mirror/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-01-16 20:12:31 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions
mariadb/sql/slave.cc
unknown 7191e77539 Fixed portability issue in my_thr_init.c (was added in my last push) 
Fixed compiler warnings (detected by VC++):
- Removed not used variables
- Added casts
- Fixed wrong assignments to bool
- Fixed wrong calls with bool arguments
- Added missing argument to store(longlong), which caused wrong store method to be called.



client/mysqldump.c:
  Removed compiler warning
heap/hp_clear.c:
  Removed compiler warning
include/my_global.h:
  Removed compiler warning
include/my_tree.h:
  Changed memory limits from int to ulong
  (Allowed me to get rid of some compiler warnings)
myisam/mi_create.c:
  Removed compiler warning
myisam/myisampack.c:
  Removed compiler warning
mysys/base64.c:
  Removed compiler warning
mysys/my_thr_init.c:
  Fixed portability issue (detected on windows)
  Added DBUG_ASSERT to detect if we call my_thread_end() too many times
  Don't wait if THR_thread_count == -1 (error condition)
mysys/tree.c:
  Removed compiler warning
sql/field.cc:
  Removed compiler warning
  Fixed wrong parameter to check_date()
  Added missing argument to store(longlong)
sql/ha_archive.cc:
  Removed compiler warning
sql/ha_federated.cc:
  Removed compiler warning
sql/ha_innodb.cc:
  Removed not used variable
sql/handler.cc:
  Removed not used variable
  Fixed wrong if (we didn't detect if rollback or commit failed). Not critical as value is not yet used
sql/item.cc:
  Removed compiler warning
sql/item_func.cc:
  Removed compiler warning
sql/item_strfunc.cc:
  Removed compiler warning
sql/item_timefunc.cc:
  Removed compiler warning
sql/log.cc:
  Removed compiler warning
sql/mysql_priv.h:
  Removed compiler warning
sql/opt_range.cc:
  Removed compiler warning
sql/password.c:
  Removed compiler warning
sql/set_var.cc:
  Removed compiler warning
sql/slave.cc:
  Removed compiler warning
sql/sp.cc:
  Removed compiler warning
sql/sp_cache.cc:
  Removed compiler warning
sql/sp_head.cc:
  Removed compiler warning
  Adjusted argument to reserve() to not use up too much memory that we are probably not going to need
sql/sql_acl.cc:
  Added missing argument to store(longlong)
sql/sql_base.cc:
  Removed compiler warning
sql/sql_db.cc:
  Removed compiler warning
sql/sql_delete.cc:
  Removed compiler warning
sql/sql_handler.cc:
  Removed not used variable
sql/sql_lex.h:
  Removed not used variable
sql/sql_prepare.cc:
  Removed not used variable
sql/sql_rename.cc:
  Removed not used variable
sql/sql_select.cc:
  Fixed that select_options are not 'cut'
  Removed some not used variables
  Removed compiler warnings by adding cast
sql/sql_show.cc:
  Removed not used variables
  Added missing argument to store(longlong)
  Removed compiler warnings
sql/sql_trigger.cc:
  Removed not used variables
  Added cast to remove compiler warnings
sql/sql_update.cc:
  Fixed wrong set of bool variable
sql/sql_view.cc:
  Removed not used variables
  Added cast to get rid of compiler warnings
sql-common/client.c:
  Fixed compiler warning
sql-common/my_time.c:
  Fixed wrong argument to check_date()
  Added casts to get rid of compiler warnings
sql/sql_yacc.yy:
  Removed not used variable
sql/uniques.cc:
  Changes memory size from uint to ulong
  Added casts to get rid of compiler warnings
strings/ctype-simple.c:
  Fixed cast to get rid of compiler warnings
2006-11-30 18:25:05 +02:00

3655 lines
122 KiB
C++
Raw Blame History

/* Copyright (C) 2000-2003 MySQL AB
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
#include "mysql_priv.h"
#include <mysql.h>
#include <myisam.h>
#include "rpl_rli.h"
#include "slave.h"
#include "sql_repl.h"
#include "rpl_filter.h"
#include "repl_failsafe.h"
#include <thr_alarm.h>
#include <my_dir.h>
#include <sql_common.h>
#ifdef HAVE_REPLICATION
#include "rpl_tblmap.h"
int queue_event(MASTER_INFO* mi,const char* buf,ulong event_len);
#define MAX_SLAVE_RETRY_PAUSE 5
bool use_slave_mask = 0;
MY_BITMAP slave_error_mask;
typedef bool (*CHECK_KILLED_FUNC)(THD*,void*);
char* slave_load_tmpdir = 0;
MASTER_INFO *active_mi= 0;
my_bool replicate_same_server_id;
ulonglong relay_log_space_limit = 0;
/*
When slave thread exits, we need to remember the temporary tables so we
can re-use them on slave start.
TODO: move the vars below under MASTER_INFO
*/
int disconnect_slave_event_count = 0, abort_slave_event_count = 0;
typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE;
static int process_io_rotate(MASTER_INFO* mi, Rotate_log_event* rev);
static int process_io_create_file(MASTER_INFO* mi, Create_file_log_event* cev);
static bool wait_for_relay_log_space(RELAY_LOG_INFO* rli);
static inline bool io_slave_killed(THD* thd,MASTER_INFO* mi);
static inline bool sql_slave_killed(THD* thd,RELAY_LOG_INFO* rli);
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type);
static int safe_connect(THD* thd, MYSQL* mysql, MASTER_INFO* mi);
static int safe_reconnect(THD* thd, MYSQL* mysql, MASTER_INFO* mi,
bool suppress_warnings);
static int connect_to_master(THD* thd, MYSQL* mysql, MASTER_INFO* mi,
bool reconnect, bool suppress_warnings);
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
void* thread_killed_arg);
static int request_table_dump(MYSQL* mysql, const char* db, const char* table);
static int create_table_from_dump(THD* thd, MYSQL *mysql, const char* db,
const char* table_name, bool overwrite);
static int get_master_version_and_clock(MYSQL* mysql, MASTER_INFO* mi);
/*
Find out which replications threads are running
SYNOPSIS
init_thread_mask()
mask Return value here
mi master_info for slave
inverse If set, returns which threads are not running
IMPLEMENTATION
Get a bit mask for which threads are running so that we can later restart
these threads.
RETURN
mask If inverse == 0, running threads
If inverse == 1, stopped threads
*/
void init_thread_mask(int* mask,MASTER_INFO* mi,bool inverse)
{
bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
register int tmp_mask=0;
DBUG_ENTER("init_thread_mask");
if (set_io)
tmp_mask |= SLAVE_IO;
if (set_sql)
tmp_mask |= SLAVE_SQL;
if (inverse)
tmp_mask^= (SLAVE_IO | SLAVE_SQL);
*mask = tmp_mask;
DBUG_VOID_RETURN;
}
/*
lock_slave_threads()
*/
void lock_slave_threads(MASTER_INFO* mi)
{
DBUG_ENTER("lock_slave_threads");
//TODO: see if we can do this without dual mutex
pthread_mutex_lock(&mi->run_lock);
pthread_mutex_lock(&mi->rli.run_lock);
DBUG_VOID_RETURN;
}
/*
unlock_slave_threads()
*/
void unlock_slave_threads(MASTER_INFO* mi)
{
DBUG_ENTER("unlock_slave_threads");
//TODO: see if we can do this without dual mutex
pthread_mutex_unlock(&mi->rli.run_lock);
pthread_mutex_unlock(&mi->run_lock);
DBUG_VOID_RETURN;
}
/* Initialize slave structures */
int init_slave()
{
DBUG_ENTER("init_slave");
/*
This is called when mysqld starts. Before client connections are
accepted. However bootstrap may conflict with us if it does START SLAVE.
So it's safer to take the lock.
*/
pthread_mutex_lock(&LOCK_active_mi);
/*
TODO: re-write this to interate through the list of files
for multi-master
*/
active_mi= new MASTER_INFO;
/*
If master_host is not specified, try to read it from the master_info file.
If master_host is specified, create the master_info file if it doesn't
exists.
*/
if (!active_mi)
{
sql_print_error("Failed to allocate memory for the master info structure");
goto err;
}
if (init_master_info(active_mi,master_info_file,relay_log_info_file,
!master_host, (SLAVE_IO | SLAVE_SQL)))
{
sql_print_error("Failed to initialize the master info structure");
goto err;
}
if (server_id && !master_host && active_mi->host[0])
master_host= active_mi->host;
/* If server id is not set, start_slave_thread() will say it */
if (master_host && !opt_skip_slave_start)
{
if (start_slave_threads(1 /* need mutex */,
0 /* no wait for start*/,
active_mi,
master_info_file,
relay_log_info_file,
SLAVE_IO | SLAVE_SQL))
{
sql_print_error("Failed to create slave threads");
goto err;
}
}
pthread_mutex_unlock(&LOCK_active_mi);
DBUG_RETURN(0);
err:
pthread_mutex_unlock(&LOCK_active_mi);
DBUG_RETURN(1);
}
/*
Init function to set up array for errors that should be skipped for slave
SYNOPSIS
init_slave_skip_errors()
arg List of errors numbers to skip, separated with ','
NOTES
Called from get_options() in mysqld.cc on start-up
*/
void init_slave_skip_errors(const char* arg)
{
const char *p;
DBUG_ENTER("init_slave_skip_errors");
if (bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
{
fprintf(stderr, "Badly out of memory, please check your system status\n");
exit(1);
}
use_slave_mask = 1;
for (;my_isspace(system_charset_info,*arg);++arg)
/* empty */;
if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
{
bitmap_set_all(&slave_error_mask);
DBUG_VOID_RETURN;
}
for (p= arg ; *p; )
{
long err_code;
if (!(p= str2int(p, 10, 0, LONG_MAX, &err_code)))
break;
if (err_code < MAX_SLAVE_ERROR)
bitmap_set_bit(&slave_error_mask,(uint)err_code);
while (!my_isdigit(system_charset_info,*p) && *p)
p++;
}
DBUG_VOID_RETURN;
}
int terminate_slave_threads(MASTER_INFO* mi,int thread_mask,bool skip_lock)
{
DBUG_ENTER("terminate_slave_threads");
if (!mi->inited)
DBUG_RETURN(0); /* successfully do nothing */
int error,force_all = (thread_mask & SLAVE_FORCE_ALL);
pthread_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock;
pthread_mutex_t *sql_cond_lock,*io_cond_lock;
sql_cond_lock=sql_lock;
io_cond_lock=io_lock;
if (skip_lock)
{
sql_lock = io_lock = 0;
}
if ((thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL)) && mi->slave_running)
{
DBUG_PRINT("info",("Terminating IO thread"));
mi->abort_slave=1;
if ((error=terminate_slave_thread(mi->io_thd,io_lock,
io_cond_lock,
&mi->stop_cond,
&mi->slave_running)) &&
!force_all)
DBUG_RETURN(error);
}
if ((thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL)) && mi->rli.slave_running)
{
DBUG_PRINT("info",("Terminating SQL thread"));
DBUG_ASSERT(mi->rli.sql_thd != 0) ;
mi->rli.abort_slave=1;
if ((error=terminate_slave_thread(mi->rli.sql_thd,sql_lock,
sql_cond_lock,
&mi->rli.stop_cond,
&mi->rli.slave_running)) &&
!force_all)
DBUG_RETURN(error);
}
DBUG_RETURN(0);
}
int terminate_slave_thread(THD* thd, pthread_mutex_t* term_lock,
pthread_mutex_t *cond_lock,
pthread_cond_t* term_cond,
volatile uint *slave_running)
{
DBUG_ENTER("terminate_slave_thread");
if (term_lock)
{
pthread_mutex_lock(term_lock);
if (!*slave_running)
{
pthread_mutex_unlock(term_lock);
DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
}
}
DBUG_ASSERT(thd != 0);
THD_CHECK_SENTRY(thd);
/*
Is is critical to test if the slave is running. Otherwise, we might
be referening freed memory trying to kick it
*/
while (*slave_running) // Should always be true
{
DBUG_PRINT("loop", ("killing slave thread"));
KICK_SLAVE(thd);
/*
There is a small chance that slave thread might miss the first
alarm. To protect againts it, resend the signal until it reacts
*/
struct timespec abstime;
set_timespec(abstime,2);
pthread_cond_timedwait(term_cond, cond_lock, &abstime);
}
if (term_lock)
pthread_mutex_unlock(term_lock);
DBUG_RETURN(0);
}
int start_slave_thread(pthread_handler h_func, pthread_mutex_t *start_lock,
pthread_mutex_t *cond_lock,
pthread_cond_t *start_cond,
volatile uint *slave_running,
volatile ulong *slave_run_id,
MASTER_INFO* mi,
bool high_priority)
{
pthread_t th;
ulong start_id;
DBUG_ENTER("start_slave_thread");
DBUG_ASSERT(mi->inited);
if (start_lock)
pthread_mutex_lock(start_lock);
if (!server_id)
{
if (start_cond)
pthread_cond_broadcast(start_cond);
if (start_lock)
pthread_mutex_unlock(start_lock);
sql_print_error("Server id not set, will not start slave");
DBUG_RETURN(ER_BAD_SLAVE);
}
if (*slave_running)
{
if (start_cond)
pthread_cond_broadcast(start_cond);
if (start_lock)
pthread_mutex_unlock(start_lock);
DBUG_RETURN(ER_SLAVE_MUST_STOP);
}
start_id= *slave_run_id;
DBUG_PRINT("info",("Creating new slave thread"));
if (high_priority)
my_pthread_attr_setprio(&connection_attrib,CONNECT_PRIOR);
if (pthread_create(&th, &connection_attrib, h_func, (void*)mi))
{
if (start_lock)
pthread_mutex_unlock(start_lock);
DBUG_RETURN(ER_SLAVE_THREAD);
}
if (start_cond && cond_lock) // caller has cond_lock
{
THD* thd = current_thd;
while (start_id == *slave_run_id)
{
DBUG_PRINT("sleep",("Waiting for slave thread to start"));
const char* old_msg = thd->enter_cond(start_cond,cond_lock,
"Waiting for slave thread to start");
pthread_cond_wait(start_cond,cond_lock);
thd->exit_cond(old_msg);
pthread_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
if (thd->killed)
DBUG_RETURN(thd->killed_errno());
}
}
if (start_lock)
pthread_mutex_unlock(start_lock);
DBUG_RETURN(0);
}
/*
start_slave_threads()
NOTES
SLAVE_FORCE_ALL is not implemented here on purpose since it does not make
sense to do that for starting a slave--we always care if it actually
started the threads that were not previously running
*/
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
MASTER_INFO* mi, const char* master_info_fname,
const char* slave_info_fname, int thread_mask)
{
pthread_mutex_t *lock_io=0,*lock_sql=0,*lock_cond_io=0,*lock_cond_sql=0;
pthread_cond_t* cond_io=0,*cond_sql=0;
int error=0;
DBUG_ENTER("start_slave_threads");
if (need_slave_mutex)
{
lock_io = &mi->run_lock;
lock_sql = &mi->rli.run_lock;
}
if (wait_for_start)
{
cond_io = &mi->start_cond;
cond_sql = &mi->rli.start_cond;
lock_cond_io = &mi->run_lock;
lock_cond_sql = &mi->rli.run_lock;
}
if (thread_mask & SLAVE_IO)
error=start_slave_thread(handle_slave_io,lock_io,lock_cond_io,
cond_io,
&mi->slave_running, &mi->slave_run_id,
mi, 1); //high priority, to read the most possible
if (!error && (thread_mask & SLAVE_SQL))
{
error=start_slave_thread(handle_slave_sql,lock_sql,lock_cond_sql,
cond_sql,
&mi->rli.slave_running, &mi->rli.slave_run_id,
mi, 0);
if (error)
terminate_slave_threads(mi, thread_mask & SLAVE_IO, 0);
}
DBUG_RETURN(error);
}
#ifdef NOT_USED_YET
static int end_slave_on_walk(MASTER_INFO* mi, gptr /*unused*/)
{
DBUG_ENTER("end_slave_on_walk");
end_master_info(mi);
DBUG_RETURN(0);
}
#endif
/*
Free all resources used by slave
SYNOPSIS
end_slave()
*/
void end_slave()
{
DBUG_ENTER("end_slave");
/*
This is called when the server terminates, in close_connections().
It terminates slave threads. However, some CHANGE MASTER etc may still be
running presently. If a START SLAVE was in progress, the mutex lock below
will make us wait until slave threads have started, and START SLAVE
returns, then we terminate them here.
*/
pthread_mutex_lock(&LOCK_active_mi);
if (active_mi)
{
/*
TODO: replace the line below with
list_walk(&master_list, (list_walk_action)end_slave_on_walk,0);
once multi-master code is ready.
*/
terminate_slave_threads(active_mi,SLAVE_FORCE_ALL);
end_master_info(active_mi);
delete active_mi;
active_mi= 0;
}
pthread_mutex_unlock(&LOCK_active_mi);
DBUG_VOID_RETURN;
}
static bool io_slave_killed(THD* thd, MASTER_INFO* mi)
{
DBUG_ENTER("io_slave_killed");
DBUG_ASSERT(mi->io_thd == thd);
DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
DBUG_RETURN(mi->abort_slave || abort_loop || thd->killed);
}
static bool sql_slave_killed(THD* thd, RELAY_LOG_INFO* rli)
{
DBUG_ENTER("sql_slave_killed");
DBUG_ASSERT(rli->sql_thd == thd);
DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
if (abort_loop || thd->killed || rli->abort_slave)
{
/*
If we are in an unsafe situation (stopping could corrupt replication),
we give one minute to the slave SQL thread of grace before really
terminating, in the hope that it will be able to read more events and
the unsafe situation will soon be left. Note that this one minute starts
from the last time anything happened in the slave SQL thread. So it's
really one minute of idleness, we don't timeout if the slave SQL thread
is actively working.
*/
if (!rli->unsafe_to_stop_at)
DBUG_RETURN(1);
DBUG_PRINT("info", ("Slave SQL thread is in an unsafe situation, giving "
"it some grace period"));
if (difftime(time(0), rli->unsafe_to_stop_at) > 60)
{
slave_print_msg(ERROR_LEVEL, rli, 0,
"SQL thread had to stop in an unsafe situation, in "
"the middle of applying updates to a "
"non-transactional table without any primary key. "
"There is a risk of duplicate updates when the slave "
"SQL thread is restarted. Please check your tables' "
"contents after restart.");
DBUG_RETURN(1);
}
}
DBUG_RETURN(0);
}
/*
Writes a message to stderr, and if it's an error message, to
rli->last_slave_error and rli->last_slave_errno (which will be displayed by
SHOW SLAVE STATUS).
SYNOPSIS
slave_print_msg()
level The severity level
rli
err_code The error code
msg The message (usually related to the error code, but can
contain more information).
... (this is printf-like format, with % symbols in msg)
RETURN VALUES
void
*/
void slave_print_msg(enum loglevel level, RELAY_LOG_INFO* rli,
int err_code, const char* msg, ...)
{
void (*report_function)(const char *, ...);
char buff[MAX_SLAVE_ERRMSG], *pbuff= buff;
uint pbuffsize= sizeof(buff);
va_list args;
DBUG_ENTER("slave_print_msg");
va_start(args,msg);
switch (level)
{
case ERROR_LEVEL:
/*
This my_error call only has effect in client threads.
Slave threads do nothing in my_error().
*/
my_error(ER_UNKNOWN_ERROR, MYF(0), msg);
/*
It's an error, it must be reported in Last_error and Last_errno in SHOW
SLAVE STATUS.
*/
pbuff= rli->last_slave_error;
pbuffsize= sizeof(rli->last_slave_error);
rli->last_slave_errno = err_code;
report_function= sql_print_error;
break;
case WARNING_LEVEL:
report_function= sql_print_warning;
break;
case INFORMATION_LEVEL:
report_function= sql_print_information;
break;
default:
DBUG_ASSERT(0); // should not come here
DBUG_VOID_RETURN; // don't crash production builds, just do nothing
}
my_vsnprintf(pbuff, pbuffsize, msg, args);
/* If the msg string ends with '.', do not add a ',' it would be ugly */
if (pbuff[0] && (*(strend(pbuff)-1) == '.'))
(*report_function)("Slave: %s Error_code: %d", pbuff, err_code);
else
(*report_function)("Slave: %s, Error_code: %d", pbuff, err_code);
DBUG_VOID_RETURN;
}
/*
skip_load_data_infile()
NOTES
This is used to tell a 3.23 master to break send_file()
*/
void skip_load_data_infile(NET *net)
{
DBUG_ENTER("skip_load_data_infile");
(void)net_request_file(net, "/dev/null");
(void)my_net_read(net); // discard response
(void)net_write_command(net, 0, "", 0, "", 0); // Send ok
DBUG_VOID_RETURN;
}
bool net_request_file(NET* net, const char* fname)
{
DBUG_ENTER("net_request_file");
DBUG_RETURN(net_write_command(net, 251, fname, strlen(fname), "", 0));
}
/*
From other comments and tests in code, it looks like
sometimes Query_log_event and Load_log_event can have db == 0
(see rewrite_db() above for example)
(cases where this happens are unclear; it may be when the master is 3.23).
*/
const char *print_slave_db_safe(const char* db)
{
DBUG_ENTER("*print_slave_db_safe");
DBUG_RETURN((db ? db : ""));
}
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
const char *default_val)
{
uint length;
DBUG_ENTER("init_strvar_from_file");
if ((length=my_b_gets(f,var, max_size)))
{
char* last_p = var + length -1;
if (*last_p == '\n')
*last_p = 0; // if we stopped on newline, kill it
else
{
/*
If we truncated a line or stopped on last char, remove all chars
up to and including newline.
*/
int c;
while (((c=my_b_get(f)) != '\n' && c != my_b_EOF));
}
DBUG_RETURN(0);
}
else if (default_val)
{
strmake(var, default_val, max_size-1);
DBUG_RETURN(0);
}
DBUG_RETURN(1);
}
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
{
char buf[32];
DBUG_ENTER("init_intvar_from_file");
if (my_b_gets(f, buf, sizeof(buf)))
{
*var = atoi(buf);
DBUG_RETURN(0);
}
else if (default_val)
{
*var = default_val;
DBUG_RETURN(0);
}
DBUG_RETURN(1);
}
/*
Note that we rely on the master's version (3.23, 4.0.14 etc) instead of
relying on the binlog's version. This is not perfect: imagine an upgrade
of the master without waiting that all slaves are in sync with the master;
then a slave could be fooled about the binlog's format. This is what happens
when people upgrade a 3.23 master to 4.0 without doing RESET MASTER: 4.0
slaves are fooled. So we do this only to distinguish between 3.23 and more
recent masters (it's too late to change things for 3.23).
RETURNS
0 ok
1 error
*/
static int get_master_version_and_clock(MYSQL* mysql, MASTER_INFO* mi)
{
const char* errmsg= 0;
DBUG_ENTER("get_master_version_and_clock");
/*
Free old description_event_for_queue (that is needed if we are in
a reconnection).
*/
delete mi->rli.relay_log.description_event_for_queue;
mi->rli.relay_log.description_event_for_queue= 0;
if (!my_isdigit(&my_charset_bin,*mysql->server_version))
errmsg = "Master reported unrecognized MySQL version";
else
{
/*
Note the following switch will bug when we have MySQL branch 30 ;)
*/
switch (*mysql->server_version)
{
case '0':
case '1':
case '2':
errmsg = "Master reported unrecognized MySQL version";
break;
case '3':
mi->rli.relay_log.description_event_for_queue= new
Format_description_log_event(1, mysql->server_version);
break;
case '4':
mi->rli.relay_log.description_event_for_queue= new
Format_description_log_event(3, mysql->server_version);
break;
default:
/*
Master is MySQL >=5.0. Give a default Format_desc event, so that we can
take the early steps (like tests for "is this a 3.23 master") which we
have to take before we receive the real master's Format_desc which will
override this one. Note that the Format_desc we create below is garbage
(it has the format of the *slave*); it's only good to help know if the
master is 3.23, 4.0, etc.
*/
mi->rli.relay_log.description_event_for_queue= new
Format_description_log_event(4, mysql->server_version);
break;
}
}
/*
This does not mean that a 5.0 slave will be able to read a 6.0 master; but
as we don't know yet, we don't want to forbid this for now. If a 5.0 slave
can't read a 6.0 master, this will show up when the slave can't read some
events sent by the master, and there will be error messages.
*/
if (errmsg)
{
sql_print_error(errmsg);
DBUG_RETURN(1);
}
/* as we are here, we tried to allocate the event */
if (!mi->rli.relay_log.description_event_for_queue)
{
sql_print_error("Slave I/O thread failed to create a default Format_description_log_event");
DBUG_RETURN(1);
}
/*
Compare the master and slave's clock. Do not die if master's clock is
unavailable (very old master not supporting UNIX_TIMESTAMP()?).
*/
MYSQL_RES *master_res= 0;
MYSQL_ROW master_row;
if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
(master_res= mysql_store_result(mysql)) &&
(master_row= mysql_fetch_row(master_res)))
{
mi->clock_diff_with_master=
(long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
}
else
{
mi->clock_diff_with_master= 0; /* The "most sensible" value */
sql_print_warning("\"SELECT UNIX_TIMESTAMP()\" failed on master, \
do not trust column Seconds_Behind_Master of SHOW SLAVE STATUS");
}
if (master_res)
mysql_free_result(master_res);
/*
Check that the master's server id and ours are different. Because if they
are equal (which can result from a simple copy of master's datadir to slave,
thus copying some my.cnf), replication will work but all events will be
skipped.
Do not die if SHOW VARIABLES LIKE 'SERVER_ID' fails on master (very old
master?).
Note: we could have put a @@SERVER_ID in the previous SELECT
UNIX_TIMESTAMP() instead, but this would not have worked on 3.23 masters.
*/
if (!mysql_real_query(mysql,
STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
(master_res= mysql_store_result(mysql)))
{
if ((master_row= mysql_fetch_row(master_res)) &&
(::server_id == strtoul(master_row[1], 0, 10)) &&
!replicate_same_server_id)
errmsg= "The slave I/O thread stops because master and slave have equal \
MySQL server ids; these ids must be different for replication to work (or \
the --replicate-same-server-id option must be used on slave but this does \
not always make sense; please check the manual before using it).";
mysql_free_result(master_res);
}
/*
Check that the master's global character_set_server and ours are the same.
Not fatal if query fails (old master?).
Note that we don't check for equality of global character_set_client and
collation_connection (neither do we prevent their setting in
set_var.cc). That's because from what I (Guilhem) have tested, the global
values of these 2 are never used (new connections don't use them).
We don't test equality of global collation_database either as it's is
going to be deprecated (made read-only) in 4.1 very soon.
The test is only relevant if master < 5.0.3 (we'll test only if it's older
than the 5 branch; < 5.0.3 was alpha...), as >= 5.0.3 master stores
charset info in each binlog event.
We don't do it for 3.23 because masters <3.23.50 hang on
SELECT @@unknown_var (BUG#7965 - see changelog of 3.23.50). So finally we
test only if master is 4.x.
*/
/* redundant with rest of code but safer against later additions */
if (*mysql->server_version == '3')
goto err;
if ((*mysql->server_version == '4') &&
!mysql_real_query(mysql,
STRING_WITH_LEN("SELECT @@GLOBAL.COLLATION_SERVER")) &&
(master_res= mysql_store_result(mysql)))
{
if ((master_row= mysql_fetch_row(master_res)) &&
strcmp(master_row[0], global_system_variables.collation_server->name))
errmsg= "The slave I/O thread stops because master and slave have \
different values for the COLLATION_SERVER global variable. The values must \
be equal for replication to work";
mysql_free_result(master_res);
}
/*
Perform analogous check for time zone. Theoretically we also should
perform check here to verify that SYSTEM time zones are the same on
slave and master, but we can't rely on value of @@system_time_zone
variable (it is time zone abbreviation) since it determined at start
time and so could differ for slave and master even if they are really
in the same system time zone. So we are omiting this check and just
relying on documentation. Also according to Monty there are many users
who are using replication between servers in various time zones. Hence
such check will broke everything for them. (And now everything will
work for them because by default both their master and slave will have
'SYSTEM' time zone).
This check is only necessary for 4.x masters (and < 5.0.4 masters but
those were alpha).
*/
if ((*mysql->server_version == '4') &&
!mysql_real_query(mysql, STRING_WITH_LEN("SELECT @@GLOBAL.TIME_ZONE")) &&
(master_res= mysql_store_result(mysql)))
{
if ((master_row= mysql_fetch_row(master_res)) &&
strcmp(master_row[0],
global_system_variables.time_zone->get_name()->ptr()))
errmsg= "The slave I/O thread stops because master and slave have \
different values for the TIME_ZONE global variable. The values must \
be equal for replication to work";
mysql_free_result(master_res);
}
err:
if (errmsg)
{
sql_print_error(errmsg);
DBUG_RETURN(1);
}
DBUG_RETURN(0);
}
/*
Used by fetch_master_table (used by LOAD TABLE tblname FROM MASTER and LOAD
DATA FROM MASTER). Drops the table (if 'overwrite' is true) and recreates it
from the dump. Honours replication inclusion/exclusion rules.
db must be non-zero (guarded by assertion).
RETURN VALUES
0 success
1 error
*/
static int create_table_from_dump(THD* thd, MYSQL *mysql, const char* db,
const char* table_name, bool overwrite)
{
ulong packet_len;
char *query, *save_db;
uint32 save_db_length;
Vio* save_vio;
HA_CHECK_OPT check_opt;
TABLE_LIST tables;
int error= 1;
handler *file;
ulonglong save_options;
NET *net= &mysql->net;
DBUG_ENTER("create_table_from_dump");
packet_len= my_net_read(net); // read create table statement
if (packet_len == packet_error)
{
my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
DBUG_RETURN(1);
}
if (net->read_pos[0] == 255) // error from master
{
char *err_msg;
err_msg= (char*) net->read_pos + ((mysql->server_capabilities &
CLIENT_PROTOCOL_41) ?
3+SQLSTATE_LENGTH+1 : 3);
my_error(ER_MASTER, MYF(0), err_msg);
DBUG_RETURN(1);
}
thd->command = COM_TABLE_DUMP;
thd->query_length= packet_len;
/* Note that we should not set thd->query until the area is initalized */
if (!(query = thd->strmake((char*) net->read_pos, packet_len)))
{
sql_print_error("create_table_from_dump: out of memory");
my_message(ER_GET_ERRNO, "Out of memory", MYF(0));
DBUG_RETURN(1);
}
thd->query= query;
thd->query_error = 0;
thd->net.no_send_ok = 1;
bzero((char*) &tables,sizeof(tables));
tables.db = (char*)db;
tables.alias= tables.table_name= (char*)table_name;
/* Drop the table if 'overwrite' is true */
if (overwrite && mysql_rm_table(thd,&tables,1,0)) /* drop if exists */
{
sql_print_error("create_table_from_dump: failed to drop the table");
goto err;
}
/* Create the table. We do not want to log the "create table" statement */
save_options = thd->options;
thd->options &= ~(ulong) (OPTION_BIN_LOG);
thd->proc_info = "Creating table from master dump";
// save old db in case we are creating in a different database
save_db = thd->db;
save_db_length= thd->db_length;
thd->db = (char*)db;
DBUG_ASSERT(thd->db != 0);
thd->db_length= strlen(thd->db);
mysql_parse(thd, thd->query, packet_len); // run create table
thd->db = save_db; // leave things the way the were before
thd->db_length= save_db_length;
thd->options = save_options;
if (thd->query_error)
goto err; // mysql_parse took care of the error send
thd->proc_info = "Opening master dump table";
tables.lock_type = TL_WRITE;
if (!open_ltable(thd, &tables, TL_WRITE))
{
sql_print_error("create_table_from_dump: could not open created table");
goto err;
}
file = tables.table->file;
thd->proc_info = "Reading master dump table data";
/* Copy the data file */
if (file->net_read_dump(net))
{
my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
sql_print_error("create_table_from_dump: failed in\
handler::net_read_dump()");
goto err;
}
check_opt.init();
check_opt.flags|= T_VERY_SILENT | T_CALC_CHECKSUM | T_QUICK;
thd->proc_info = "Rebuilding the index on master dump table";
/*
We do not want repair() to spam us with messages
just send them to the error log, and report the failure in case of
problems.
*/
save_vio = thd->net.vio;
thd->net.vio = 0;
/* Rebuild the index file from the copied data file (with REPAIR) */
error=file->ha_repair(thd,&check_opt) != 0;
thd->net.vio = save_vio;
if (error)
my_error(ER_INDEX_REBUILD, MYF(0), tables.table->s->table_name.str);
err:
close_thread_tables(thd);
thd->net.no_send_ok = 0;
DBUG_RETURN(error);
}
int fetch_master_table(THD *thd, const char *db_name, const char *table_name,
MASTER_INFO *mi, MYSQL *mysql, bool overwrite)
{
int error= 1;
const char *errmsg=0;
bool called_connected= (mysql != NULL);
DBUG_ENTER("fetch_master_table");
DBUG_PRINT("enter", ("db_name: '%s' table_name: '%s'",
db_name,table_name));
if (!called_connected)
{
if (!(mysql = mysql_init(NULL)))
{
DBUG_RETURN(1);
}
if (connect_to_master(thd, mysql, mi))
{
my_error(ER_CONNECT_TO_MASTER, MYF(0), mysql_error(mysql));
/*
We need to clear the active VIO since, theoretically, somebody
might issue an awake() on this thread. If we are then in the
middle of closing and destroying the VIO inside the
mysql_close(), we will have a problem.
*/
#ifdef SIGNAL_WITH_VIO_CLOSE
thd->clear_active_vio();
#endif
mysql_close(mysql);
DBUG_RETURN(1);
}
if (thd->killed)
goto err;
}
if (request_table_dump(mysql, db_name, table_name))
{
error= ER_UNKNOWN_ERROR;
errmsg= "Failed on table dump request";
goto err;
}
if (create_table_from_dump(thd, mysql, db_name,
table_name, overwrite))
goto err; // create_table_from_dump have sent the error already
error = 0;
err:
thd->net.no_send_ok = 0; // Clear up garbage after create_table_from_dump
if (!called_connected)
mysql_close(mysql);
if (errmsg && thd->vio_ok())
my_message(error, errmsg, MYF(0));
DBUG_RETURN(test(error)); // Return 1 on error
}
static bool wait_for_relay_log_space(RELAY_LOG_INFO* rli)
{
bool slave_killed=0;
MASTER_INFO* mi = rli->mi;
const char *save_proc_info;
THD* thd = mi->io_thd;
DBUG_ENTER("wait_for_relay_log_space");
pthread_mutex_lock(&rli->log_space_lock);
save_proc_info= thd->enter_cond(&rli->log_space_cond,
&rli->log_space_lock,
"\
Waiting for the slave SQL thread to free enough relay log space");
while (rli->log_space_limit < rli->log_space_total &&
!(slave_killed=io_slave_killed(thd,mi)) &&
!rli->ignore_log_space_limit)
pthread_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
thd->exit_cond(save_proc_info);
DBUG_RETURN(slave_killed);
}
/*
Builds a Rotate from the ignored events' info and writes it to relay log.
SYNOPSIS
write_ignored_events_info_to_relay_log()
thd pointer to I/O thread's thd
mi
DESCRIPTION
Slave I/O thread, going to die, must leave a durable trace of the
ignored events' end position for the use of the slave SQL thread, by
calling this function. Only that thread can call it (see assertion).
*/
static void write_ignored_events_info_to_relay_log(THD *thd, MASTER_INFO *mi)
{
RELAY_LOG_INFO *rli= &mi->rli;
pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
DBUG_ENTER("write_ignored_events_info_to_relay_log");
DBUG_ASSERT(thd == mi->io_thd);
pthread_mutex_lock(log_lock);
if (rli->ign_master_log_name_end[0])
{
DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
0, rli->ign_master_log_pos_end,
Rotate_log_event::DUP_NAME);
rli->ign_master_log_name_end[0]= 0;
/* can unlock before writing as slave SQL thd will soon see our Rotate */
pthread_mutex_unlock(log_lock);
if (likely((bool)ev))
{
ev->server_id= 0; // don't be ignored by slave SQL thread
if (unlikely(rli->relay_log.append(ev)))
sql_print_error("Slave I/O thread failed to write a Rotate event"
" to the relay log, "
"SHOW SLAVE STATUS may be inaccurate");
rli->relay_log.harvest_bytes_written(&rli->log_space_total);
if (flush_master_info(mi, 1))
sql_print_error("Failed to flush master info file");
delete ev;
}
else
sql_print_error("Slave I/O thread failed to create a Rotate event"
" (out of memory?), "
"SHOW SLAVE STATUS may be inaccurate");
}
else
pthread_mutex_unlock(log_lock);
DBUG_VOID_RETURN;
}
int register_slave_on_master(MYSQL* mysql)
{
char buf[1024], *pos= buf;
uint report_host_len, report_user_len=0, report_password_len=0;
DBUG_ENTER("register_slave_on_master");
if (!report_host)
DBUG_RETURN(0);
report_host_len= strlen(report_host);
if (report_user)
report_user_len= strlen(report_user);
if (report_password)
report_password_len= strlen(report_password);
/* 30 is a good safety margin */
if (report_host_len + report_user_len + report_password_len + 30 >
sizeof(buf))
DBUG_RETURN(0); // safety
int4store(pos, server_id); pos+= 4;
pos= net_store_data(pos, report_host, report_host_len);
pos= net_store_data(pos, report_user, report_user_len);
pos= net_store_data(pos, report_password, report_password_len);
int2store(pos, (uint16) report_port); pos+= 2;
int4store(pos, rpl_recovery_rank); pos+= 4;
/* The master will fill in master_id */
int4store(pos, 0); pos+= 4;
if (simple_command(mysql, COM_REGISTER_SLAVE, (char*) buf,
(uint) (pos- buf), 0))
{
sql_print_error("Error on COM_REGISTER_SLAVE: %d '%s'",
mysql_errno(mysql),
mysql_error(mysql));
DBUG_RETURN(1);
}
DBUG_RETURN(0);
}
bool show_master_info(THD* thd, MASTER_INFO* mi)
{
// TODO: fix this for multi-master
List<Item> field_list;
Protocol *protocol= thd->protocol;
DBUG_ENTER("show_master_info");
field_list.push_back(new Item_empty_string("Slave_IO_State",
14));
field_list.push_back(new Item_empty_string("Master_Host",
sizeof(mi->host)));
field_list.push_back(new Item_empty_string("Master_User",
sizeof(mi->user)));
field_list.push_back(new Item_return_int("Master_Port", 7,
MYSQL_TYPE_LONG));
field_list.push_back(new Item_return_int("Connect_Retry", 10,
MYSQL_TYPE_LONG));
field_list.push_back(new Item_empty_string("Master_Log_File",
FN_REFLEN));
field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
MYSQL_TYPE_LONGLONG));
field_list.push_back(new Item_empty_string("Relay_Log_File",
FN_REFLEN));
field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
MYSQL_TYPE_LONGLONG));
field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
FN_REFLEN));
field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
field_list.push_back(new Item_empty_string("Replicate_Do_DB", 20));
field_list.push_back(new Item_empty_string("Replicate_Ignore_DB", 20));
field_list.push_back(new Item_empty_string("Replicate_Do_Table", 20));
field_list.push_back(new Item_empty_string("Replicate_Ignore_Table", 23));
field_list.push_back(new Item_empty_string("Replicate_Wild_Do_Table", 24));
field_list.push_back(new Item_empty_string("Replicate_Wild_Ignore_Table",
28));
field_list.push_back(new Item_return_int("Last_Errno", 4, MYSQL_TYPE_LONG));
field_list.push_back(new Item_empty_string("Last_Error", 20));
field_list.push_back(new Item_return_int("Skip_Counter", 10,
MYSQL_TYPE_LONG));
field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
MYSQL_TYPE_LONGLONG));
field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
MYSQL_TYPE_LONGLONG));
field_list.push_back(new Item_empty_string("Until_Condition", 6));
field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
MYSQL_TYPE_LONGLONG));
field_list.push_back(new Item_empty_string("Master_SSL_Allowed", 7));
field_list.push_back(new Item_empty_string("Master_SSL_CA_File",
sizeof(mi->ssl_ca)));
field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
sizeof(mi->ssl_capath)));
field_list.push_back(new Item_empty_string("Master_SSL_Cert",
sizeof(mi->ssl_cert)));
field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
sizeof(mi->ssl_cipher)));
field_list.push_back(new Item_empty_string("Master_SSL_Key",
sizeof(mi->ssl_key)));
field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
MYSQL_TYPE_LONGLONG));
if (protocol->send_fields(&field_list,
Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
DBUG_RETURN(TRUE);
if (mi->host[0])
{
DBUG_PRINT("info",("host is set: '%s'", mi->host));
String *packet= &thd->packet;
protocol->prepare_for_resend();
/*
TODO: we read slave_running without run_lock, whereas these variables
are updated under run_lock and not data_lock. In 5.0 we should lock
run_lock on top of data_lock (with good order).
*/
pthread_mutex_lock(&mi->data_lock);
pthread_mutex_lock(&mi->rli.data_lock);
protocol->store(mi->io_thd ? mi->io_thd->proc_info : "", &my_charset_bin);
protocol->store(mi->host, &my_charset_bin);
protocol->store(mi->user, &my_charset_bin);
protocol->store((uint32) mi->port);
protocol->store((uint32) mi->connect_retry);
protocol->store(mi->master_log_name, &my_charset_bin);
protocol->store((ulonglong) mi->master_log_pos);
protocol->store(mi->rli.group_relay_log_name +
dirname_length(mi->rli.group_relay_log_name),
&my_charset_bin);
protocol->store((ulonglong) mi->rli.group_relay_log_pos);
protocol->store(mi->rli.group_master_log_name, &my_charset_bin);
protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
"Yes" : "No", &my_charset_bin);
protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
protocol->store(rpl_filter->get_do_db());
protocol->store(rpl_filter->get_ignore_db());
char buf[256];
String tmp(buf, sizeof(buf), &my_charset_bin);
rpl_filter->get_do_table(&tmp);
protocol->store(&tmp);
rpl_filter->get_ignore_table(&tmp);
protocol->store(&tmp);
rpl_filter->get_wild_do_table(&tmp);
protocol->store(&tmp);
rpl_filter->get_wild_ignore_table(&tmp);
protocol->store(&tmp);
protocol->store((uint32) mi->rli.last_slave_errno);
protocol->store(mi->rli.last_slave_error, &my_charset_bin);
protocol->store((uint32) mi->rli.slave_skip_counter);
protocol->store((ulonglong) mi->rli.group_master_log_pos);
protocol->store((ulonglong) mi->rli.log_space_total);
protocol->store(
mi->rli.until_condition==RELAY_LOG_INFO::UNTIL_NONE ? "None":
( mi->rli.until_condition==RELAY_LOG_INFO::UNTIL_MASTER_POS? "Master":
"Relay"), &my_charset_bin);
protocol->store(mi->rli.until_log_name, &my_charset_bin);
protocol->store((ulonglong) mi->rli.until_log_pos);
#ifdef HAVE_OPENSSL
protocol->store(mi->ssl? "Yes":"No", &my_charset_bin);
#else
protocol->store(mi->ssl? "Ignored":"No", &my_charset_bin);
#endif
protocol->store(mi->ssl_ca, &my_charset_bin);
protocol->store(mi->ssl_capath, &my_charset_bin);
protocol->store(mi->ssl_cert, &my_charset_bin);
protocol->store(mi->ssl_cipher, &my_charset_bin);
protocol->store(mi->ssl_key, &my_charset_bin);
/*
Seconds_Behind_Master: if SQL thread is running and I/O thread is
connected, we can compute it otherwise show NULL (i.e. unknown).
*/
if ((mi->slave_running == MYSQL_SLAVE_RUN_CONNECT) &&
mi->rli.slave_running)
{
long tmp= (long)((time_t)time((time_t*) 0)
- mi->rli.last_master_timestamp)
- mi->clock_diff_with_master;
/*
Apparently on some systems tmp can be <0. Here are possible reasons
related to MySQL:
- the master is itself a slave of another master whose time is ahead.
- somebody used an explicit SET TIMESTAMP on the master.
Possible reason related to granularity-to-second of time functions
(nothing to do with MySQL), which can explain a value of -1:
assume the master's and slave's time are perfectly synchronized, and
that at slave's connection time, when the master's timestamp is read,
it is at the very end of second 1, and (a very short time later) when
the slave's timestamp is read it is at the very beginning of second
2. Then the recorded value for master is 1 and the recorded value for
slave is 2. At SHOW SLAVE STATUS time, assume that the difference
between timestamp of slave and rli->last_master_timestamp is 0
(i.e. they are in the same second), then we get 0-(2-1)=-1 as a result.
This confuses users, so we don't go below 0: hence the max().
last_master_timestamp == 0 (an "impossible" timestamp 1970) is a
special marker to say "consider we have caught up".
*/
protocol->store((longlong)(mi->rli.last_master_timestamp ? max(0, tmp)
: 0));
}
else
protocol->store_null();
pthread_mutex_unlock(&mi->rli.data_lock);
pthread_mutex_unlock(&mi->data_lock);
if (my_net_write(&thd->net, (char*)thd->packet.ptr(), packet->length()))
DBUG_RETURN(TRUE);
}
send_eof(thd);
DBUG_RETURN(FALSE);
}
void set_slave_thread_options(THD* thd)
{
DBUG_ENTER("set_slave_thread_options");
/*
It's nonsense to constrain the slave threads with max_join_size; if a
query succeeded on master, we HAVE to execute it. So set
OPTION_BIG_SELECTS. Setting max_join_size to HA_POS_ERROR is not enough
(and it's not needed if we have OPTION_BIG_SELECTS) because an INSERT
SELECT examining more than 4 billion rows would still fail (yes, because
when max_join_size is 4G, OPTION_BIG_SELECTS is automatically set, but
only for client threads.
*/
ulonglong options= thd->options | OPTION_BIG_SELECTS;
if (opt_log_slave_updates)
options|= OPTION_BIN_LOG;
else
options&= ~OPTION_BIN_LOG;
thd->options= options;
thd->variables.completion_type= 0;
DBUG_VOID_RETURN;
}
void set_slave_thread_default_charset(THD* thd, RELAY_LOG_INFO *rli)
{
DBUG_ENTER("set_slave_thread_default_charset");
thd->variables.character_set_client=
global_system_variables.character_set_client;
thd->variables.collation_connection=
global_system_variables.collation_connection;
thd->variables.collation_server=
global_system_variables.collation_server;
thd->update_charset();
rli->cached_charset_invalidate();
DBUG_VOID_RETURN;
}
/*
init_slave_thread()
*/
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type)
{
DBUG_ENTER("init_slave_thread");
thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
thd->security_ctx->skip_grants();
my_net_init(&thd->net, 0);
/*
Adding MAX_LOG_EVENT_HEADER_LEN to the max_allowed_packet on all
slave threads, since a replication event can become this much larger
than the corresponding packet (query) sent from client to master.
*/
thd->variables.max_allowed_packet= global_system_variables.max_allowed_packet
+ MAX_LOG_EVENT_HEADER; /* note, incr over the global not session var */
thd->net.read_timeout = slave_net_timeout;
thd->slave_thread = 1;
set_slave_thread_options(thd);
thd->client_capabilities = CLIENT_LOCAL_FILES;
thd->real_id=pthread_self();
pthread_mutex_lock(&LOCK_thread_count);
thd->thread_id = thread_id++;
pthread_mutex_unlock(&LOCK_thread_count);
if (init_thr_lock() || thd->store_globals())
{
thd->cleanup();
delete thd;
DBUG_RETURN(-1);
}
#if !defined(__WIN__) && !defined(__NETWARE__)
sigset_t set;
VOID(sigemptyset(&set)); // Get mask in use
VOID(pthread_sigmask(SIG_UNBLOCK,&set,&thd->block_signals));
#endif
if (thd_type == SLAVE_THD_SQL)
thd->proc_info= "Waiting for the next event in relay log";
else
thd->proc_info= "Waiting for master update";
thd->version=refresh_version;
thd->set_time();
DBUG_RETURN(0);
}
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
void* thread_killed_arg)
{
int nap_time;
thr_alarm_t alarmed;
DBUG_ENTER("safe_sleep");
thr_alarm_init(&alarmed);
time_t start_time= time((time_t*) 0);
time_t end_time= start_time+sec;
while ((nap_time= (int) (end_time - start_time)) > 0)
{
ALARM alarm_buff;
/*
The only reason we are asking for alarm is so that
we will be woken up in case of murder, so if we do not get killed,
set the alarm so it goes off after we wake up naturally
*/
thr_alarm(&alarmed, 2 * nap_time, &alarm_buff);
sleep(nap_time);
thr_end_alarm(&alarmed);
if ((*thread_killed)(thd,thread_killed_arg))
DBUG_RETURN(1);
start_time=time((time_t*) 0);
}
DBUG_RETURN(0);
}
static int request_dump(MYSQL* mysql, MASTER_INFO* mi,
bool *suppress_warnings)
{
char buf[FN_REFLEN + 10];
int len;
int binlog_flags = 0; // for now
char* logname = mi->master_log_name;
DBUG_ENTER("request_dump");
// TODO if big log files: Change next to int8store()
int4store(buf, (ulong) mi->master_log_pos);
int2store(buf + 4, binlog_flags);
int4store(buf + 6, server_id);
len = (uint) strlen(logname);
memcpy(buf + 10, logname,len);
if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
{
/*
Something went wrong, so we will just reconnect and retry later
in the future, we should do a better error analysis, but for
now we just fill up the error log :-)
*/
if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
*suppress_warnings= 1; // Suppress reconnect warning
else
sql_print_error("Error on COM_BINLOG_DUMP: %d %s, will retry in %d secs",
mysql_errno(mysql), mysql_error(mysql),
master_connect_retry);
DBUG_RETURN(1);
}
DBUG_RETURN(0);
}
static int request_table_dump(MYSQL* mysql, const char* db, const char* table)
{
char buf[1024];
DBUG_ENTER("request_table_dump");
char * p = buf;
uint table_len = (uint) strlen(table);
uint db_len = (uint) strlen(db);
if (table_len + db_len > sizeof(buf) - 2)
{
sql_print_error("request_table_dump: Buffer overrun");
DBUG_RETURN(1);
}
*p++ = db_len;
memcpy(p, db, db_len);
p += db_len;
*p++ = table_len;
memcpy(p, table, table_len);
if (simple_command(mysql, COM_TABLE_DUMP, buf, p - buf + table_len, 1))
{
sql_print_error("request_table_dump: Error sending the table dump \
command");
DBUG_RETURN(1);
}
DBUG_RETURN(0);
}
/*
Read one event from the master
SYNOPSIS
read_event()
mysql MySQL connection
mi Master connection information
suppress_warnings TRUE when a normal net read timeout has caused us to
try a reconnect. We do not want to print anything to
the error log in this case because this a anormal
event in an idle server.
RETURN VALUES
'packet_error' Error
number Length of packet
*/
static ulong read_event(MYSQL* mysql, MASTER_INFO *mi, bool* suppress_warnings)
{
ulong len;
DBUG_ENTER("read_event");
*suppress_warnings= 0;
/*
my_real_read() will time us out
We check if we were told to die, and if not, try reading again
*/
#ifndef DBUG_OFF
if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
DBUG_RETURN(packet_error);
#endif
len = cli_safe_read(mysql);
if (len == packet_error || (long) len < 1)
{
if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
{
/*
We are trying a normal reconnect after a read timeout;
we suppress prints to .err file as long as the reconnect
happens without problems
*/
*suppress_warnings= TRUE;
}
else
sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
mysql_error(mysql), mysql_errno(mysql));
DBUG_RETURN(packet_error);
}
/* Check if eof packet */
if (len < 8 && mysql->net.read_pos[0] == 254)
{
sql_print_information("Slave: received end packet from server, apparent "
"master shutdown: %s",
mysql_error(mysql));
DBUG_RETURN(packet_error);
}
DBUG_PRINT("exit", ("len: %lu net->read_pos[4]: %d",
len, mysql->net.read_pos[4]));
DBUG_RETURN(len - 1);
}
int check_expected_error(THD* thd, RELAY_LOG_INFO* rli, int expected_error)
{
DBUG_ENTER("check_expected_error");
switch (expected_error) {
case ER_NET_READ_ERROR:
case ER_NET_ERROR_ON_WRITE:
case ER_SERVER_SHUTDOWN:
case ER_NEW_ABORTING_CONNECTION:
DBUG_RETURN(1);
default:
DBUG_RETURN(0);
}
}
/*
Check if the current error is of temporary nature of not.
Some errors are temporary in nature, such as
ER_LOCK_DEADLOCK and ER_LOCK_WAIT_TIMEOUT. Ndb also signals
that the error is temporary by pushing a warning with the error code
ER_GET_TEMPORARY_ERRMSG, if the originating error is temporary.
*/
static int has_temporary_error(THD *thd)
{
DBUG_ENTER("has_temporary_error");
if (thd->is_fatal_error)
DBUG_RETURN(0);
/*
Temporary error codes:
currently, InnoDB deadlock detected by InnoDB or lock
wait timeout (innodb_lock_wait_timeout exceeded
*/
if (thd->net.last_errno == ER_LOCK_DEADLOCK ||
thd->net.last_errno == ER_LOCK_WAIT_TIMEOUT)
DBUG_RETURN(1);
#ifdef HAVE_NDB_BINLOG
/*
currently temporary error set in ndbcluster
*/
List_iterator_fast<MYSQL_ERROR> it(thd->warn_list);
MYSQL_ERROR *err;
while ((err= it++))
{
DBUG_PRINT("info", ("has warning %d %s", err->code, err->msg));
switch (err->code)
{
case ER_GET_TEMPORARY_ERRMSG:
DBUG_RETURN(1);
default:
break;
}
}
#endif
DBUG_RETURN(0);
}
static int exec_relay_log_event(THD* thd, RELAY_LOG_INFO* rli)
{
DBUG_ENTER("exec_relay_log_event");
/*
We acquire this mutex since we need it for all operations except
event execution. But we will release it in places where we will
wait for something for example inside of next_event().
*/
pthread_mutex_lock(&rli->data_lock);
/*
This tests if the position of the end of the last previous executed event
hits the UNTIL barrier.
We would prefer to test if the position of the start (or possibly) end of
the to-be-read event hits the UNTIL barrier, this is different if there
was an event ignored by the I/O thread just before (BUG#13861 to be
fixed).
*/
if (rli->until_condition!=RELAY_LOG_INFO::UNTIL_NONE &&
rli->is_until_satisfied())
{
char buf[22];
sql_print_information("Slave SQL thread stopped because it reached its"
" UNTIL position %s", llstr(rli->until_pos(), buf));
/*
Setting abort_slave flag because we do not want additional message about
error in query execution to be printed.
*/
rli->abort_slave= 1;
pthread_mutex_unlock(&rli->data_lock);
DBUG_RETURN(1);
}
Log_event * ev = next_event(rli);
DBUG_ASSERT(rli->sql_thd==thd);
if (sql_slave_killed(thd,rli))
{
pthread_mutex_unlock(&rli->data_lock);
delete ev;
DBUG_RETURN(1);
}
if (ev)
{
int type_code = ev->get_type_code();
int exec_res;
/*
Queries originating from this server must be skipped.
Low-level events (Format_desc, Rotate, Stop) from this server
must also be skipped. But for those we don't want to modify
group_master_log_pos, because these events did not exist on the master.
Format_desc is not completely skipped.
Skip queries specified by the user in slave_skip_counter.
We can't however skip events that has something to do with the
log files themselves.
Filtering on own server id is extremely important, to ignore execution of
events created by the creation/rotation of the relay log (remember that
now the relay log starts with its Format_desc, has a Rotate etc).
*/
DBUG_PRINT("info",("type_code=%d, server_id=%d",type_code,ev->server_id));
if ((ev->server_id == (uint32) ::server_id &&
!replicate_same_server_id &&
type_code != FORMAT_DESCRIPTION_EVENT) ||
(rli->slave_skip_counter &&
type_code != ROTATE_EVENT && type_code != STOP_EVENT &&
type_code != START_EVENT_V3 && type_code!= FORMAT_DESCRIPTION_EVENT))
{
DBUG_PRINT("info", ("event skipped"));
/*
We only skip the event here and do not increase the group log
position. In the event that we have to restart, this means
that we might have to skip the event again, but that is a
minor issue.
If we were to increase the group log position when skipping an
event, it might be that we are restarting at the wrong
position and have events before that we should have executed,
so not increasing the group log position is a sure bet in this
case.
In this way, we just step the group log position when we
*know* that we are at the end of a group.
*/
rli->inc_event_relay_log_pos();
/*
Protect against common user error of setting the counter to 1
instead of 2 while recovering from an insert which used auto_increment,
rand or user var.
*/
if (rli->slave_skip_counter &&
!((type_code == INTVAR_EVENT ||
type_code == RAND_EVENT ||
type_code == USER_VAR_EVENT) &&
rli->slave_skip_counter == 1) &&
/*
The events from ourselves which have something to do with the relay
log itself must be skipped, true, but they mustn't decrement
rli->slave_skip_counter, because the user is supposed to not see
these events (they are not in the master's binlog) and if we
decremented, START SLAVE would for example decrement when it sees
the Rotate, so the event which the user probably wanted to skip
would not be skipped.
*/
!(ev->server_id == (uint32) ::server_id &&
(type_code == ROTATE_EVENT || type_code == STOP_EVENT ||
type_code == START_EVENT_V3 || type_code == FORMAT_DESCRIPTION_EVENT)))
--rli->slave_skip_counter;
pthread_mutex_unlock(&rli->data_lock);
delete ev;
DBUG_RETURN(0); // avoid infinite update loops
}
pthread_mutex_unlock(&rli->data_lock);
thd->server_id = ev->server_id; // use the original server id for logging
thd->set_time(); // time the query
thd->lex->current_select= 0;
if (!ev->when)
ev->when = time(NULL);
ev->thd = thd; // because up to this point, ev->thd == 0
exec_res = ev->exec_event(rli);
DBUG_PRINT("info", ("exec_event result: %d", exec_res));
DBUG_ASSERT(rli->sql_thd==thd);
/*
Format_description_log_event should not be deleted because it will be
used to read info about the relay log's format; it will be deleted when
the SQL thread does not need it, i.e. when this thread terminates.
*/
if (ev->get_type_code() != FORMAT_DESCRIPTION_EVENT)
{
DBUG_PRINT("info", ("Deleting the event after it has been executed"));
delete ev;
}
if (slave_trans_retries)
{
if (exec_res && has_temporary_error(thd))
{
const char *errmsg;
/*
We were in a transaction which has been rolled back because of a
temporary error;
let's seek back to BEGIN log event and retry it all again.
Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
there is no rollback since 5.0.13 (ref: manual).
We have to not only seek but also
a) init_master_info(), to seek back to hot relay log's start for later
(for when we will come back to this hot log after re-processing the
possibly existing old logs where BEGIN is: check_binlog_magic() will
then need the cache to be at position 0 (see comments at beginning of
init_master_info()).
b) init_relay_log_pos(), because the BEGIN may be an older relay log.
*/
if (rli->trans_retries < slave_trans_retries)
{
if (init_master_info(rli->mi, 0, 0, 0, SLAVE_SQL))
sql_print_error("Failed to initialize the master info structure");
else if (init_relay_log_pos(rli,
rli->group_relay_log_name,
rli->group_relay_log_pos,
1, &errmsg, 1))
sql_print_error("Error initializing relay log position: %s",
errmsg);
else
{
exec_res= 0;
end_trans(thd, ROLLBACK);
/* chance for concurrent connection to get more locks */
safe_sleep(thd, min(rli->trans_retries, MAX_SLAVE_RETRY_PAUSE),
(CHECK_KILLED_FUNC)sql_slave_killed, (void*)rli);
pthread_mutex_lock(&rli->data_lock); // because of SHOW STATUS
rli->trans_retries++;
rli->retried_trans++;
pthread_mutex_unlock(&rli->data_lock);
DBUG_PRINT("info", ("Slave retries transaction "
"rli->trans_retries: %lu", rli->trans_retries));
}
}
else
sql_print_error("Slave SQL thread retried transaction %lu time(s) "
"in vain, giving up. Consider raising the value of "
"the slave_transaction_retries variable.",
slave_trans_retries);
}
else if (!((thd->options & OPTION_BEGIN) && opt_using_transactions))
{
/*
Only reset the retry counter if the event succeeded or
failed with a non-transient error. On a successful event,
the execution will proceed as usual; in the case of a
non-transient error, the slave will stop with an error.
*/
rli->trans_retries= 0; // restart from fresh
}
}
DBUG_RETURN(exec_res);
}
pthread_mutex_unlock(&rli->data_lock);
slave_print_msg(ERROR_LEVEL, rli, 0, "\
Could not parse relay log event entry. The possible reasons are: the master's \
binary log is corrupted (you can check this by running 'mysqlbinlog' on the \
binary log), the slave's relay log is corrupted (you can check this by running \
'mysqlbinlog' on the relay log), a network problem, or a bug in the master's \
or slave's MySQL code. If you want to check the master's binary log or slave's \
relay log, you will be able to know their names by issuing 'SHOW SLAVE STATUS' \
on this slave.\
");
DBUG_RETURN(1);
}
/* Slave I/O Thread entry point */
pthread_handler_t handle_slave_io(void *arg)
{
THD *thd; // needs to be first for thread_stack
MYSQL *mysql;
MASTER_INFO *mi = (MASTER_INFO*)arg;
RELAY_LOG_INFO *rli= &mi->rli;
char llbuff[22];
uint retry_count;
// needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
my_thread_init();
DBUG_ENTER("handle_slave_io");
DBUG_ASSERT(mi->inited);
mysql= NULL ;
retry_count= 0;
pthread_mutex_lock(&mi->run_lock);
/* Inform waiting threads that slave has started */
mi->slave_run_id++;
#ifndef DBUG_OFF
mi->events_till_disconnect = disconnect_slave_event_count;
#endif
thd= new THD; // note that contructor of THD uses DBUG_ !
THD_CHECK_SENTRY(thd);
pthread_detach_this_thread();
thd->thread_stack= (char*) &thd; // remember where our stack is
if (init_slave_thread(thd, SLAVE_THD_IO))
{
pthread_cond_broadcast(&mi->start_cond);
pthread_mutex_unlock(&mi->run_lock);
sql_print_error("Failed during slave I/O thread initialization");
goto err;
}
mi->io_thd = thd;
pthread_mutex_lock(&LOCK_thread_count);
threads.append(thd);
pthread_mutex_unlock(&LOCK_thread_count);
mi->slave_running = 1;
mi->abort_slave = 0;
pthread_mutex_unlock(&mi->run_lock);
pthread_cond_broadcast(&mi->start_cond);
DBUG_PRINT("master_info",("log_file_name: '%s' position: %s",
mi->master_log_name,
llstr(mi->master_log_pos,llbuff)));
if (!(mi->mysql = mysql = mysql_init(NULL)))
{
sql_print_error("Slave I/O thread: error in mysql_init()");
goto err;
}
thd->proc_info = "Connecting to master";
// we can get killed during safe_connect
if (!safe_connect(thd, mysql, mi))
{
sql_print_information("Slave I/O thread: connected to master '%s@%s:%d',"
"replication started in log '%s' at position %s",
mi->user, mi->host, mi->port,
IO_RPL_LOG_NAME,
llstr(mi->master_log_pos,llbuff));
/*
Adding MAX_LOG_EVENT_HEADER_LEN to the max_packet_size on the I/O
thread, since a replication event can become this much larger than
the corresponding packet (query) sent from client to master.
*/
mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
}
else
{
sql_print_information("Slave I/O thread killed while connecting to master");
goto err;
}
connected:
// TODO: the assignment below should be under mutex (5.0)
mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
thd->slave_net = &mysql->net;
thd->proc_info = "Checking master version";
if (get_master_version_and_clock(mysql, mi))
goto err;
if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
{
/*
Register ourselves with the master.
If fails, this is not fatal - we just print the error message and go
on with life.
*/
thd->proc_info = "Registering slave on master";
if (register_slave_on_master(mysql) || update_slave_list(mysql, mi))
goto err;
}
DBUG_PRINT("info",("Starting reading binary log from master"));
while (!io_slave_killed(thd,mi))
{
bool suppress_warnings= 0;
thd->proc_info = "Requesting binlog dump";
if (request_dump(mysql, mi, &suppress_warnings))
{
sql_print_error("Failed on request_dump()");
if (io_slave_killed(thd,mi))
{
sql_print_information("Slave I/O thread killed while requesting master \
dump");
goto err;
}
mi->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT;
thd->proc_info= "Waiting to reconnect after a failed binlog dump request";
#ifdef SIGNAL_WITH_VIO_CLOSE
thd->clear_active_vio();
#endif
end_server(mysql);
/*
First time retry immediately, assuming that we can recover
right away - if first time fails, sleep between re-tries
hopefuly the admin can fix the problem sometime
*/
if (retry_count++)
{
if (retry_count > master_retry_count)
goto err; // Don't retry forever
safe_sleep(thd,mi->connect_retry,(CHECK_KILLED_FUNC)io_slave_killed,
(void*)mi);
}
if (io_slave_killed(thd,mi))
{
sql_print_information("Slave I/O thread killed while retrying master \
dump");
goto err;
}
thd->proc_info = "Reconnecting after a failed binlog dump request";
if (!suppress_warnings)
sql_print_error("Slave I/O thread: failed dump request, \
reconnecting to try again, log '%s' at postion %s", IO_RPL_LOG_NAME,
llstr(mi->master_log_pos,llbuff));
if (safe_reconnect(thd, mysql, mi, suppress_warnings) ||
io_slave_killed(thd,mi))
{
sql_print_information("Slave I/O thread killed during or \
after reconnect");
goto err;
}
goto connected;
}
while (!io_slave_killed(thd,mi))
{
bool suppress_warnings= 0;
/*
We say "waiting" because read_event() will wait if there's nothing to
read. But if there's something to read, it will not wait. The
important thing is to not confuse users by saying "reading" whereas
we're in fact receiving nothing.
*/
thd->proc_info = "Waiting for master to send event";
ulong event_len = read_event(mysql, mi, &suppress_warnings);
if (io_slave_killed(thd,mi))
{
if (global_system_variables.log_warnings)
sql_print_information("Slave I/O thread killed while reading event");
goto err;
}
if (event_len == packet_error)
{
uint mysql_error_number= mysql_errno(mysql);
if (mysql_error_number == ER_NET_PACKET_TOO_LARGE)
{
sql_print_error("\
Log entry on master is longer than max_allowed_packet (%ld) on \
slave. If the entry is correct, restart the server with a higher value of \
max_allowed_packet",
thd->variables.max_allowed_packet);
goto err;
}
if (mysql_error_number == ER_MASTER_FATAL_ERROR_READING_BINLOG)
{
sql_print_error(ER(mysql_error_number), mysql_error_number,
mysql_error(mysql));
goto err;
}
mi->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT;
thd->proc_info = "Waiting to reconnect after a failed master event read";
#ifdef SIGNAL_WITH_VIO_CLOSE
thd->clear_active_vio();
#endif
end_server(mysql);
if (retry_count++)
{
if (retry_count > master_retry_count)
goto err; // Don't retry forever
safe_sleep(thd,mi->connect_retry,(CHECK_KILLED_FUNC)io_slave_killed,
(void*) mi);
}
if (io_slave_killed(thd,mi))
{
if (global_system_variables.log_warnings)
sql_print_information("Slave I/O thread killed while waiting to \
reconnect after a failed read");
goto err;
}
thd->proc_info = "Reconnecting after a failed master event read";
if (!suppress_warnings)
sql_print_information("Slave I/O thread: Failed reading log event, \
reconnecting to retry, log '%s' position %s", IO_RPL_LOG_NAME,
llstr(mi->master_log_pos, llbuff));
if (safe_reconnect(thd, mysql, mi, suppress_warnings) ||
io_slave_killed(thd,mi))
{
if (global_system_variables.log_warnings)
sql_print_information("Slave I/O thread killed during or after a \
reconnect done to recover from failed read");
goto err;
}
goto connected;
} // if (event_len == packet_error)
retry_count=0; // ok event, reset retry counter
thd->proc_info = "Queueing master event to the relay log";
if (queue_event(mi,(const char*)mysql->net.read_pos + 1,
event_len))
{
sql_print_error("Slave I/O thread could not queue event from master");
goto err;
}
if (flush_master_info(mi, 1))
{
sql_print_error("Failed to flush master info file");
goto err;
}
/*
See if the relay logs take too much space.
We don't lock mi->rli.log_space_lock here; this dirty read saves time
and does not introduce any problem:
- if mi->rli.ignore_log_space_limit is 1 but becomes 0 just after (so
the clean value is 0), then we are reading only one more event as we
should, and we'll block only at the next event. No big deal.
- if mi->rli.ignore_log_space_limit is 0 but becomes 1 just after (so
the clean value is 1), then we are going into wait_for_relay_log_space()
for no reason, but this function will do a clean read, notice the clean
value and exit immediately.
*/
#ifndef DBUG_OFF
{
char llbuf1[22], llbuf2[22];
DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
llstr(rli->log_space_limit,llbuf1),
llstr(rli->log_space_total,llbuf2),
(int) rli->ignore_log_space_limit));
}
#endif
if (rli->log_space_limit && rli->log_space_limit <
rli->log_space_total &&
!rli->ignore_log_space_limit)
if (wait_for_relay_log_space(rli))
{
sql_print_error("Slave I/O thread aborted while waiting for relay \
log space");
goto err;
}
}
}
// error = 0;
err:
// print the current replication position
sql_print_information("Slave I/O thread exiting, read up to log '%s', position %s",
IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
VOID(pthread_mutex_lock(&LOCK_thread_count));
thd->query = thd->db = 0; // extra safety
thd->query_length= thd->db_length= 0;
VOID(pthread_mutex_unlock(&LOCK_thread_count));
if (mysql)
{
/*
Here we need to clear the active VIO before closing the
connection with the master. The reason is that THD::awake()
might be called from terminate_slave_thread() because somebody
issued a STOP SLAVE. If that happends, the close_active_vio()
can be called in the middle of closing the VIO associated with
the 'mysql' object, causing a crash.
*/
#ifdef SIGNAL_WITH_VIO_CLOSE
thd->clear_active_vio();
#endif
mysql_close(mysql);
mi->mysql=0;
}
write_ignored_events_info_to_relay_log(thd, mi);
thd->proc_info = "Waiting for slave mutex on exit";
pthread_mutex_lock(&mi->run_lock);
/* Forget the relay log's format */
delete mi->rli.relay_log.description_event_for_queue;
mi->rli.relay_log.description_event_for_queue= 0;
// TODO: make rpl_status part of MASTER_INFO
change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
DBUG_ASSERT(thd->net.buff != 0);
net_end(&thd->net); // destructor will not free it, because net.vio is 0
close_thread_tables(thd, 0);
pthread_mutex_lock(&LOCK_thread_count);
THD_CHECK_SENTRY(thd);
delete thd;
pthread_mutex_unlock(&LOCK_thread_count);
mi->abort_slave = 0;
mi->slave_running = 0;
mi->io_thd = 0;
pthread_mutex_unlock(&mi->run_lock);
pthread_cond_broadcast(&mi->stop_cond); // tell the world we are done
my_thread_end();
pthread_exit(0);
DBUG_RETURN(0); // Can't return anything here
}
/* Slave SQL Thread entry point */
pthread_handler_t handle_slave_sql(void *arg)
{
THD *thd; /* needs to be first for thread_stack */
char llbuff[22],llbuff1[22];
RELAY_LOG_INFO* rli = &((MASTER_INFO*)arg)->rli;
const char *errmsg;
// needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
my_thread_init();
DBUG_ENTER("handle_slave_sql");
DBUG_ASSERT(rli->inited);
pthread_mutex_lock(&rli->run_lock);
DBUG_ASSERT(!rli->slave_running);
errmsg= 0;
#ifndef DBUG_OFF
rli->events_till_abort = abort_slave_event_count;
#endif
thd = new THD; // note that contructor of THD uses DBUG_ !
thd->thread_stack = (char*)&thd; // remember where our stack is
/* Inform waiting threads that slave has started */
rli->slave_run_id++;
pthread_detach_this_thread();
if (init_slave_thread(thd, SLAVE_THD_SQL))
{
/*
TODO: this is currently broken - slave start and change master
will be stuck if we fail here
*/
pthread_cond_broadcast(&rli->start_cond);
pthread_mutex_unlock(&rli->run_lock);
sql_print_error("Failed during slave thread initialization");
goto err;
}
thd->init_for_queries();
rli->sql_thd= thd;
thd->temporary_tables = rli->save_temporary_tables; // restore temp tables
pthread_mutex_lock(&LOCK_thread_count);
threads.append(thd);
pthread_mutex_unlock(&LOCK_thread_count);
/*
We are going to set slave_running to 1. Assuming slave I/O thread is
alive and connected, this is going to make Seconds_Behind_Master be 0
i.e. "caught up". Even if we're just at start of thread. Well it's ok, at
the moment we start we can think we are caught up, and the next second we
start receiving data so we realize we are not caught up and
Seconds_Behind_Master grows. No big deal.
*/
rli->slave_running = 1;
rli->abort_slave = 0;
pthread_mutex_unlock(&rli->run_lock);
pthread_cond_broadcast(&rli->start_cond);
/*
Reset errors for a clean start (otherwise, if the master is idle, the SQL
thread may execute no Query_log_event, so the error will remain even
though there's no problem anymore). Do not reset the master timestamp
(imagine the slave has caught everything, the STOP SLAVE and START SLAVE:
as we are not sure that we are going to receive a query, we want to
remember the last master timestamp (to say how many seconds behind we are
now.
But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
*/
rli->clear_slave_error();
//tell the I/O thread to take relay_log_space_limit into account from now on
pthread_mutex_lock(&rli->log_space_lock);
rli->ignore_log_space_limit= 0;
pthread_mutex_unlock(&rli->log_space_lock);
rli->trans_retries= 0; // start from "no error"
if (init_relay_log_pos(rli,
rli->group_relay_log_name,
rli->group_relay_log_pos,
1 /*need data lock*/, &errmsg,
1 /*look for a description_event*/))
{
sql_print_error("Error initializing relay log position: %s",
errmsg);
goto err;
}
THD_CHECK_SENTRY(thd);
#ifndef DBUG_OFF
{
char llbuf1[22], llbuf2[22];
DBUG_PRINT("info", ("my_b_tell(rli->cur_log)=%s rli->event_relay_log_pos=%s",
llstr(my_b_tell(rli->cur_log),llbuf1),
llstr(rli->event_relay_log_pos,llbuf2)));
DBUG_ASSERT(rli->event_relay_log_pos >= BIN_LOG_HEADER_SIZE);
/*
Wonder if this is correct. I (Guilhem) wonder if my_b_tell() returns the
correct position when it's called just after my_b_seek() (the questionable
stuff is those "seek is done on next read" comments in the my_b_seek()
source code).
The crude reality is that this assertion randomly fails whereas
replication seems to work fine. And there is no easy explanation why it
fails (as we my_b_seek(rli->event_relay_log_pos) at the very end of
init_relay_log_pos() called above). Maybe the assertion would be
meaningful if we held rli->data_lock between the my_b_seek() and the
DBUG_ASSERT().
*/
#ifdef SHOULD_BE_CHECKED
DBUG_ASSERT(my_b_tell(rli->cur_log) == rli->event_relay_log_pos);
#endif
}
#endif
DBUG_ASSERT(rli->sql_thd == thd);
DBUG_PRINT("master_info",("log_file_name: %s position: %s",
rli->group_master_log_name,
llstr(rli->group_master_log_pos,llbuff)));
if (global_system_variables.log_warnings)
sql_print_information("Slave SQL thread initialized, starting replication in \
log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
llstr(rli->group_relay_log_pos,llbuff1));
/* execute init_slave variable */
if (sys_init_slave.value_length)
{
execute_init_command(thd, &sys_init_slave, &LOCK_sys_init_slave);
if (thd->query_error)
{
sql_print_error("\
Slave SQL thread aborted. Can't execute init_slave query");
goto err;
}
}
/* Read queries from the IO/THREAD until this thread is killed */
while (!sql_slave_killed(thd,rli))
{
thd->proc_info = "Reading event from the relay log";
DBUG_ASSERT(rli->sql_thd == thd);
THD_CHECK_SENTRY(thd);
if (exec_relay_log_event(thd,rli))
{
// do not scare the user if SQL thread was simply killed or stopped
if (!sql_slave_killed(thd,rli))
{
/*
retrieve as much info as possible from the thd and, error codes and warnings
and print this to the error log as to allow the user to locate the error
*/
if (thd->net.last_errno != 0)
{
if (rli->last_slave_errno == 0)
{
slave_print_msg(ERROR_LEVEL, rli, thd->net.last_errno,
thd->net.last_error ?
thd->net.last_error : "<no message>");
}
else if (rli->last_slave_errno != (int) thd->net.last_errno)
{
sql_print_error("Slave (additional info): %s Error_code: %d",
thd->net.last_error ?
thd->net.last_error : "<no message>",
thd->net.last_errno);
}
}
/* Print any warnings issued */
List_iterator_fast<MYSQL_ERROR> it(thd->warn_list);
MYSQL_ERROR *err;
while ((err= it++))
sql_print_warning("Slave: %s Error_code: %d",err->msg, err->code);
sql_print_error("\
Error running query, slave SQL thread aborted. Fix the problem, and restart \
the slave SQL thread with \"SLAVE START\". We stopped at log \
'%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
}
goto err;
}
}
/* Thread stopped. Print the current replication position to the log */
sql_print_information("Slave SQL thread exiting, replication stopped in log "
"'%s' at position %s",
RPL_LOG_NAME, llstr(rli->group_master_log_pos,llbuff));
err:
/*
Some events set some playgrounds, which won't be cleared because thread
stops. Stopping of this thread may not be known to these events ("stop"
request is detected only by the present function, not by events), so we
must "proactively" clear playgrounds:
*/
rli->cleanup_context(thd, 1);
VOID(pthread_mutex_lock(&LOCK_thread_count));
/*
Some extra safety, which should not been needed (normally, event deletion
should already have done these assignments (each event which sets these
variables is supposed to set them to 0 before terminating)).
*/
thd->query= thd->db= thd->catalog= 0;
thd->query_length= thd->db_length= 0;
VOID(pthread_mutex_unlock(&LOCK_thread_count));
thd->proc_info = "Waiting for slave mutex on exit";
pthread_mutex_lock(&rli->run_lock);
/* We need data_lock, at least to wake up any waiting master_pos_wait() */
pthread_mutex_lock(&rli->data_lock);
DBUG_ASSERT(rli->slave_running == 1); // tracking buffer overrun
/* When master_pos_wait() wakes up it will check this and terminate */
rli->slave_running= 0;
/* Forget the relay log's format */
delete rli->relay_log.description_event_for_exec;
rli->relay_log.description_event_for_exec= 0;
/* Wake up master_pos_wait() */
pthread_mutex_unlock(&rli->data_lock);
DBUG_PRINT("info",("Signaling possibly waiting master_pos_wait() functions"));
pthread_cond_broadcast(&rli->data_cond);
rli->ignore_log_space_limit= 0; /* don't need any lock */
/* we die so won't remember charset - re-update them on next thread start */
rli->cached_charset_invalidate();
rli->save_temporary_tables = thd->temporary_tables;
/*
TODO: see if we can do this conditionally in next_event() instead
to avoid unneeded position re-init
*/
thd->temporary_tables = 0; // remove tempation from destructor to close them
DBUG_ASSERT(thd->net.buff != 0);
net_end(&thd->net); // destructor will not free it, because we are weird
DBUG_ASSERT(rli->sql_thd == thd);
THD_CHECK_SENTRY(thd);
rli->sql_thd= 0;
pthread_mutex_lock(&LOCK_thread_count);
THD_CHECK_SENTRY(thd);
delete thd;
pthread_mutex_unlock(&LOCK_thread_count);
pthread_cond_broadcast(&rli->stop_cond);
// tell the world we are done
pthread_mutex_unlock(&rli->run_lock);
my_thread_end();
pthread_exit(0);
DBUG_RETURN(0); // Can't return anything here
}
/*
process_io_create_file()
*/
static int process_io_create_file(MASTER_INFO* mi, Create_file_log_event* cev)
{
int error = 1;
ulong num_bytes;
bool cev_not_written;
THD *thd = mi->io_thd;
NET *net = &mi->mysql->net;
DBUG_ENTER("process_io_create_file");
if (unlikely(!cev->is_valid()))
DBUG_RETURN(1);
if (!rpl_filter->db_ok(cev->db))
{
skip_load_data_infile(net);
DBUG_RETURN(0);
}
DBUG_ASSERT(cev->inited_from_old);
thd->file_id = cev->file_id = mi->file_id++;
thd->server_id = cev->server_id;
cev_not_written = 1;
if (unlikely(net_request_file(net,cev->fname)))
{
sql_print_error("Slave I/O: failed requesting download of '%s'",
cev->fname);
goto err;
}
/*
This dummy block is so we could instantiate Append_block_log_event
once and then modify it slightly instead of doing it multiple times
in the loop
*/
{
Append_block_log_event aev(thd,0,0,0,0);
for (;;)
{
if (unlikely((num_bytes=my_net_read(net)) == packet_error))
{
sql_print_error("Network read error downloading '%s' from master",
cev->fname);
goto err;
}
if (unlikely(!num_bytes)) /* eof */
{
net_write_command(net, 0, "", 0, "", 0);/* 3.23 master wants it */
/*
If we wrote Create_file_log_event, then we need to write
Execute_load_log_event. If we did not write Create_file_log_event,
then this is an empty file and we can just do as if the LOAD DATA
INFILE had not existed, i.e. write nothing.
*/
if (unlikely(cev_not_written))
break;
Execute_load_log_event xev(thd,0,0);
xev.log_pos = cev->log_pos;
if (unlikely(mi->rli.relay_log.append(&xev)))
{
sql_print_error("Slave I/O: error writing Exec_load event to \
relay log");
goto err;
}
mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
break;
}
if (unlikely(cev_not_written))
{
cev->block = (char*)net->read_pos;
cev->block_len = num_bytes;
if (unlikely(mi->rli.relay_log.append(cev)))
{
sql_print_error("Slave I/O: error writing Create_file event to \
relay log");
goto err;
}
cev_not_written=0;
mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
}
else
{
aev.block = (char*)net->read_pos;
aev.block_len = num_bytes;
aev.log_pos = cev->log_pos;
if (unlikely(mi->rli.relay_log.append(&aev)))
{
sql_print_error("Slave I/O: error writing Append_block event to \
relay log");
goto err;
}
mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
}
}
}
error=0;
err:
DBUG_RETURN(error);
}
/*
Start using a new binary log on the master
SYNOPSIS
process_io_rotate()
mi master_info for the slave
rev The rotate log event read from the binary log
DESCRIPTION
Updates the master info with the place in the next binary
log where we should start reading.
Rotate the relay log to avoid mixed-format relay logs.
NOTES
We assume we already locked mi->data_lock
RETURN VALUES
0 ok
1 Log event is illegal
*/
static int process_io_rotate(MASTER_INFO *mi, Rotate_log_event *rev)
{
DBUG_ENTER("process_io_rotate");
safe_mutex_assert_owner(&mi->data_lock);
if (unlikely(!rev->is_valid()))
DBUG_RETURN(1);
/* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
mi->master_log_pos= rev->pos;
DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
mi->master_log_name, (ulong) mi->master_log_pos));
#ifndef DBUG_OFF
/*
If we do not do this, we will be getting the first
rotate event forever, so we need to not disconnect after one.
*/
if (disconnect_slave_event_count)
mi->events_till_disconnect++;
#endif
/*
If description_event_for_queue is format <4, there is conversion in the
relay log to the slave's format (4). And Rotate can mean upgrade or
nothing. If upgrade, it's to 5.0 or newer, so we will get a Format_desc, so
no need to reset description_event_for_queue now. And if it's nothing (same
master version as before), no need (still using the slave's format).
*/
if (mi->rli.relay_log.description_event_for_queue->binlog_version >= 4)
{
delete mi->rli.relay_log.description_event_for_queue;
/* start from format 3 (MySQL 4.0) again */
mi->rli.relay_log.description_event_for_queue= new
Format_description_log_event(3);
}
/*
Rotate the relay log makes binlog format detection easier (at next slave
start or mysqlbinlog)
*/
rotate_relay_log(mi); /* will take the right mutexes */
DBUG_RETURN(0);
}
/*
Reads a 3.23 event and converts it to the slave's format. This code was
copied from MySQL 4.0.
*/
static int queue_binlog_ver_1_event(MASTER_INFO *mi, const char *buf,
ulong event_len)
{
const char *errmsg = 0;
ulong inc_pos;
bool ignore_event= 0;
char *tmp_buf = 0;
RELAY_LOG_INFO *rli= &mi->rli;
DBUG_ENTER("queue_binlog_ver_1_event");
/*
If we get Load event, we need to pass a non-reusable buffer
to read_log_event, so we do a trick
*/
if (buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
{
if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
{
sql_print_error("Slave I/O: out of memory for Load event");
DBUG_RETURN(1);
}
memcpy(tmp_buf,buf,event_len);
/*
Create_file constructor wants a 0 as last char of buffer, this 0 will
serve as the string-termination char for the file's name (which is at the
end of the buffer)
We must increment event_len, otherwise the event constructor will not see
this end 0, which leads to segfault.
*/
tmp_buf[event_len++]=0;
int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
buf = (const char*)tmp_buf;
}
/*
This will transform LOAD_EVENT into CREATE_FILE_EVENT, ask the master to
send the loaded file, and write it to the relay log in the form of
Append_block/Exec_load (the SQL thread needs the data, as that thread is not
connected to the master).
*/
Log_event *ev = Log_event::read_log_event(buf,event_len, &errmsg,
mi->rli.relay_log.description_event_for_queue);
if (unlikely(!ev))
{
sql_print_error("Read invalid event from master: '%s',\
master could be corrupt but a more likely cause of this is a bug",
errmsg);
my_free((char*) tmp_buf, MYF(MY_ALLOW_ZERO_PTR));
DBUG_RETURN(1);
}
pthread_mutex_lock(&mi->data_lock);
ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
switch (ev->get_type_code()) {
case STOP_EVENT:
ignore_event= 1;
inc_pos= event_len;
break;
case ROTATE_EVENT:
if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
{
delete ev;
pthread_mutex_unlock(&mi->data_lock);
DBUG_RETURN(1);
}
inc_pos= 0;
break;
case CREATE_FILE_EVENT:
/*
Yes it's possible to have CREATE_FILE_EVENT here, even if we're in
queue_old_event() which is for 3.23 events which don't comprise
CREATE_FILE_EVENT. This is because read_log_event() above has just
transformed LOAD_EVENT into CREATE_FILE_EVENT.
*/
{
/* We come here when and only when tmp_buf != 0 */
DBUG_ASSERT(tmp_buf != 0);
inc_pos=event_len;
ev->log_pos+= inc_pos;
int error = process_io_create_file(mi,(Create_file_log_event*)ev);
delete ev;
mi->master_log_pos += inc_pos;
DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
pthread_mutex_unlock(&mi->data_lock);
my_free((char*)tmp_buf, MYF(0));
DBUG_RETURN(error);
}
default:
inc_pos= event_len;
break;
}
if (likely(!ignore_event))
{
if (ev->log_pos)
/*
Don't do it for fake Rotate events (see comment in
Log_event::Log_event(const char* buf...) in log_event.cc).
*/
ev->log_pos+= event_len; /* make log_pos be the pos of the end of the event */
if (unlikely(rli->relay_log.append(ev)))
{
delete ev;
pthread_mutex_unlock(&mi->data_lock);
DBUG_RETURN(1);
}
rli->relay_log.harvest_bytes_written(&rli->log_space_total);
}
delete ev;
mi->master_log_pos+= inc_pos;
DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
pthread_mutex_unlock(&mi->data_lock);
DBUG_RETURN(0);
}
/*
Reads a 4.0 event and converts it to the slave's format. This code was copied
from queue_binlog_ver_1_event(), with some affordable simplifications.
*/
static int queue_binlog_ver_3_event(MASTER_INFO *mi, const char *buf,
ulong event_len)
{
const char *errmsg = 0;
ulong inc_pos;
char *tmp_buf = 0;
RELAY_LOG_INFO *rli= &mi->rli;
DBUG_ENTER("queue_binlog_ver_3_event");
/* read_log_event() will adjust log_pos to be end_log_pos */
Log_event *ev = Log_event::read_log_event(buf,event_len, &errmsg,
mi->rli.relay_log.description_event_for_queue);
if (unlikely(!ev))
{
sql_print_error("Read invalid event from master: '%s',\
master could be corrupt but a more likely cause of this is a bug",
errmsg);
my_free((char*) tmp_buf, MYF(MY_ALLOW_ZERO_PTR));
DBUG_RETURN(1);
}
pthread_mutex_lock(&mi->data_lock);
switch (ev->get_type_code()) {
case STOP_EVENT:
goto err;
case ROTATE_EVENT:
if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
{
delete ev;
pthread_mutex_unlock(&mi->data_lock);
DBUG_RETURN(1);
}
inc_pos= 0;
break;
default:
inc_pos= event_len;
break;
}
if (unlikely(rli->relay_log.append(ev)))
{
delete ev;
pthread_mutex_unlock(&mi->data_lock);
DBUG_RETURN(1);
}
rli->relay_log.harvest_bytes_written(&rli->log_space_total);
delete ev;
mi->master_log_pos+= inc_pos;
err:
DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
pthread_mutex_unlock(&mi->data_lock);
DBUG_RETURN(0);
}
/*
queue_old_event()
Writes a 3.23 or 4.0 event to the relay log, after converting it to the 5.0
(exactly, slave's) format. To do the conversion, we create a 5.0 event from
the 3.23/4.0 bytes, then write this event to the relay log.
TODO:
Test this code before release - it has to be tested on a separate
setup with 3.23 master or 4.0 master
*/
static int queue_old_event(MASTER_INFO *mi, const char *buf,
ulong event_len)
{
DBUG_ENTER("queue_old_event");
switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
{
case 1:
DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
case 3:
DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
default: /* unsupported format; eg version 2 */
DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
mi->rli.relay_log.description_event_for_queue->binlog_version));
DBUG_RETURN(1);
}
}
/*
queue_event()
If the event is 3.23/4.0, passes it to queue_old_event() which will convert
it. Otherwise, writes a 5.0 (or newer) event to the relay log. Then there is
no format conversion, it's pure read/write of bytes.
So a 5.0.0 slave's relay log can contain events in the slave's format or in
any >=5.0.0 format.
*/
int queue_event(MASTER_INFO* mi,const char* buf, ulong event_len)
{
int error= 0;
ulong inc_pos;
RELAY_LOG_INFO *rli= &mi->rli;
pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
DBUG_ENTER("queue_event");
if (mi->rli.relay_log.description_event_for_queue->binlog_version<4 &&
buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT /* a way to escape */)
DBUG_RETURN(queue_old_event(mi,buf,event_len));
LINT_INIT(inc_pos);
pthread_mutex_lock(&mi->data_lock);
switch (buf[EVENT_TYPE_OFFSET]) {
case STOP_EVENT:
/*
We needn't write this event to the relay log. Indeed, it just indicates a
master server shutdown. The only thing this does is cleaning. But
cleaning is already done on a per-master-thread basis (as the master
server is shutting down cleanly, it has written all DROP TEMPORARY TABLE
prepared statements' deletion are TODO only when we binlog prep stmts).
We don't even increment mi->master_log_pos, because we may be just after
a Rotate event. Btw, in a few milliseconds we are going to have a Start
event from the next binlog (unless the master is presently running
without --log-bin).
*/
goto err;
case ROTATE_EVENT:
{
Rotate_log_event rev(buf,event_len,mi->rli.relay_log.description_event_for_queue);
if (unlikely(process_io_rotate(mi,&rev)))
{
error= 1;
goto err;
}
/*
Now the I/O thread has just changed its mi->master_log_name, so
incrementing mi->master_log_pos is nonsense.
*/
inc_pos= 0;
break;
}
case FORMAT_DESCRIPTION_EVENT:
{
/*
Create an event, and save it (when we rotate the relay log, we will have
to write this event again).
*/
/*
We are the only thread which reads/writes description_event_for_queue.
The relay_log struct does not move (though some members of it can
change), so we needn't any lock (no rli->data_lock, no log lock).
*/
Format_description_log_event* tmp;
const char* errmsg;
if (!(tmp= (Format_description_log_event*)
Log_event::read_log_event(buf, event_len, &errmsg,
mi->rli.relay_log.description_event_for_queue)))
{
error= 2;
goto err;
}
delete mi->rli.relay_log.description_event_for_queue;
mi->rli.relay_log.description_event_for_queue= tmp;
/*
Though this does some conversion to the slave's format, this will
preserve the master's binlog format version, and number of event types.
*/
/*
If the event was not requested by the slave (the slave did not ask for
it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
*/
inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
DBUG_PRINT("info",("binlog format is now %d",
mi->rli.relay_log.description_event_for_queue->binlog_version));
}
break;
default:
inc_pos= event_len;
break;
}
/*
If this event is originating from this server, don't queue it.
We don't check this for 3.23 events because it's simpler like this; 3.23
will be filtered anyway by the SQL slave thread which also tests the
server id (we must also keep this test in the SQL thread, in case somebody
upgrades a 4.0 slave which has a not-filtered relay log).
ANY event coming from ourselves can be ignored: it is obvious for queries;
for STOP_EVENT/ROTATE_EVENT/START_EVENT: these cannot come from ourselves
(--log-slave-updates would not log that) unless this slave is also its
direct master (an unsupported, useless setup!).
*/
pthread_mutex_lock(log_lock);
if ((uint4korr(buf + SERVER_ID_OFFSET) == ::server_id) &&
!replicate_same_server_id)
{
/*
Do not write it to the relay log.
a) We still want to increment mi->master_log_pos, so that we won't
re-read this event from the master if the slave IO thread is now
stopped/restarted (more efficient if the events we are ignoring are big
LOAD DATA INFILE).
b) We want to record that we are skipping events, for the information of
the slave SQL thread, otherwise that thread may let
rli->group_relay_log_pos stay too small if the last binlog's event is
ignored.
But events which were generated by this slave and which do not exist in
the master's binlog (i.e. Format_desc, Rotate & Stop) should not increment
mi->master_log_pos.
*/
if (buf[EVENT_TYPE_OFFSET]!=FORMAT_DESCRIPTION_EVENT &&
buf[EVENT_TYPE_OFFSET]!=ROTATE_EVENT &&
buf[EVENT_TYPE_OFFSET]!=STOP_EVENT)
{
mi->master_log_pos+= inc_pos;
memcpy(rli->ign_master_log_name_end, mi->master_log_name, FN_REFLEN);
DBUG_ASSERT(rli->ign_master_log_name_end[0]);
rli->ign_master_log_pos_end= mi->master_log_pos;
}
rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
DBUG_PRINT("info", ("master_log_pos: %lu, event originating from the same server, ignored",
(ulong) mi->master_log_pos));
}
else
{
/* write the event to the relay log */
if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
{
mi->master_log_pos+= inc_pos;
DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
rli->relay_log.harvest_bytes_written(&rli->log_space_total);
}
else
error= 3;
rli->ign_master_log_name_end[0]= 0; // last event is not ignored
}
pthread_mutex_unlock(log_lock);
err:
pthread_mutex_unlock(&mi->data_lock);
DBUG_PRINT("info", ("error=%d", error));
DBUG_RETURN(error);
}
void end_relay_log_info(RELAY_LOG_INFO* rli)
{
DBUG_ENTER("end_relay_log_info");
if (!rli->inited)
DBUG_VOID_RETURN;
if (rli->info_fd >= 0)
{
end_io_cache(&rli->info_file);
(void) my_close(rli->info_fd, MYF(MY_WME));
rli->info_fd = -1;
}
if (rli->cur_log_fd >= 0)
{
end_io_cache(&rli->cache_buf);
(void)my_close(rli->cur_log_fd, MYF(MY_WME));
rli->cur_log_fd = -1;
}
rli->inited = 0;
rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
rli->relay_log.harvest_bytes_written(&rli->log_space_total);
/*
Delete the slave's temporary tables from memory.
In the future there will be other actions than this, to ensure persistance
of slave's temp tables after shutdown.
*/
rli->close_temporary_tables();
DBUG_VOID_RETURN;
}
/*
Try to connect until successful or slave killed
SYNPOSIS
safe_connect()
thd Thread handler for slave
mysql MySQL connection handle
mi Replication handle
RETURN
0 ok
# Error
*/
static int safe_connect(THD* thd, MYSQL* mysql, MASTER_INFO* mi)
{
DBUG_ENTER("safe_connect");
DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
}
/*
SYNPOSIS
connect_to_master()
IMPLEMENTATION
Try to connect until successful or slave killed or we have retried
master_retry_count times
*/
static int connect_to_master(THD* thd, MYSQL* mysql, MASTER_INFO* mi,
bool reconnect, bool suppress_warnings)
{
int slave_was_killed;
int last_errno= -2; // impossible error
ulong err_count=0;
char llbuff[22];
DBUG_ENTER("connect_to_master");
#ifndef DBUG_OFF
mi->events_till_disconnect = disconnect_slave_event_count;
#endif
ulong client_flag= CLIENT_REMEMBER_OPTIONS;
if (opt_slave_compressed_protocol)
client_flag=CLIENT_COMPRESS; /* We will use compression */
mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout);
mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout);
#ifdef HAVE_OPENSSL
if (mi->ssl)
mysql_ssl_set(mysql,
mi->ssl_key[0]?mi->ssl_key:0,
mi->ssl_cert[0]?mi->ssl_cert:0,
mi->ssl_ca[0]?mi->ssl_ca:0,
mi->ssl_capath[0]?mi->ssl_capath:0,
mi->ssl_cipher[0]?mi->ssl_cipher:0);
#endif
mysql_options(mysql, MYSQL_SET_CHARSET_NAME, default_charset_info->csname);
/* This one is not strictly needed but we have it here for completeness */
mysql_options(mysql, MYSQL_SET_CHARSET_DIR, (char *) charsets_dir);
while (!(slave_was_killed = io_slave_killed(thd,mi)) &&
(reconnect ? mysql_reconnect(mysql) != 0 :
mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0,
mi->port, 0, client_flag) == 0))
{
/* Don't repeat last error */
if ((int)mysql_errno(mysql) != last_errno)
{
last_errno=mysql_errno(mysql);
suppress_warnings= 0;
sql_print_error("Slave I/O thread: error %s to master "
"'%s@%s:%d': \
Error: '%s' errno: %d retry-time: %d retries: %lu",
(reconnect ? "reconnecting" : "connecting"),
mi->user, mi->host, mi->port,
mysql_error(mysql), last_errno,
mi->connect_retry,
master_retry_count);
}
/*
By default we try forever. The reason is that failure will trigger
master election, so if the user did not set master_retry_count we
do not want to have election triggered on the first failure to
connect
*/
if (++err_count == master_retry_count)
{
slave_was_killed=1;
if (reconnect)
change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
break;
}
safe_sleep(thd,mi->connect_retry,(CHECK_KILLED_FUNC)io_slave_killed,
(void*)mi);
}
if (!slave_was_killed)
{
if (reconnect)
{
if (!suppress_warnings && global_system_variables.log_warnings)
sql_print_information("Slave: connected to master '%s@%s:%d',\
replication resumed in log '%s' at position %s", mi->user,
mi->host, mi->port,
IO_RPL_LOG_NAME,
llstr(mi->master_log_pos,llbuff));
}
else
{
change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
mi->user, mi->host, mi->port);
}
#ifdef SIGNAL_WITH_VIO_CLOSE
thd->set_active_vio(mysql->net.vio);
#endif
}
mysql->reconnect= 1;
DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
DBUG_RETURN(slave_was_killed);
}
/*
safe_reconnect()
IMPLEMENTATION
Try to connect until successful or slave killed or we have retried
master_retry_count times
*/
static int safe_reconnect(THD* thd, MYSQL* mysql, MASTER_INFO* mi,
bool suppress_warnings)
{
DBUG_ENTER("safe_reconnect");
DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
}
/*
Store the file and position where the execute-slave thread are in the
relay log.
SYNOPSIS
flush_relay_log_info()
rli Relay log information
NOTES
- As this is only called by the slave thread, we don't need to
have a lock on this.
- If there is an active transaction, then we don't update the position
in the relay log. This is to ensure that we re-execute statements
if we die in the middle of an transaction that was rolled back.
- As a transaction never spans binary logs, we don't have to handle the
case where we do a relay-log-rotation in the middle of the transaction.
If this would not be the case, we would have to ensure that we
don't delete the relay log file where the transaction started when
we switch to a new relay log file.
TODO
- Change the log file information to a binary format to avoid calling
longlong2str.
RETURN VALUES
0 ok
1 write error
*/
bool flush_relay_log_info(RELAY_LOG_INFO* rli)
{
bool error=0;
DBUG_ENTER("flush_relay_log_info");
if (unlikely(rli->no_storage))
DBUG_RETURN(0);
IO_CACHE *file = &rli->info_file;
char buff[FN_REFLEN*2+22*2+4], *pos;
my_b_seek(file, 0L);
pos=strmov(buff, rli->group_relay_log_name);
*pos++='\n';
pos=longlong2str(rli->group_relay_log_pos, pos, 10);
*pos++='\n';
pos=strmov(pos, rli->group_master_log_name);
*pos++='\n';
pos=longlong2str(rli->group_master_log_pos, pos, 10);
*pos='\n';
if (my_b_write(file, (byte*) buff, (ulong) (pos-buff)+1))
error=1;
if (flush_io_cache(file))
error=1;
/* Flushing the relay log is done by the slave I/O thread */
DBUG_RETURN(error);
}
/*
Called when we notice that the current "hot" log got rotated under our feet.
*/
static IO_CACHE *reopen_relay_log(RELAY_LOG_INFO *rli, const char **errmsg)
{
DBUG_ENTER("reopen_relay_log");
DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
DBUG_ASSERT(rli->cur_log_fd == -1);
IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
errmsg)) <0)
DBUG_RETURN(0);
/*
We want to start exactly where we was before:
relay_log_pos Current log pos
pending Number of bytes already processed from the event
*/
rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
my_b_seek(cur_log,rli->event_relay_log_pos);
DBUG_RETURN(cur_log);
}
static Log_event* next_event(RELAY_LOG_INFO* rli)
{
Log_event* ev;
IO_CACHE* cur_log = rli->cur_log;
pthread_mutex_t *log_lock = rli->relay_log.get_log_lock();
const char* errmsg=0;
THD* thd = rli->sql_thd;
DBUG_ENTER("next_event");
DBUG_ASSERT(thd != 0);
#ifndef DBUG_OFF
if (abort_slave_event_count && !rli->events_till_abort--)
DBUG_RETURN(0);
#endif
/*
For most operations we need to protect rli members with data_lock,
so we assume calling function acquired this mutex for us and we will
hold it for the most of the loop below However, we will release it
whenever it is worth the hassle, and in the cases when we go into a
pthread_cond_wait() with the non-data_lock mutex
*/
safe_mutex_assert_owner(&rli->data_lock);
while (!sql_slave_killed(thd,rli))
{
/*
We can have two kinds of log reading:
hot_log:
rli->cur_log points at the IO_CACHE of relay_log, which
is actively being updated by the I/O thread. We need to be careful
in this case and make sure that we are not looking at a stale log that
has already been rotated. If it has been, we reopen the log.
The other case is much simpler:
We just have a read only log that nobody else will be updating.
*/
bool hot_log;
if ((hot_log = (cur_log != &rli->cache_buf)))
{
DBUG_ASSERT(rli->cur_log_fd == -1); // foreign descriptor
pthread_mutex_lock(log_lock);
/*
Reading xxx_file_id is safe because the log will only
be rotated when we hold relay_log.LOCK_log
*/
if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
{
// The master has switched to a new log file; Reopen the old log file
cur_log=reopen_relay_log(rli, &errmsg);
pthread_mutex_unlock(log_lock);
if (!cur_log) // No more log files
goto err;
hot_log=0; // Using old binary log
}
}
#ifndef DBUG_OFF
{
/* This is an assertion which sometimes fails, let's try to track it */
char llbuf1[22], llbuf2[22];
DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
llstr(my_b_tell(cur_log),llbuf1),
llstr(rli->event_relay_log_pos,llbuf2)));
DBUG_ASSERT(my_b_tell(cur_log) >= BIN_LOG_HEADER_SIZE);
DBUG_ASSERT(my_b_tell(cur_log) == rli->event_relay_log_pos);
}
#endif
/*
Relay log is always in new format - if the master is 3.23, the
I/O thread will convert the format for us.
A problem: the description event may be in a previous relay log. So if
the slave has been shutdown meanwhile, we would have to look in old relay
logs, which may even have been deleted. So we need to write this
description event at the beginning of the relay log.
When the relay log is created when the I/O thread starts, easy: the
master will send the description event and we will queue it.
But if the relay log is created by new_file(): then the solution is:
MYSQL_BIN_LOG::open() will write the buffered description event.
*/
if ((ev=Log_event::read_log_event(cur_log,0,
rli->relay_log.description_event_for_exec)))
{
DBUG_ASSERT(thd==rli->sql_thd);
/*
read it while we have a lock, to avoid a mutex lock in
inc_event_relay_log_pos()
*/
rli->future_event_relay_log_pos= my_b_tell(cur_log);
if (hot_log)
pthread_mutex_unlock(log_lock);
DBUG_RETURN(ev);
}
DBUG_ASSERT(thd==rli->sql_thd);
if (opt_reckless_slave) // For mysql-test
cur_log->error = 0;
if (cur_log->error < 0)
{
errmsg = "slave SQL thread aborted because of I/O error";
if (hot_log)
pthread_mutex_unlock(log_lock);
goto err;
}
if (!cur_log->error) /* EOF */
{
/*
On a hot log, EOF means that there are no more updates to
process and we must block until I/O thread adds some and
signals us to continue
*/
if (hot_log)
{
/*
We say in Seconds_Behind_Master that we have "caught up". Note that
for example if network link is broken but I/O slave thread hasn't
noticed it (slave_net_timeout not elapsed), then we'll say "caught
up" whereas we're not really caught up. Fixing that would require
internally cutting timeout in smaller pieces in network read, no
thanks. Another example: SQL has caught up on I/O, now I/O has read
a new event and is queuing it; the false "0" will exist until SQL
finishes executing the new event; it will be look abnormal only if
the events have old timestamps (then you get "many", 0, "many").
Transient phases like this can't really be fixed.
*/
time_t save_timestamp= rli->last_master_timestamp;
rli->last_master_timestamp= 0;
DBUG_ASSERT(rli->relay_log.get_open_count() ==
rli->cur_log_old_open_count);
if (rli->ign_master_log_name_end[0])
{
/* We generate and return a Rotate, to make our positions advance */
DBUG_PRINT("info",("seeing an ignored end segment"));
ev= new Rotate_log_event(rli->ign_master_log_name_end,
0, rli->ign_master_log_pos_end,
Rotate_log_event::DUP_NAME);
rli->ign_master_log_name_end[0]= 0;
pthread_mutex_unlock(log_lock);
if (unlikely(!ev))
{
errmsg= "Slave SQL thread failed to create a Rotate event "
"(out of memory?), SHOW SLAVE STATUS may be inaccurate";
goto err;
}
ev->server_id= 0; // don't be ignored by slave SQL thread
DBUG_RETURN(ev);
}
/*
We can, and should release data_lock while we are waiting for
update. If we do not, show slave status will block
*/
pthread_mutex_unlock(&rli->data_lock);
/*
Possible deadlock :
- the I/O thread has reached log_space_limit
- the SQL thread has read all relay logs, but cannot purge for some
reason:
* it has already purged all logs except the current one
* there are other logs than the current one but they're involved in
a transaction that finishes in the current one (or is not finished)
Solution :
Wake up the possibly waiting I/O thread, and set a boolean asking
the I/O thread to temporarily ignore the log_space_limit
constraint, because we do not want the I/O thread to block because of
space (it's ok if it blocks for any other reason (e.g. because the
master does not send anything). Then the I/O thread stops waiting
and reads more events.
The SQL thread decides when the I/O thread should take log_space_limit
into account again : ignore_log_space_limit is reset to 0
in purge_first_log (when the SQL thread purges the just-read relay
log), and also when the SQL thread starts. We should also reset
ignore_log_space_limit to 0 when the user does RESET SLAVE, but in
fact, no need as RESET SLAVE requires that the slave
be stopped, and the SQL thread sets ignore_log_space_limit to 0 when
it stops.
*/
pthread_mutex_lock(&rli->log_space_lock);
// prevent the I/O thread from blocking next times
rli->ignore_log_space_limit= 1;
/*
If the I/O thread is blocked, unblock it.
Ok to broadcast after unlock, because the mutex is only destroyed in
~st_relay_log_info(), i.e. when rli is destroyed, and rli will not be
destroyed before we exit the present function.
*/
pthread_mutex_unlock(&rli->log_space_lock);
pthread_cond_broadcast(&rli->log_space_cond);
// Note that wait_for_update unlocks lock_log !
rli->relay_log.wait_for_update(rli->sql_thd, 1);
// re-acquire data lock since we released it earlier
pthread_mutex_lock(&rli->data_lock);
rli->last_master_timestamp= save_timestamp;
continue;
}
/*
If the log was not hot, we need to move to the next log in
sequence. The next log could be hot or cold, we deal with both
cases separately after doing some common initialization
*/
end_io_cache(cur_log);
DBUG_ASSERT(rli->cur_log_fd >= 0);
my_close(rli->cur_log_fd, MYF(MY_WME));
rli->cur_log_fd = -1;
if (relay_log_purge)
{
/*
purge_first_log will properly set up relay log coordinates in rli.
If the group's coordinates are equal to the event's coordinates
(i.e. the relay log was not rotated in the middle of a group),
we can purge this relay log too.
We do ulonglong and string comparisons, this may be slow but
- purging the last relay log is nice (it can save 1GB of disk), so we
like to detect the case where we can do it, and given this,
- I see no better detection method
- purge_first_log is not called that often
*/
if (rli->relay_log.purge_first_log
(rli,
rli->group_relay_log_pos == rli->event_relay_log_pos
&& !strcmp(rli->group_relay_log_name,rli->event_relay_log_name)))
{
errmsg = "Error purging processed logs";
goto err;
}
}
else
{
/*
If hot_log is set, then we already have a lock on
LOCK_log. If not, we have to get the lock.
According to Sasha, the only time this code will ever be executed
is if we are recovering from a bug.
*/
if (rli->relay_log.find_next_log(&rli->linfo, !hot_log))
{
errmsg = "error switching to the next log";
goto err;
}
rli->event_relay_log_pos = BIN_LOG_HEADER_SIZE;
strmake(rli->event_relay_log_name,rli->linfo.log_file_name,
sizeof(rli->event_relay_log_name)-1);
flush_relay_log_info(rli);
}
/*
Now we want to open this next log. To know if it's a hot log (the one
being written by the I/O thread now) or a cold log, we can use
is_active(); if it is hot, we use the I/O cache; if it's cold we open
the file normally. But if is_active() reports that the log is hot, this
may change between the test and the consequence of the test. So we may
open the I/O cache whereas the log is now cold, which is nonsense.
To guard against this, we need to have LOCK_log.
*/
DBUG_PRINT("info",("hot_log: %d",hot_log));
if (!hot_log) /* if hot_log, we already have this mutex */
pthread_mutex_lock(log_lock);
if (rli->relay_log.is_active(rli->linfo.log_file_name))
{
#ifdef EXTRA_DEBUG
if (global_system_variables.log_warnings)
sql_print_information("next log '%s' is currently active",
rli->linfo.log_file_name);
#endif
rli->cur_log= cur_log= rli->relay_log.get_log_file();
rli->cur_log_old_open_count= rli->relay_log.get_open_count();
DBUG_ASSERT(rli->cur_log_fd == -1);
/*
Read pointer has to be at the start since we are the only
reader.
We must keep the LOCK_log to read the 4 first bytes, as this is a hot
log (same as when we call read_log_event() above: for a hot log we
take the mutex).
*/
if (check_binlog_magic(cur_log,&errmsg))
{
if (!hot_log) pthread_mutex_unlock(log_lock);
goto err;
}
if (!hot_log) pthread_mutex_unlock(log_lock);
continue;
}
if (!hot_log) pthread_mutex_unlock(log_lock);
/*
if we get here, the log was not hot, so we will have to open it
ourselves. We are sure that the log is still not hot now (a log can get
from hot to cold, but not from cold to hot). No need for LOCK_log.
*/
#ifdef EXTRA_DEBUG
if (global_system_variables.log_warnings)
sql_print_information("next log '%s' is not active",
rli->linfo.log_file_name);
#endif
// open_binlog() will check the magic header
if ((rli->cur_log_fd=open_binlog(cur_log,rli->linfo.log_file_name,
&errmsg)) <0)
goto err;
}
else
{
/*
Read failed with a non-EOF error.
TODO: come up with something better to handle this error
*/
if (hot_log)
pthread_mutex_unlock(log_lock);
sql_print_error("Slave SQL thread: I/O error reading \
event(errno: %d cur_log->error: %d)",
my_errno,cur_log->error);
// set read position to the beginning of the event
my_b_seek(cur_log,rli->event_relay_log_pos);
/* otherwise, we have had a partial read */
errmsg = "Aborting slave SQL thread because of partial event read";
break; // To end of function
}
}
if (!errmsg && global_system_variables.log_warnings)
{
sql_print_information("Error reading relay log event: %s",
"slave SQL thread was killed");
DBUG_RETURN(0);
}
err:
if (errmsg)
sql_print_error("Error reading relay log event: %s", errmsg);
DBUG_RETURN(0);
}
/*
Rotate a relay log (this is used only by FLUSH LOGS; the automatic rotation
because of size is simpler because when we do it we already have all relevant
locks; here we don't, so this function is mainly taking locks).
Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
is void).
*/
void rotate_relay_log(MASTER_INFO* mi)
{
DBUG_ENTER("rotate_relay_log");
RELAY_LOG_INFO* rli= &mi->rli;
/* We don't lock rli->run_lock. This would lead to deadlocks. */
pthread_mutex_lock(&mi->run_lock);
/*
We need to test inited because otherwise, new_file() will attempt to lock
LOCK_log, which may not be inited (if we're not a slave).
*/
if (!rli->inited)
{
DBUG_PRINT("info", ("rli->inited == 0"));
goto end;
}
/* If the relay log is closed, new_file() will do nothing. */
rli->relay_log.new_file();
/*
We harvest now, because otherwise BIN_LOG_HEADER_SIZE will not immediately
be counted, so imagine a succession of FLUSH LOGS and assume the slave
threads are started:
relay_log_space decreases by the size of the deleted relay log, but does
not increase, so flush-after-flush we may become negative, which is wrong.
Even if this will be corrected as soon as a query is replicated on the
slave (because the I/O thread will then call harvest_bytes_written() which
will harvest all these BIN_LOG_HEADER_SIZE we forgot), it may give strange
output in SHOW SLAVE STATUS meanwhile. So we harvest now.
If the log is closed, then this will just harvest the last writes, probably
0 as they probably have been harvested.
*/
rli->relay_log.harvest_bytes_written(&rli->log_space_total);
end:
pthread_mutex_unlock(&mi->run_lock);
DBUG_VOID_RETURN;
}
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
template class I_List_iterator<i_string>;
template class I_List_iterator<i_string_pair>;
#endif
#endif /* HAVE_REPLICATION */
Reference in a new issue View git blame Copy permalink