一、基本框架
整体的框架其实和普通的C/S服务器模式的应用程序没有本质的不同，只是MySql更显复杂一些。服务端启动网络服务进行网络监听，用户通过客户端复用网络通信来和服务端进行通信。在服务端要有一些基础的方法解释，这有些类似于编译器对源码的处理，但此处的SQL要简单很多。主要的功能框架如下图：
这幅图和上篇文章的图其实是一样的，只是这里进行了一些细节的表述，更容易理解。
二、主要流程
主要的流程基本如下图：
流程其实就是整个程序的最基本的数据和业务流动的方向，掌握了流程图基本就是明白了骨架，余下的就是不断的填充细节到上面，就可以全方位的掌握MySql的源码了。本文是基于Boost版本的Mysql8.0源码分析。
三、源码
一个程序的启动，当然是从Main函数开始的，特别是C/c++程序。在MySql中也不例外，下面先看main.cc中的Main函数：
//就下面几行代码
extern int mysqld_main(int argc, char **argv);
int main(int argc, char **argv)
{
return mysqld_main(argc, argv);
}
//mysqld.cc主要的动作都在这个文件中完成
int mysqld_main(int argc, char **argv) {
bool mysqld_monitor = false;
mysqld_early_option = is_early_option(argc, argv);
if (!mysqld_early_option) {
initialize_mysqld_monitor();
mysqld_monitor = is_mysqld_monitor();
}
if (mysqld_early_option || !mysqld_monitor) {
/*
When several instances are running on the same machine, we
need to have an unique named hEventShudown through the
application PID e.g.: MySQLShutdown1890; MySQLShutdown2342
*/
longlong10_to_str(GetCurrentProcessId(),
my_stpcpy(shutdown_event_name, "MYSQLShutdown"), 10);
longlong10_to_str(GetCurrentProcessId(),
my_stpcpy(restart_event_name, "MYSQLRestart"), 10);
}
/* Must be initialized early for comparison of service name */
system_charset_info = &my_charset_utf8_general_ci;
if (mysqld_early_option || !mysqld_monitor) {
#ifdef WITH_PERFSCHEMA_STORAGE_ENGINE
pre_initialize_performance_schema();
#endif /*WITH_PERFSCHEMA_STORAGE_ENGINE */
if (my_init()) {
LogErr(ERROR_LEVEL, ER_MYINIT_FAILED);
flush_error_log_messages();
return 1;
}
}
if (Service.GetOS() && mysqld_monitor) /* true NT family */
{
char file_path[FN_REFLEN];
my_path(file_path, argv[0], ""); /* Find name in path */
fn_format(file_path, argv[0], file_path, "",
MY_REPLACE_DIR | MY_UNPACK_FILENAME | MY_RESOLVE_SYMLINKS);
if (argc == 2) {
if (!default_service_handling(argv, MYSQL_SERVICENAME, MYSQL_SERVICENAME,
file_path, "", NULL))
return 0;
if (Service.IsService(argv[1])) /* Start an optional service */
{
/*
Only add the service name to the groups read from the config file
if it's not "MySQL". (The default service name should be 'mysqld'
but we started a bad tradition by calling it MySQL from the start
and we are now stuck with it.
*/
if (my_strcasecmp(system_charset_info, argv[1], "mysql"))
load_default_groups[load_default_groups_sz - 2] = argv[1];
windows_service = true;
Service.Init(argv[1], mysql_service);
return 0;
}
} else if (argc == 3) /* install or remove any optional service */
{
if (!default_service_handling(argv, argv[2], argv[2], file_path, "",
NULL))
return 0;
if (Service.IsService(argv[2])) {
/*
mysqld was started as
mysqld --defaults-file=my_path\my.ini service-name
*/
use_opt_args = 1;
opt_argc = 2; // Skip service-name
opt_argv = argv;
windows_service = true;
if (my_strcasecmp(system_charset_info, argv[2], "mysql"))
load_default_groups[load_default_groups_sz - 2] = argv[2];
Service.Init(argv[2], mysql_service);
return 0;
}
} else if (argc == 4 || argc == 5) {
/*
This may seem strange, because we handle --local-service while
preserving 4.1's behavior of allowing any one other argument that is
passed to the service on startup. (The assumption is that this is
--defaults-file=file, but that was not enforced in 4.1, so we don't
enforce it here.)
*/
const char *extra_opt = NullS;
const char *account_name = NullS;
int index;
for (index = 3; index < argc; index++) {
if (!strcmp(argv[index], "--local-service"))
account_name = "NT AUTHORITY\\LocalService";
else
extra_opt = argv[index];
}
if (argc == 4 || account_name)
if (!default_service_handling(argv, argv[2], argv[2], file_path,
extra_opt, account_name))
return 0;
} else if (argc == 1 && Service.IsService(MYSQL_SERVICENAME)) {
/* start the default service */
windows_service = true;
Service.Init(MYSQL_SERVICENAME, mysql_service);
return 0;
}
}
// Set windows_service value in mysqld
if (!mysqld_monitor) {
windows_service = is_monitor_win_service();
if (windows_service) {
if (argc == 2 && Service.IsService(argv[1])) {
if (my_strcasecmp(system_charset_info, argv[1], "mysql"))
load_default_groups[load_default_groups_sz - 2] = argv[1];
argc--;
} else if (argc == 3 && Service.IsService(argv[2])) {
/*
mysqld was started as
mysqld --defaults-file=my_path\my.ini service-name
*/
if (my_strcasecmp(system_charset_info, argv[2], "mysql"))
load_default_groups[load_default_groups_sz - 2] = argv[2];
argc--;
}
}
my_global_argc = argc;
my_global_argv = argv;
} else {
Service.my_argc = argc;
Service.my_argv = argv;
}
return mysql_service(NULL);
}
int mysql_service(void *p) {
int my_argc;
char **my_argv;
if (use_opt_args) {
my_argc = opt_argc;
my_argv = opt_argv;
} else if (is_mysqld_monitor()) {
my_argc = Service.my_argc;
my_argv = Service.my_argv;
} else {
my_argc = my_global_argc;
my_argv = my_global_argv;
}
if (!mysqld_early_option) {
int res = start_monitor();
if (res != -1) {
deinitialize_mysqld_monitor();
return res;
}
}
if (my_thread_init()) {
flush_error_log_messages();
return 1;
}
win_main(my_argc, my_argv);
my_thread_end();
return 0;
}
最后调用win_main,这才是真正干活的，跟下去：
#ifdef _WIN32
int win_main(int argc, char **argv)
#else
int mysqld_main(int argc, char **argv)
#endif
{
// Substitute the full path to the executable in argv[0]
substitute_progpath(argv);
sysd::notify_connect();
sysd::notify("STATUS=Server startup in progress\n");
/*
Perform basic thread library and malloc initialization,
to be able to read defaults files and parse options.
*/
my_progname = argv[0];
calculate_mysql_home_from_my_progname();
#ifndef _WIN32
#ifdef WITH_PERFSCHEMA_STORAGE_ENGINE
pre_initialize_performance_schema();
#endif /*WITH_PERFSCHEMA_STORAGE_ENGINE */
// For windows, my_init() is called from the win specific mysqld_main
if (my_init()) // init my_sys library & pthreads
{
LogErr(ERROR_LEVEL, ER_MYINIT_FAILED);
flush_error_log_messages();
return 1;
}
#endif /* _WIN32 */
orig_argc = argc;
orig_argv = argv;
my_getopt_use_args_separator = true;
my_defaults_read_login_file = false;
//加载my.cnf,my.cnf.d及相关命令行参数
if (load_defaults(MYSQL_CONFIG_NAME, load_default_groups, &argc, &argv,
&argv_alloc)) {
flush_error_log_messages();
return 1;
}
/* Set data dir directory paths */
strmake(mysql_real_data_home, get_relative_path(MYSQL_DATADIR),
sizeof(mysql_real_data_home) - 1);
/*
Initialize variables cache for persisted variables, load persisted
config file and append read only persisted variables to command line
options if present.
*/
if (persisted_variables_cache.init(&argc, &argv) ||
persisted_variables_cache.load_persist_file() ||
persisted_variables_cache.append_read_only_variables(&argc, &argv)) {
flush_error_log_messages();
return 1;
}
my_getopt_use_args_separator = false;
remaining_argc = argc;
remaining_argv = argv;
init_variable_default_paths();
/* Must be initialized early for comparison of options name */
system_charset_info = &my_charset_utf8_general_ci;
/* Write mysys error messages to the error log. */
local_message_hook = error_log_print;
int heo_error;
#ifdef WITH_PERFSCHEMA_STORAGE_ENGINE
/*
Initialize the array of performance schema instrument configurations.
*/
init_pfs_instrument_array();
#endif /* WITH_PERFSCHEMA_STORAGE_ENGINE */
heo_error = handle_early_options();
init_sql_statement_names();
sys_var_init();
ulong requested_open_files = 0;
// Init error log subsystem. This does not actually open the log yet.
if (init_error_log()) unireg_abort(MYSQLD_ABORT_EXIT);
if (!opt_validate_config) adjust_related_options(&requested_open_files);
#ifdef WITH_PERFSCHEMA_STORAGE_ENGINE
if (heo_error == 0) {
if (!is_help_or_validate_option() && !opt_initialize) {
int pfs_rc;
/* Add sizing hints from the server sizing parameters. */
pfs_param.m_hints.m_table_definition_cache = table_def_size;
pfs_param.m_hints.m_table_open_cache = table_cache_size;
pfs_param.m_hints.m_max_connections = max_connections;
pfs_param.m_hints.m_open_files_limit = requested_open_files;
pfs_param.m_hints.m_max_prepared_stmt_count = max_prepared_stmt_count;
pfs_rc = initialize_performance_schema(
&pfs_param, &psi_thread_hook, &psi_mutex_hook, &psi_rwlock_hook,
&psi_cond_hook, &psi_file_hook, &psi_socket_hook, &psi_table_hook,
&psi_mdl_hook, &psi_idle_hook, &psi_stage_hook, &psi_statement_hook,
&psi_transaction_hook, &psi_memory_hook, &psi_error_hook,
&psi_data_lock_hook, &psi_system_hook, &psi_tls_channel_hook);
if ((pfs_rc != 0) && pfs_param.m_enabled) {
pfs_param.m_enabled = false;
LogErr(WARNING_LEVEL, ER_PERFSCHEMA_INIT_FAILED);
}
}
}
#endif /* WITH_PERFSCHEMA_STORAGE_ENGINE */
#ifdef WITH_LOCK_ORDER
if (heo_error == 0) {
if (lo_param.m_enabled && !opt_help && !opt_initialize) {
int lo_rc;
lo_rc = LO_init(&lo_param, &psi_thread_hook, &psi_mutex_hook,
&psi_rwlock_hook, &psi_cond_hook, &psi_file_hook,
&psi_socket_hook, &psi_table_hook, &psi_mdl_hook,
&psi_idle_hook, &psi_stage_hook, &psi_statement_hook,
&psi_transaction_hook, &psi_memory_hook);
if (lo_rc != 0) {
LogErr(WARNING_LEVEL, ER_LOCK_ORDER_INIT_FAILED);
}
}
}
#endif /* WITH_LOCK_ORDER */
/*
Other provider of the instrumentation interface should
initialize PSI_hook here:
- HAVE_PSI_INTERFACE is for the instrumentation interface
- WITH_PERFSCHEMA_STORAGE_ENGINE is for one implementation
of the interface,
but there could be alternate implementations, which is why
these two defines are kept separate.
*/
#ifdef HAVE_PSI_INTERFACE
/*
Obtain the current performance schema instrumentation interface,
if available.
*/
void *service;
if (psi_thread_hook != nullptr) {
service = psi_thread_hook->get_interface(PSI_CURRENT_THREAD_VERSION);
if (service != nullptr) {
set_psi_thread_service(service);
}
}
if (psi_mutex_hook != nullptr) {
service = psi_mutex_hook->get_interface(PSI_CURRENT_MUTEX_VERSION);
if (service != nullptr) {
set_psi_mutex_service(service);
}
}
if (psi_rwlock_hook != nullptr) {
service = psi_rwlock_hook->get_interface(PSI_CURRENT_RWLOCK_VERSION);
if (service != nullptr) {
set_psi_rwlock_service(service);
}
}
if (psi_cond_hook != nullptr) {
service = psi_cond_hook->get_interface(PSI_CURRENT_COND_VERSION);
if (service != nullptr) {
set_psi_cond_service(service);
}
}
if (psi_file_hook != nullptr) {
service = psi_file_hook->get_interface(PSI_CURRENT_FILE_VERSION);
if (service != nullptr) {
set_psi_file_service(service);
}
}
if (psi_socket_hook != nullptr) {
service = psi_socket_hook->get_interface(PSI_CURRENT_SOCKET_VERSION);
if (service != nullptr) {
set_psi_socket_service(service);
}
}
if (psi_table_hook != nullptr) {
service = psi_table_hook->get_interface(PSI_CURRENT_TABLE_VERSION);
if (service != nullptr) {
set_psi_table_service(service);
}
}
if (psi_mdl_hook != nullptr) {
service = psi_mdl_hook->get_interface(PSI_CURRENT_MDL_VERSION);
if (service != nullptr) {
set_psi_mdl_service(service);
}
}
if (psi_idle_hook != nullptr) {
service = psi_idle_hook->get_interface(PSI_CURRENT_IDLE_VERSION);
if (service != nullptr) {
set_psi_idle_service(service);
}
}
if (psi_stage_hook != nullptr) {
service = psi_stage_hook->get_interface(PSI_CURRENT_STAGE_VERSION);
if (service != nullptr) {
set_psi_stage_service(service);
}
}
if (psi_statement_hook != nullptr) {
service = psi_statement_hook->get_interface(PSI_CURRENT_STATEMENT_VERSION);
if (service != nullptr) {
set_psi_statement_service(service);
}
}
if (psi_transaction_hook != nullptr) {
service =
psi_transaction_hook->get_interface(PSI_CURRENT_TRANSACTION_VERSION);
if (service != nullptr) {
set_psi_transaction_service(service);
}
}
if (psi_memory_hook != nullptr) {
service = psi_memory_hook->get_interface(PSI_CURRENT_MEMORY_VERSION);
if (service != nullptr) {
set_psi_memory_service(service);
}
}
if (psi_error_hook != nullptr) {
service = psi_error_hook->get_interface(PSI_CURRENT_ERROR_VERSION);
if (service != nullptr) {
set_psi_error_service(service);
}
}
if (psi_data_lock_hook != nullptr) {
service = psi_data_lock_hook->get_interface(PSI_CURRENT_DATA_LOCK_VERSION);
if (service != nullptr) {
set_psi_data_lock_service(service);
}
}
if (psi_system_hook != nullptr) {
service = psi_system_hook->get_interface(PSI_CURRENT_SYSTEM_VERSION);
if (service != nullptr) {
set_psi_system_service(service);
}
}
if (psi_tls_channel_hook != nullptr) {
service =
psi_tls_channel_hook->get_interface(PSI_CURRENT_TLS_CHANNEL_VERSION);
if (service != nullptr) {
set_psi_tls_channel_service(service);
}
}
/*
Now that we have parsed the command line arguments, and have initialized
the performance schema itself, the next step is to register all the
server instruments.
*/
init_server_psi_keys();
/*
Now that some instrumentation is in place,
recreate objects which were initialised early,
so that they are instrumented as well.
*/
my_thread_global_reinit();
#endif /* HAVE_PSI_INTERFACE */
/*
Initialize Components core subsystem early on, once we have PSI, which it
uses. This part doesn't use any more MySQL-specific functionalities but
error logging and PFS.
*/
if (component_infrastructure_init()) unireg_abort(MYSQLD_ABORT_EXIT);
/*
Initialize Performance Schema component services.
*/
#ifdef HAVE_PSI_THREAD_INTERFACE
if (!is_help_or_validate_option() && !opt_initialize) {
register_pfs_notification_service();
register_pfs_resource_group_service();
}
#endif
// Initialize the resource group subsystem.
auto res_grp_mgr = resourcegroups::Resource_group_mgr::instance();
if (!is_help_or_validate_option() && !opt_initialize) {
if (res_grp_mgr->init()) {
LogErr(ERROR_LEVEL, ER_RESOURCE_GROUP_SUBSYSTEM_INIT_FAILED);
unireg_abort(MYSQLD_ABORT_EXIT);
}
}
#ifdef HAVE_PSI_THREAD_INTERFACE
/* Instrument the main thread */
PSI_thread *psi = PSI_THREAD_CALL(new_thread)(key_thread_main, nullptr, 0);
PSI_THREAD_CALL(set_thread_os_id)(psi);
PSI_THREAD_CALL(set_thread)(psi);
#endif /* HAVE_PSI_THREAD_INTERFACE */
/* Initialize audit interface globals. Audit plugins are inited later. */
mysql_audit_initialize();
Srv_session::module_init();
/*
Perform basic query log initialization. Should be called after
MY_INIT, as it initializes mutexes.
*/
query_logger.init();
if (heo_error) {
/*
Parsing command line option failed,
Since we don't have a workable remaining_argc/remaining_argv
to continue the server initialization, this is as far as this
code can go.
This is the best effort to log meaningful messages:
- messages will be printed to stderr, which is not redirected yet,
- messages will be printed in the NT event log, for windows.
*/
flush_error_log_messages();
/*
Not enough initializations for unireg_abort()
Using exit() for windows.
*/
exit(MYSQLD_ABORT_EXIT);
}
//读取和设置相关配置文件及命令参数，并根据其具体的设置开始操作，内容很丰富，在后面会分析
if (init_common_variables()) {
setup_error_log();
unireg_abort(MYSQLD_ABORT_EXIT); // Will do exit
}
keyring_lockable_init();
my_init_signals();
size_t guardize = 0;
#ifndef _WIN32
int retval = pthread_attr_getguardsize(&connection_attrib, &guardize);
assert(retval == 0);
if (retval != 0) guardize = my_thread_stack_size;
#endif
#if defined(__ia64__) || defined(__ia64)
/*
Peculiar things with ia64 platforms - it seems we only have half the
stack size in reality, so we have to double it here
*/
guardize = my_thread_stack_size;
#endif
if (0 != my_thread_attr_setstacksize(&connection_attrib,
my_thread_stack_size + guardize)) {
assert(false);
}
{
/* Retrieve used stack size; Needed for checking stack overflows */
size_t stack_size = 0;
my_thread_attr_getstacksize(&connection_attrib, &stack_size);
/* We must check if stack_size = 0 as Solaris 2.9 can return 0 here */
if (stack_size && stack_size < (my_thread_stack_size + guardize)) {
LogErr(WARNING_LEVEL, ER_STACKSIZE_UNEXPECTED,
my_thread_stack_size + guardize, (long)stack_size);
#if defined(__ia64__) || defined(__ia64)
my_thread_stack_size = stack_size / 2;
#else
my_thread_stack_size = static_cast<ulong>(stack_size - guardize);
#endif
}
}
#ifndef NDEBUG
test_lc_time_sz();
srand(static_cast<uint>(time(nullptr)));
#endif
#if !defined(_WIN32)
if (opt_initialize && opt_daemonize) {
fprintf(stderr, "Initialize and daemon options are incompatible.\n");
unireg_abort(MYSQLD_ABORT_EXIT);
}
if (opt_daemonize && log_error_dest == disabled_my_option &&
(isatty(STDOUT_FILENO) || isatty(STDERR_FILENO))) {
// Just use the default in this case.
log_error_dest = "";
}
if (opt_daemonize && !opt_validate_config) {
if (chdir("/") < 0) {
LogErr(ERROR_LEVEL, ER_CANNOT_CHANGE_TO_ROOT_DIR, strerror(errno));
unireg_abort(MYSQLD_ABORT_EXIT);
}
if ((pipe_write_fd = mysqld::runtime::mysqld_daemonize()) < -1) {
LogErr(ERROR_LEVEL, ER_FAILED_START_MYSQLD_DAEMON);
unireg_abort(MYSQLD_ABORT_EXIT);
}
if (pipe_write_fd < 0) {
// This is the launching process and the daemon appears to have
// started ok (Need to call unireg_abort with success here to
// clean up resources in the lauching process.
unireg_abort(MYSQLD_SUCCESS_EXIT);
}
// Need to update the value of current_pid so that it reflects the
// pid of the daemon (the previous value was set by unireg_init()
// while still in the launcher process.
current_pid = static_cast<ulong>(getpid());
}
#endif
#ifndef _WIN32
user_info = check_user(mysqld_user);
if (!user_info.IsVoid()) {
#if HAVE_CHOWN
if (unlikely(opt_initialize)) {
/* need to change the owner of the freshly created data directory */
MY_STAT stat;
char errbuf[MYSYS_STRERROR_SIZE];
bool must_chown = true;
/* fetch the directory's owner */
if (!my_stat(mysql_real_data_home, &stat, MYF(0))) {
LogErr(INFORMATION_LEVEL, ER_CANT_STAT_DATADIR, my_errno(),
my_strerror(errbuf, sizeof(errbuf), my_errno()));
}
/* Don't change it if it's already the same as SElinux stops this */
else if (stat.st_uid == user_info.pw_uid &&
stat.st_gid == user_info.pw_gid)
must_chown = false;
if (must_chown &&
chown(mysql_real_data_home, user_info.pw_uid, user_info.pw_gid)) {
LogErr(ERROR_LEVEL, ER_CANT_CHOWN_DATADIR, mysqld_user);
unireg_abort(1);
}
}
#endif
#if defined(HAVE_MLOCKALL) && defined(MCL_CURRENT)
if (locked_in_memory) // getuid() == 0 here
set_effective_user(user_info);
else
#endif
set_user(mysqld_user, user_info);
}
#endif // !_WIN32
/*
initiate key migration if any one of the migration specific
options are provided.
*/
if (opt_keyring_migration_source || opt_keyring_migration_destination ||
migrate_connect_options) {
int exit_state = MYSQLD_ABORT_EXIT;
while (true) {
Migrate_keyring mk;
my_getopt_skip_unknown = true;
if (mk.init(remaining_argc, remaining_argv, opt_keyring_migration_source,
opt_keyring_migration_destination, opt_keyring_migration_user,
opt_keyring_migration_host, opt_keyring_migration_password,
opt_keyring_migration_socket, opt_keyring_migration_port,
opt_keyring_migration_to_component)) {
LogErr(ERROR_LEVEL, ER_KEYRING_MIGRATION_FAILED);
log_error_dest = "stderr";
flush_error_log_messages();
break;
}
if (mk.execute()) {
LogErr(ERROR_LEVEL, ER_KEYRING_MIGRATION_FAILED);
log_error_dest = "stderr";
flush_error_log_messages();
break;
}
my_getopt_skip_unknown = false;
LogErr(INFORMATION_LEVEL, ER_KEYRING_MIGRATION_SUCCESSFUL);
log_error_dest = "stderr";
flush_error_log_messages();
exit_state = MYSQLD_SUCCESS_EXIT;
break;
}
unireg_abort(exit_state);
}
/*
We have enough space for fiddling with the argv, continue
*/
if (!(is_help_or_validate_option()) &&
my_setwd(mysql_real_data_home, MYF(0))) {
char errbuf[MYSYS_STRERROR_SIZE];
LogErr(ERROR_LEVEL, ER_CANT_SET_DATA_DIR, mysql_real_data_home, errno,
my_strerror(errbuf, sizeof(errbuf), errno));
unireg_abort(MYSQLD_ABORT_EXIT); /* purecov: inspected */
}
if (initialize_manifest_file_components()) unireg_abort(MYSQLD_ABORT_EXIT);
/*
If keyring component was loaded through manifest file, services provided
by such a component should get priority over keyring plugin. That's why
we have to set defaults before proxy keyring services are loaded.
*/
set_srv_keyring_implementation_as_default();
/*
The subsequent calls may take a long time : e.g. innodb log read.
Thus set the long running service control manager timeout
*/
#if defined(_WIN32)
if (windows_service) {
if (setup_service_status_cmd_processed_handle())
unireg_abort(MYSQLD_ABORT_EXIT);
char buf[32];
snprintf(buf, sizeof(buf), "T %lu", slow_start_timeout);
Service_status_msg msg(buf);
send_service_status(msg);
}
#endif
/* Determine default TCP port and unix socket name */
set_ports();
if (init_server_components()) unireg_abort(MYSQLD_ABORT_EXIT);
if (!server_id_supplied)
LogErr(INFORMATION_LEVEL, ER_WARN_NO_SERVERID_SPECIFIED);
/*
Add server_uuid to the sid_map. This must be done after
server_uuid has been initialized in init_server_auto_options and
after the binary log (and sid_map file) has been initialized in
init_server_components().
No error message is needed: init_sid_map() prints a message.
Strictly speaking, this is not currently needed when
opt_bin_log==0, since the variables that gtid_state->init
initializes are not currently used in that case. But we call it
regardless to avoid possible future bugs if gtid_state ever
needs to do anything else.
*/
global_sid_lock->wrlock();
int gtid_ret = gtid_state->init();
global_sid_lock->unlock();
if (gtid_ret) unireg_abort(MYSQLD_ABORT_EXIT);
if (!opt_initialize && !opt_initialize_insecure) {
// Initialize executed_gtids from mysql.gtid_executed table.
if (gtid_state->read_gtid_executed_from_table() == -1) unireg_abort(1);
}
if (opt_bin_log) {
/*
Initialize GLOBAL.GTID_EXECUTED and GLOBAL.GTID_PURGED from
gtid_executed table and binlog files during server startup.
*/
Gtid_set *executed_gtids =
const_cast<Gtid_set *>(gtid_state->get_executed_gtids());
Gtid_set *lost_gtids = const_cast<Gtid_set *>(gtid_state->get_lost_gtids());
Gtid_set *gtids_only_in_table =
const_cast<Gtid_set *>(gtid_state->get_gtids_only_in_table());
Gtid_set *previous_gtids_logged =
const_cast<Gtid_set *>(gtid_state->get_previous_gtids_logged());
Gtid_set purged_gtids_from_binlog(global_sid_map, global_sid_lock);
Gtid_set gtids_in_binlog(global_sid_map, global_sid_lock);
Gtid_set gtids_in_binlog_not_in_table(global_sid_map, global_sid_lock);
if (mysql_bin_log.init_gtid_sets(
>ids_in_binlog, &purged_gtids_from_binlog,
opt_master_verify_checksum, true /*true=need lock*/,
nullptr /*trx_parser*/, nullptr /*partial_trx*/,
true /*is_server_starting*/))
unireg_abort(MYSQLD_ABORT_EXIT);
global_sid_lock->wrlock();
purged_gtids_from_binlog.dbug_print("purged_gtids_from_binlog");
gtids_in_binlog.dbug_print("gtids_in_binlog");
if (!gtids_in_binlog.is_empty() &&
!gtids_in_binlog.is_subset(executed_gtids)) {
gtids_in_binlog_not_in_table.add_gtid_set(>ids_in_binlog);
if (!executed_gtids->is_empty())
gtids_in_binlog_not_in_table.remove_gtid_set(executed_gtids);
/*
Save unsaved GTIDs into gtid_executed table, in the following
four cases:
1. the upgrade case.
2. the case that a slave is provisioned from a backup of
the master and the slave is cleaned by RESET MASTER
and RESET SLAVE before this.
3. the case that no binlog rotation happened from the
last RESET MASTER on the server before it crashes.
4. The set of GTIDs of the last binlog is not saved into the
gtid_executed table if server crashes, so we save it into
gtid_executed table and executed_gtids during recovery
from the crash.
*/
if (gtid_state->save(>ids_in_binlog_not_in_table) == -1) {
global_sid_lock->unlock();
unireg_abort(MYSQLD_ABORT_EXIT);
}
executed_gtids->add_gtid_set(>ids_in_binlog_not_in_table);
}
/* gtids_only_in_table= executed_gtids - gtids_in_binlog */
if (gtids_only_in_table->add_gtid_set(executed_gtids) != RETURN_STATUS_OK) {
global_sid_lock->unlock();
unireg_abort(MYSQLD_ABORT_EXIT);
}
gtids_only_in_table->remove_gtid_set(>ids_in_binlog);
/*
lost_gtids = executed_gtids -
(gtids_in_binlog - purged_gtids_from_binlog)
= gtids_only_in_table + purged_gtids_from_binlog;
*/
assert(lost_gtids->is_empty());
if (lost_gtids->add_gtid_set(gtids_only_in_table) != RETURN_STATUS_OK ||
lost_gtids->add_gtid_set(&purged_gtids_from_binlog) !=
RETURN_STATUS_OK) {
global_sid_lock->unlock();
unireg_abort(MYSQLD_ABORT_EXIT);
}
/* Prepare previous_gtids_logged for next binlog */
if (previous_gtids_logged->add_gtid_set(>ids_in_binlog) !=
RETURN_STATUS_OK) {
global_sid_lock->unlock();
unireg_abort(MYSQLD_ABORT_EXIT);
}
/*
Write the previous set of gtids at this point because during
the creation of the binary log this is not done as we cannot
move the init_gtid_sets() to a place before openning the binary
log. This requires some investigation.
/Alfranio
*/
Previous_gtids_log_event prev_gtids_ev(>ids_in_binlog);
global_sid_lock->unlock();
(prev_gtids_ev.common_footer)->checksum_alg =
static_cast<enum_binlog_checksum_alg>(binlog_checksum_options);
if (mysql_bin_log.write_event_to_binlog_and_sync(&prev_gtids_ev))
unireg_abort(MYSQLD_ABORT_EXIT);
if (expire_logs_days > 0 || binlog_expire_logs_seconds > 0) {
time_t purge_time = my_time(0) - binlog_expire_logs_seconds -
expire_logs_days * 24 * 60 * 60;
DBUG_EXECUTE_IF("expire_logs_always_at_start",
{ purge_time = my_time(0); });
mysql_bin_log.purge_logs_before_date(purge_time, true);
}
(void)RUN_HOOK(server_state, after_engine_recovery, (nullptr));
}
if (init_ssl_communication()) unireg_abort(MYSQLD_ABORT_EXIT);
//网络初始化，获取监听地址并启动监听
if (network_init()) unireg_abort(MYSQLD_ABORT_EXIT);
#ifdef _WIN32
if (opt_require_secure_transport && !opt_enable_shared_memory &&
!have_ssl() && !opt_initialize) {
LogErr(ERROR_LEVEL, ER_TRANSPORTS_WHAT_TRANSPORTS);
unireg_abort(MYSQLD_ABORT_EXIT);
}
#endif
/*
Initialize my_str_malloc(), my_str_realloc() and my_str_free()
*/
my_str_malloc = &my_str_malloc_mysqld;
my_str_free = &my_str_free_mysqld;
my_str_realloc = &my_str_realloc_mysqld;
error_handler_hook = my_message_sql;
bool abort = false;
/* Save pid of this process in a file */
if (!opt_initialize) {
if (create_pid_file()) abort = true;
}
/* Read the optimizer cost model configuration tables */
if (!opt_initialize) reload_optimizer_cost_constants();
if (
/*
Read components table to restore previously installed components. This
requires read access to mysql.component table.
*/
(!opt_initialize && mysql_component_infrastructure_init()) ||
mysql_rm_tmp_tables()) {
abort = true;
}
/* we do want to exit if there are any other unknown options */
if (remaining_argc > 1) {
int ho_error;
struct my_option no_opts[] = {{nullptr, 0, nullptr, nullptr, nullptr,
nullptr, GET_NO_ARG, NO_ARG, 0, 0, 0,
nullptr, 0, nullptr}};
/*
We need to eat any 'loose' arguments first before we conclude
that there are unprocessed options.
*/
my_getopt_skip_unknown = false;
if ((ho_error = handle_options(&remaining_argc, &remaining_argv, no_opts,
mysqld_get_one_option)))
abort = true;
else {
/* Add back the program name handle_options removes */
remaining_argc++;
remaining_argv--;
my_getopt_skip_unknown = true;
if (remaining_argc > 1) {
LogErr(ERROR_LEVEL, ER_EXCESS_ARGUMENTS, remaining_argv[1]);
LogErr(INFORMATION_LEVEL, ER_VERBOSE_HINT);
abort = true;
}
}
}
if (abort || acl_init(opt_noacl)) {
if (!abort) LogErr(ERROR_LEVEL, ER_PRIVILEGE_SYSTEM_INIT_FAILED);
abort = true;
opt_noacl = true;
}
/*
if running with --initialize, explicitly allocate the memory
to be used by ACL objects.
*/
if (opt_initialize) init_acl_memory();
if (abort || my_tz_init((THD *)nullptr, default_tz_name, opt_initialize) ||
grant_init(opt_noacl)) {
set_connection_events_loop_aborted(true);
delete_pid_file(MYF(MY_WME));
unireg_abort(MYSQLD_ABORT_EXIT);
}
/*
Bootstrap the dynamic privilege service implementation
*/
if (dynamic_privilege_init()) {
LogErr(WARNING_LEVEL, ER_PERSISTENT_PRIVILEGES_BOOTSTRAP);
}
if (!opt_initialize) servers_init(false);
if (!opt_noacl) {
udf_read_functions_table();
}
init_status_vars();
/* If running with --initialize, do not start replication. */
if (opt_initialize) opt_skip_slave_start = true;
check_binlog_cache_size(nullptr);
check_binlog_stmt_cache_size(nullptr);
binlog_unsafe_map_init();
/* If running with --initialize, do not start replication. */
if (!opt_initialize) {
// Make @@slave_skip_errors show the nice human-readable value.
set_slave_skip_errors(&opt_slave_skip_errors);
/*
Group replication filters should be discarded before init_slave(),
otherwise the pre-configured filters will be referenced by group
replication channels.
*/
rpl_channel_filters.discard_group_replication_filters();
/*
init_slave() must be called after the thread keys are created.
*/
if (server_id != 0)
init_slave(); /* Ignoring errors while configuring replication. */
/*
If the user specifies a per-channel replication filter through a
command-line option (or in a configuration file) for a slave
replication channel which does not exist as of now (i.e not
present in slave info tables yet), then the per-channel
replication filter is discarded with a warning.
If the user specifies a per-channel replication filter through
a command-line option (or in a configuration file) for group
replication channels 'group_replication_recovery' and
'group_replication_applier' which is disallowed, then the
per-channel replication filter is discarded with a warning.
*/
rpl_channel_filters.discard_all_unattached_filters();
}
#ifdef WITH_LOCK_ORDER
if (!opt_initialize) {
LO_activate();
}
#endif /* WITH_LOCK_ORDER */
#ifdef WITH_PERFSCHEMA_STORAGE_ENGINE
initialize_performance_schema_acl(opt_initialize);
#endif /* WITH_PERFSCHEMA_STORAGE_ENGINE */
initialize_information_schema_acl();
(void)RUN_HOOK(server_state, after_recovery, (nullptr));
if (Events::init(opt_noacl || opt_initialize))
unireg_abort(MYSQLD_ABORT_EXIT);
#ifndef _WIN32
// Start signal handler thread.
start_signal_handler();
#endif
/* set all persistent options */
if (persisted_variables_cache.set_persist_options()) {
LogErr(ERROR_LEVEL, ER_CANT_SET_UP_PERSISTED_VALUES);
flush_error_log_messages();
return 1;
}
/*
Try to read the previous run's error log and make it available in
performance_schema.error_log. Activate all error logging services
requested by the user in @@global.log_error_services (now that both
the component infrastructure and InnoDB are available), flush the
buffered error messages to performance schema and to configured services,
and end error log buffering.
*/
if (setup_error_log_components()) unireg_abort(MYSQLD_ABORT_EXIT);
/*
Invoke the bootstrap thread, if required.
*/
process_bootstrap();
/*
Event must be invoked after error_handler_hook is assigned to
my_message_sql, otherwise my_message will not cause the event to abort.
*/
void *argv_p = argv;
if (mysql_audit_notify(AUDIT_EVENT(MYSQL_AUDIT_SERVER_STARTUP_STARTUP),
static_cast<const char **>(argv_p), argc))
unireg_abort(MYSQLD_ABORT_EXIT);
#ifdef _WIN32
create_shutdown_and_restart_thread();
#endif
if (mysqld_process_must_end_at_startup) {
#if !defined(_WIN32)
if (opt_daemonize) mysqld::runtime::signal_parent(pipe_write_fd, 1);
#endif
unireg_abort(MYSQLD_SUCCESS_EXIT);
}
start_handle_manager();
create_compress_gtid_table_thread();
LogEvent()
.type(LOG_TYPE_ERROR)
.subsys(LOG_SUBSYSTEM_TAG)
.prio(SYSTEM_LEVEL)
.lookup(ER_SERVER_STARTUP_MSG, my_progname, server_version,
#ifdef HAVE_SYS_UN_H
(opt_initialize ? "" : mysqld_unix_port),
#else
"",
#endif
mysqld_port, MYSQL_COMPILATION_COMMENT_SERVER);
if (!opt_disable_networking && my_admin_bind_addr_str)
LogEvent()
.type(LOG_TYPE_ERROR)
.subsys(LOG_SUBSYSTEM_TAG)
.prio(SYSTEM_LEVEL)
.lookup(ER_SERVER_STARTUP_ADMIN_INTERFACE, my_admin_bind_addr_str,
mysqld_admin_port, MYSQL_COMPILATION_COMMENT);
#if defined(_WIN32)
if (windows_service) {
Service_status_msg s("R");
send_service_status(s);
}
#endif
server_components_initialized();
/*
Set opt_super_readonly here because if opt_super_readonly is set
in get_option, it will create problem while setting up event scheduler.
*/
set_super_read_only_post_init();
DBUG_PRINT("info", ("Block, listening for incoming connections"));
(void)MYSQL_SET_STAGE(0, __FILE__, __LINE__);
server_operational_state = SERVER_OPERATING;
sysd::notify("READY=1\nSTATUS=Server is operational\nMAIN_PID=", getpid(),
"\n");
(void)RUN_HOOK(server_state, before_handle_connection, (nullptr));
#if defined(_WIN32)
if (mysqld_socket_acceptor != nullptr)
mysqld_socket_acceptor->check_and_spawn_admin_connection_handler_thread();
setup_conn_event_handler_threads();
#else
mysql_mutex_lock(&LOCK_socket_listener_active);
// Make it possible for the signal handler to kill the listener.
socket_listener_active = true;
mysql_mutex_unlock(&LOCK_socket_listener_active);
if (opt_daemonize) {
if (nstdout != nullptr) {
// Show the pid on stdout if deamonizing and connected to tty
fprintf(nstdout, "mysqld is running as pid %lu\n", current_pid);
fclose(nstdout);
nstdout = nullptr;
}
mysqld::runtime::signal_parent(pipe_write_fd, 1);
}
mysqld_socket_acceptor->check_and_spawn_admin_connection_handler_thread();
mysqld_socket_acceptor->connection_event_loop();
#endif /* _WIN32 */
server_operational_state = SERVER_SHUTTING_DOWN;
sysd::notify("STOPPING=1\nSTATUS=Server shutdown in progress\n");
DBUG_PRINT("info", ("No longer listening for incoming connections"));
mysql_audit_notify(MYSQL_AUDIT_SERVER_SHUTDOWN_SHUTDOWN,
MYSQL_AUDIT_SERVER_SHUTDOWN_REASON_SHUTDOWN,
MYSQLD_SUCCESS_EXIT);
terminate_compress_gtid_table_thread();
/*
Save set of GTIDs of the last binlog into gtid_executed table
on server shutdown.
*/
if (opt_bin_log)
if (gtid_state->save_gtids_of_last_binlog_into_table())
LogErr(WARNING_LEVEL, ER_CANT_SAVE_GTIDS);
#ifndef _WIN32
mysql_mutex_lock(&LOCK_socket_listener_active);
// Notify the signal handler that we have stopped listening for connections.
socket_listener_active = false;
mysql_cond_broadcast(&COND_socket_listener_active);
mysql_mutex_unlock(&LOCK_socket_listener_active);
#endif // !_WIN32
#ifdef HAVE_PSI_THREAD_INTERFACE
/*
Disable the main thread instrumentation,
to avoid recording events during the shutdown.
*/
PSI_THREAD_CALL(delete_current_thread)();
#endif /* HAVE_PSI_THREAD_INTERFACE */
DBUG_PRINT("info", ("Waiting for shutdown proceed"));
int ret = 0;
#ifdef _WIN32
if (shutdown_restart_thr_handle.handle)
ret = my_thread_join(&shutdown_restart_thr_handle, NULL);
shutdown_restart_thr_handle.handle = NULL;
if (0 != ret)
LogErr(WARNING_LEVEL, ER_CANT_JOIN_SHUTDOWN_THREAD, "shutdown ", ret);
#else
if (signal_thread_id.thread != 0)
ret = my_thread_join(&signal_thread_id, nullptr);
signal_thread_id.thread = 0;
if (0 != ret)
LogErr(WARNING_LEVEL, ER_CANT_JOIN_SHUTDOWN_THREAD, "signal_", ret);
#endif // _WIN32
clean_up(true);
sysd::notify("STATUS=Server shutdown complete");
mysqld_exit(signal_hand_thr_exit_code);
}
看到最后一行代码没有，整个服务停止，这意味着启动的代码到此就结束了，那么也就意味着，正常的代码是走不到这里的。一般来说，在源码中看到类似这种代码，说明整个流程就基本完成了。最后看一下收尾是如何做的：
static void mysqld_exit(int exit_code) {
assert((exit_code >= MYSQLD_SUCCESS_EXIT && exit_code <= MYSQLD_ABORT_EXIT) ||
exit_code == MYSQLD_RESTART_EXIT);
mysql_audit_finalize();
Srv_session::module_deinit();
delete_optimizer_cost_module();
clean_up_mutexes();
my_end(opt_endinfo ? MY_CHECK_ERROR | MY_GIVE_INFO : 0);
destroy_error_log();
log_error_read_log_exit();
#ifdef WITH_PERFSCHEMA_STORAGE_ENGINE
shutdown_performance_schema();
#endif
#ifdef WITH_LOCK_ORDER
LO_cleanup();
#endif
#if defined(_WIN32)
if (hEventShutdown) CloseHandle(hEventShutdown);
close_service_status_pipe_in_mysqld();
#endif // _WIN32
exit(exit_code); /* purecov: inspected */
}
很典型的，断言判断后，开始各种运行模块的关闭，然后开始处理日志的关闭并停止各种支持，回收各种资源，最后区分平台性的进行关闭并发射退出代码。在下文将对整体的流程逐一进行分析介绍。
四、总结
MySql做为传统的关系型数据库中一个典型的代表，其代码对网络操作、Sql处理解析、优化控制等都有着独到之处，特别是后期中对分布式的支持，也可以做为向互联网大数据迈进的一步，都值得去认真的学习分析。
学习别人的源码，等于是在别人的智慧之河中畅游，除了锻炼了身体，更重要的是学习别人的智慧并融入自己的思想。如果上岸后，除了惊叹别人的聪明才智之后，没有任何的收获，那么读源码的初衷就大打折扣了。每个有名的开源代码，都有自己的典型的烙印，抓住这些特征，通过不断的比较分析，就会有不少的所得。
他山之石，可以攻玉，古人诚不我欺！