_current_thread)();
#endif /* HAVE_PSI_THREAD_INTERFACE */

    delete thd;

    // Server is shutting down so end the pthread.
    if (connection_events_loop_aborted()) break;

    channel_info = Per_thread_connection_handler::block_until_new_connection();
    if (channel_info == NULL) break;
    pthread_reused = true;
    if (connection_events_loop_aborted()) {
      ...
      //错误处理，略
    }
  }

  my_thread_end();
  my_thread_exit(0);
  return NULL;
}

• 会对连接进行thd_prepare_connection预处理操作，没问题后继续下面的逻辑。
• 当连接未被关闭，就会一直do_command处理请求。
• 当连接关闭，则走下面关闭逻辑

执行

sql/sql_parse.cc

bool do_command(THD *thd) {
  bool return_value;
  int rc;
  NET *net = NULL;
  enum enum_server_command command;
  COM_DATA com_data;
  DBUG_ENTER("do_command");
  DBUG_ASSERT(thd->is_classic_protocol());

  /*
    indicator of uninitialized lex => normal flow of errors handling
    (see my_message_sql)
  */
  thd->lex->set_current_select(0);

  /*
    XXX: this code is here only to clear possible errors of init_connect.
    Consider moving to prepare_new_connection_state() instead.
    That requires making sure the DA is cleared before non-parsing statements
    such as COM_QUIT.
  */
  thd->clear_error();  // Clear error message
  thd->get_stmt_da()->reset_diagnostics_area();

  /*
    This thread will do a blocking read from the client which
    will be interrupted when the next command is received from
    the client, the connection is closed or "net_wait_timeout"
    number of seconds has passed.
  */
  net = thd->get_protocol_classic()->get_net();
  my_net_set_read_timeout(net, thd->variables.net_wait_timeout);
  net_new_transaction(net);

  /*
    Synchronization point for testing of KILL_CONNECTION.
    This sync point can wait here, to simulate slow code execution
    between the last test of thd->killed and blocking in read().

    The goal of this test is to verify that a connection does not
    hang, if it is killed at this point of execution.
    (Bug#37780 - main.kill fails randomly)

    Note that the sync point wait itself will be terminated by a
    kill. In this case it consumes a condition broadcast, but does
    not change anything else. The consumed broadcast should not
    matter here, because the read/recv() below doesn't use it.
  */
  DEBUG_SYNC(thd, "before_do_command_net_read");

  /*
    Because of networking layer callbacks in place,
    this call will maintain the following instrumentation:
    - IDLE events
    - SOCKET events
    - STATEMENT events
    - STAGE events
    when reading a new network packet.
    In particular, a new instrumented statement is started.
    See init_net_server_extension()
  */
  thd->m_server_idle = true;
  rc = thd->get_protocol()->get_command(&com_data, &command);
  thd->m_server_idle = false;

  if (rc) {
    ...
    //错误处理，略
  }

  char desc[VIO_DESCRIPTION_SIZE];
  vio_description(net->vio, desc);
  DBUG_PRINT("info", ("Command on %s = %d (%s)", desc, command,
                      command_name[command].str));

  DBUG_PRINT("info", ("packet: '%*.s'; command: %d",
                      thd->get_protocol_classic()->get_packet_length(),
                      thd->get_protocol_classic()->get_raw_packet(), command));
  if (thd->get_protocol_classic()->bad_packet)
    DBUG_ASSERT(0);  // Should be caught earlier

  // Reclaim some memory
  thd->get_protocol_classic()->get_output_packet()->shrink(
      thd->variables.net_buffer_length);
  /* Restore read timeout value */
  my_net_set_read_timeout(net, thd->variables.net_read_timeout);

  return_value = dispatch_command(thd, &com_data, command);
  thd->get_protocol_classic()->get_output_packet()->shrink(
      thd->variables.net_buffer_length);

out:
  /* The statement instrumentation must be closed in all cases. */
  DBUG_ASSERT(thd->m_digest == NULL);
  DBUG_ASSERT(thd->m_statement_psi == NULL);
  DBUG_RETURN(return_value);
}

• 主要的处理逻辑为dispatch_command，根据不同的command类型进行分发。

/**
  Perform one connection-level (COM_XXXX) command.

  @param thd             connection handle
  @param command         type of command to perform
  @param com_data        com_data union to store the generated command

  @todo
    set thd->lex->sql_command to SQLCOM_END here.
  @todo
    The following has to be changed to an 8 byte integer

  @retval
    0   ok
  @retval
    1   request of thread shutdown, i. e. if command is
        COM_QUIT
*/
bool dispatch_command(THD *thd, const COM_DATA *com_data,
                      enum enum_server_command command) {
  ... //太长不看
  
  switch (command) {
    case ... //太长不看
    case COM_QUERY: {
      ... //太长不看
      mysql_parse(thd, &parser_state);

      ... //太长不看

      DBUG_PRINT("info", ("query ready"));
      break;
    }
    case ... //太长不看
    default:
      my_error(ER_UNKNOWN_COM_ERROR, MYF(0));
      break;
  }
}

• 主要看COM_QUERY这个逻辑，我们要用到的DDL、DML都会走这个流程，这个流程中主要是调用mysql_parse方法

/**
  Parse a query.

  @param thd          Current session.
  @param parser_state Parser state.
*/

void mysql_parse(THD *thd, Parser_state *parser_state) {
  ... //太长不看

  mysql_reset_thd_for_next_command(thd);
  
  if (!err) {
    err = parse_sql(thd, parser_state, NULL);
    ... //太长不看
  }

  if (!err) {
    
    mysql_rewrite_query(thd);

    ... //太长不看
  }

  if (!err) {
    ...
    error = mysql_execute_command(thd, true);
    ...
  }
}

• 主要是SQL语法解析和执行
• mysql_reset_thd_for_next_command是对下一次执行做准备，重置线程各变量
• mysql_rewrite_query看着像是SQL优化？待定 还没追进去，记个TODO
• 词法解析前不应该有缓存吗？没有找到缓存的逻辑，记个TODO（后续：原来MySQL8.0取消了query cache，详见：https://mysqlserverteam.com/mysql-8-0-retiring-support-for-the-query-cache/）

关闭连接

sql/conn_handler/connection_handler_pre_thread.cc

Channel_info *Per_thread_connection_handler::block_until_new_connection() {
  Channel_info *new_conn = NULL;
  mysql_mutex_lock(&LOCK_thread_cache);
  if (blocked_pthread_count < max_blocked_pthreads && !shrink_cache) {
    /* Don't kill the pthread, just block it for reuse */
    DBUG_PRINT("info", ("Blocking pthread for reuse"));

    /*
      mysys_var is bound to the physical thread,
      so make sure mysys_var->dbug is reset to a clean state
      before picking another session in the thread cache.
    */
    DBUG_POP();
    DBUG_ASSERT(!_db_is_pushed_());

    // Block pthread
    blocked_pthread_count++;
    while (!connection_events_loop_aborted() && !wake_pthread && !shrink_cache)
      mysql_cond_wait(&COND_thread_cache, &LOCK_thread_cache);
    blocked_pthread_count--;

    if (shrink_cache && blocked_pthread_count <= max_blocked_pthreads) {
      mysql_cond_signal(&COND_flush_thread_cache);
    }

    if (wake_pthread) {
      wake_pthread--;
      if (!waiting_channel_info_list->empty()) {
        new_conn = waiting_channel_info_list->front();
        waiting_channel_info_list->pop_front();
        DBUG_PRINT("info", ("waiting_channel_info_list->pop %p", new_conn));
      } else {
        DBUG_ASSERT(0);  // We should not get here.
      }
    }
  }
  mysql_mutex_unlock(&LOCK_thread_cache);
  return new_conn;
}

• 如果阻塞的线程数小于最大阻塞线程数，则此线程不回收，而是进入阻塞状态（等待），等待新连接来的时候重复使用。
• 否则关闭线程。

客户端

【从入门到放弃-MySQL】数据库连接过程分析-客户端

参考文献：
https://www.cnblogs.com/FateTHarlaown/p/8676166.html

更多文章见：https://nc2era.com