Commit 73adaadb authored by unknown's avatar unknown

mysql-test/mysql-test-run

    cleanup - remove unneeded daeamon restarts
sql/log_event.cc
    fixed bug - do not confuse confused slaves even further if they try to 
    read less than LOG_EVENT_HEADER_LEN before the end of file
sql/mysqlbinlog.cc
    give error if constructor fails
sql/mysqld.cc
    --disconnect-slave-event-count for replication debugging
sql/slave.cc
    --disconnect-slave-event-count
    cleanup to remove unneeded sleeps when things go wrong - sleep only 
    when we have to
sql/slave.h
    --disconnect-slave-event-count

test cases for ALTER TABLE ORDER BY and replication offset off by 22 
confusion mystery bug (both pass)


mysql-test/mysql-test-run:
  cleanup - remove unneeded daeamon restarts
sql/log_event.cc:
  fixed bug - do not confuse confused slaves even further if they try to 
  read less than LOG_EVENT_HEADER_LEN before the end of file
sql/mysqlbinlog.cc:
  give error if constructor fails
sql/mysqld.cc:
  --disconnect-slave-event-count for replication debugging
sql/slave.cc:
  --disconnect-slave-event-count
  cleanup to remove unneeded sleeps when things go wrong - sleep only 
  when we have to
sql/slave.h:
  --disconnect-slave-event-count
sql/sql_repl.cc:
  added unknown error
parent 350c792d
......@@ -278,6 +278,7 @@ start_master()
start_slave()
{
[ x$SKIP_SLAVE = x1 ] && return
[ -d $GCOV_SLAVE_SRC ] && cd $GCOV_SLAVE_SRC
slave_args="--no-defaults --server-id=2 \
--master-user=root \
......@@ -362,7 +363,7 @@ run_testcase ()
tname=`$ECHO $tname | $CUT -d . -f 1`
master_opt_file=$TESTDIR/$tname-master.opt
slave_opt_file=$TESTDIR/$tname-slave.opt
SKIP_SLAVE=`$EXPR \( match $tname rpl \) = 0`
if [ -f $master_opt_file ] ;
then
......@@ -443,7 +444,9 @@ run_testcase ()
mysql_install_db
if [ -z $DO_GDB ]
#do not automagically start deamons if we are in gdb or running only one test
#case
if [ -z $DO_GDB ] && [ -z $1 ]
then
$SETCOLOR_NORMAL && $ECHO -n "Starting mysqld for Testing"
mysql_start
......
use test;
drop table if exists x;
create table x (n int);
insert into x values(9),(3),(12),(10);
alter table x order by n;
@r/3.23/alt000001.result select * from x;
--disconnect-slave-event-count=1
#this tests the offset off by 22 mystery bug
#must run slave with --disconnect-slave-event-count=1 --master-connect-retry=1
source t/include/master-slave.inc;
connection slave;
drop table if exists foo;
connection master;
drop table if exists foo;
create table foo (n int not null auto_increment primary key);
insert into foo values(NULL);
insert into foo values(2);
connection slave;
sleep 5;
@r/3.23/rpl000010.result select n from foo;
......@@ -81,7 +81,8 @@ int Log_event::read_log_event(IO_CACHE* file, String* packet,
// if the read hits eof, we must report it as eof
// so the caller will know it can go into cond_wait to be woken up
// on the next update to the log
return file->error >= 0 ? LOG_READ_EOF: LOG_READ_IO;
if(!file->error) return LOG_READ_EOF;
return file->error > 0 ? LOG_READ_TRUNC: LOG_READ_IO;
}
data_len = uint4korr(buf + EVENT_LEN_OFFSET);
if (data_len < LOG_EVENT_HEADER_LEN || data_len > MAX_EVENT_LEN)
......
......@@ -361,6 +361,8 @@ static void dump_local_log_entries(const char* logname)
die("Could not read entry at offset %ld : Error in log format or \
read error",
my_b_tell(file));
else
die("Could not construct event object");
break;
}
if (rec_count >= offset)
......
......@@ -2231,7 +2231,8 @@ enum options {
OPT_REPLICATE_REWRITE_DB, OPT_SERVER_ID, OPT_SKIP_SLAVE_START,
OPT_SKIP_INNOBASE,OPT_SAFEMALLOC_MEM_LIMIT,
OPT_REPLICATE_DO_TABLE, OPT_REPLICATE_IGNORE_TABLE,
OPT_REPLICATE_WILD_DO_TABLE, OPT_REPLICATE_WILD_IGNORE_TABLE
OPT_REPLICATE_WILD_DO_TABLE, OPT_REPLICATE_WILD_IGNORE_TABLE,
OPT_DISCONNECT_SLAVE_EVENT_COUNT
};
static struct option long_options[] = {
......@@ -2287,6 +2288,10 @@ static struct option long_options[] = {
{"master-info-file", required_argument, 0, (int) OPT_MASTER_INFO_FILE},
{"myisam-recover", optional_argument, 0, (int) OPT_MYISAM_RECOVER},
{"memlock", no_argument, 0, (int) OPT_MEMLOCK},
// needs to be available for the test case to pass in non-debugging mode
// is a no-op
{"disconnect-slave-event-count", required_argument, 0,
(int) OPT_DISCONNECT_SLAVE_EVENT_COUNT},
#if !defined(DBUG_OFF) && defined(SAFEMALLOC)
{"safemalloc-mem-limit", required_argument, 0, (int)
OPT_SAFEMALLOC_MEM_LIMIT},
......@@ -2888,6 +2893,12 @@ static void get_options(int argc,char **argv)
if (optarg && optarg[0])
opt_bin_logname=my_strdup(optarg,MYF(0));
break;
// needs to be handled (as no-op) in non-debugging mode for test suite
case (int)OPT_DISCONNECT_SLAVE_EVENT_COUNT:
#ifndef DBUG_OFF
disconnect_slave_event_count = atoi(optarg);
#endif
break;
case (int) OPT_LOG_SLAVE_UPDATES:
opt_log_slave_updates = 1;
break;
......
......@@ -30,7 +30,11 @@ DYNAMIC_ARRAY replicate_wild_do_table, replicate_wild_ignore_table;
bool do_table_inited = 0, ignore_table_inited = 0;
bool wild_do_table_inited = 0, wild_ignore_table_inited = 0;
bool table_rules_on = 0;
#ifndef DBUG_OFF
int disconnect_slave_event_count = 0;
static int events_till_disconnect = -1;
static int stuck_count = 0;
#endif
static inline void skip_load_data_infile(NET* net);
......@@ -668,6 +672,11 @@ static uint read_event(MYSQL* mysql, MASTER_INFO *mi)
// being in the interrupted state :-)
// 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 && !(events_till_disconnect--))
return packet_error;
#endif
while (!abort_loop && !abort_slave && len == packet_error && read_errno == EINTR )
{
len = mc_net_safe_read(mysql);
......@@ -970,6 +979,8 @@ pthread_handler_decl(handle_slave,arg __attribute__((unused)))
pthread_mutex_unlock(&LOCK_slave);
int error = 1;
bool retried_once = 0;
ulonglong last_failed_pos = 0;
my_thread_init(); // needs to be up here, otherwise we get a coredump
// trying to use DBUG_ stuff
......@@ -1009,12 +1020,20 @@ pthread_handler_decl(handle_slave,arg __attribute__((unused)))
thd->proc_info = "waiting to reconnect after a failed dump request";
if(mysql->net.vio)
vio_close(mysql->net.vio);
// 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(retried_once)
safe_sleep(thd, glob_mi.connect_retry);
else
retried_once = 1;
if(slave_killed(thd))
goto err;
thd->proc_info = "reconnecting after a failed dump request";
sql_print_error("Slave: failed dump request, reconnecting to \
try again, master_log_pos=%ld", last_failed_pos = glob_mi.pos );
safe_reconnect(thd, mysql, &glob_mi);
if(slave_killed(thd))
goto err;
......@@ -1025,7 +1044,6 @@ pthread_handler_decl(handle_slave,arg __attribute__((unused)))
while(!slave_killed(thd))
{
bool reset = 0;
thd->proc_info = "reading master update";
uint event_len = read_event(mysql, &glob_mi);
if(slave_killed(thd))
......@@ -1036,18 +1054,21 @@ pthread_handler_decl(handle_slave,arg __attribute__((unused)))
thd->proc_info = "waiting to reconnect after a failed read";
if(mysql->net.vio)
vio_close(mysql->net.vio);
if(retried_once) // punish repeat offender with sleep
safe_sleep(thd, glob_mi.connect_retry);
else
retried_once = 1;
if(slave_killed(thd))
goto err;
thd->proc_info = "reconnecting after a failed read";
sql_print_error("Slave: Failed reading log event, \
reconnecting to retry, master_log_pos=%ld", last_failed_pos = glob_mi.pos);
safe_reconnect(thd, mysql, &glob_mi);
if(slave_killed(thd))
goto err;
reset = 1;
}
if(reset)
break;
}
thd->proc_info = "processing master log event";
if(exec_event(thd, &mysql->net, &glob_mi, event_len))
......@@ -1060,6 +1081,28 @@ pthread_handler_decl(handle_slave,arg __attribute__((unused)))
// should restart the slave with mysqladmin start-slave
}
// successful exec with offset advance,
// the slave repents and his sins are forgiven!
if(glob_mi.pos > last_failed_pos)
{
retried_once = 0;
#ifndef DBUG_OFF
stuck_count = 0;
#endif
}
#ifndef DBUG_OFF
else
{
stuck_count++;
// show a little mercy, allow slave to read one more event
// before cutting him off - otherwise he gets stuck
// on Invar events, since they do not advance the offset
// immediately
if(stuck_count > 2)
events_till_disconnect++;
}
#endif
}
}
......@@ -1086,6 +1129,9 @@ pthread_handler_decl(handle_slave,arg __attribute__((unused)))
static void safe_connect(THD* thd, MYSQL* mysql, MASTER_INFO* mi)
// will try to connect until successful
{
#ifndef DBUG_OFF
events_till_disconnect = disconnect_slave_event_count;
#endif
while(!slave_killed(thd) &&
!mc_mysql_connect(mysql, mi->host, mi->user, mi->password, 0,
mi->port, 0, 0))
......@@ -1107,10 +1153,13 @@ static void safe_reconnect(THD* thd, MYSQL* mysql, MASTER_INFO* mi)
{
mi->pending = 0; // if we lost connection after reading a state set event
// we will be re-reading it, so pending needs to be cleared
#ifndef DBUG_OFF
events_till_disconnect = disconnect_slave_event_count;
#endif
while(!slave_killed(thd) && mc_mysql_reconnect(mysql))
{
sql_print_error(
"Slave thread: error connecting to master:%s, retry in %d sec",
sql_print_error("Slave thread: error connecting to master:\
%s, retry in %d sec",
mc_mysql_error(mysql), mi->connect_retry);
safe_sleep(thd, mi->connect_retry);
}
......
......@@ -94,6 +94,10 @@ extern bool do_table_inited, ignore_table_inited,
wild_do_table_inited, wild_ignore_table_inited;
extern bool table_rules_on;
#ifndef DBUG_OFF
extern int disconnect_slave_event_count ;
#endif
// the master variables are defaults read from my.cnf or command line
extern uint master_port, master_connect_retry;
extern my_string master_user, master_password, master_host,
......
......@@ -309,6 +309,9 @@ sweepstakes if you report the bug";
case LOG_READ_TRUNC:
errmsg = "binlog truncated in the middle of event";
break;
default:
errmsg = "unknown error reading log event on the master";
break;
}
goto err;
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment