1、背景
测试mysql5.7和mysql8.0 分别在读写、只读、只写模式下不同并发时的性能(tps,qps)
2、前提
- 测试使用版本为mysql5.7.22和mysql8.0.15
- sysbench测试前先重启mysql服务,并清空os的cache(避免多次测试时命中缓存)
- 每次进行测试都是新生成测试数据后再进行mysql5.7和mysql8.0的测试
每次测试时保证mysql5.7和mysql8.0的配置参数一致
3、环境
机器
cat /etc/redhat-release | xargs echo '版本 ' && dmidecode -s system-product-name | xargs echo '是否虚拟化 ' && cat /proc/cpuinfo |grep "processor"|wc -l | xargs echo 'cpu核数 '
版本 CentOS Linux release 7.5.1804 (Core) 是否虚拟化 KVM cpu核数 4
myql5.7.22
5.7.22-log
innodb_buffer_pool_size 128M
innodb_log_buffer_size 64M
innodb_log_file_size 48M
binlog_format ROW
log_bin ON
transaction_isolation REPEATABLE-READ
mysql8.0.15
8.0.15
innodb_buffer_pool_size 128M
innodb_log_buffer_size 64M
innodb_log_file_size 48M
binlog_format ROW
log_bin ON
transaction_isolation REPEATABLE-READ
sysbench
sysbench -V
sysbench 1.1.0 (using bundled LuaJIT 2.1.0-beta3)
4、测试
在不同的持久化策略下(
binlog
,redo log持久化)mysql5.7和mysql8.0 在读写模式、只读模式、只写模式(oltp_read_write
,oltp_read_only
,oltp_write_only
)下的性能表现- sysbench 测试时间为60s,测试的表数量为20
- 测试分别在双1模式(安全性)和0 2模式(高性能)下进行
| 参数 | 可选值 | 含义 |
| —- | —- | —- |
|
sync_binlog
| 0 | binlog刷盘持久化由操作系统完成,性能好,存在丢失binlog的风险 | |sync_binlog
| 1 | 事务提交后刷盘持久化,最安全 | |sync_binlog
| N | 在每N个事务提交后进行刷盘持久化 | |innodb_flush_log_at_trx_commit
| 0 | 每秒钟写redo log 并刷盘持久化 | |innodb_flush_log_at_trx_commit
| 1 | 事务提交后写redo log 并刷盘持久化,最安全 | |innodb_flush_log_at_trx_commit
| 2 | 事务提交后写redo log , 每秒刷盘持久化 |
双1模式下
SHOW GLOBAL VARIABLES WHERE Variable_name IN('sync_binlog','innodb_flush_log_at_trx_commit');
+--------------------------------+-------+
| Variable_name | Value |
+--------------------------------+-------+
| innodb_flush_log_at_trx_commit | 1 |
| sync_binlog | 1 |
+--------------------------------+-------+
mysql5.7和mysql8.0 在读写模式下的表现
双1 配置,读写模式下,mysql5.7.22 和mysql8.0.15 tps 、qps 性能差不多,mysql8.0.15 在120 线程并发时,性能出现了下降抖动
mysql5.7和mysql8.0 在只读模式下的表现
双1 配置,只读模式下,mysql5.7.22 的tps、qps比mysql8.0.15 好1/3 左右;并发线程数增加后,tps、qps并没有随着增加,反而出现了下降的趋势
mysql5.7和mysql8.0 在只写模式下的表现
双1 配置,只写模式下,随着并发数的上升,mysql5.7.22 的性能比mysql8.0.15 好1/4左右
0 2 模式下
SHOW GLOBAL VARIABLES WHERE Variable_name IN('sync_binlog','innodb_flush_log_at_trx_commit');
+--------------------------------+-------+
| Variable_name | Value |
+--------------------------------+-------+
| innodb_flush_log_at_trx_commit | 2 |
| sync_binlog | 0 |
+--------------------------------+-------+
mysql5.7和mysql8.0 在读写模式下的表现
0 2配置,读写模式下,并发数低时,mysql5.7.22性能好于mysql8.0.15; 并发数比较高时,mysql8.0.15 性能好于mysql5.7.22;在80 线程的并发以上时,性能开始下降
mysql5.7和mysql8.0 在只读模式下的表现
0 2配置,只读模式下,mysql5.7.22性能比mysql8.0.15 好1/3左右;随着并发数的上升,性能也没有上升,反而有下降的趋势
mysql5.7和mysql8.0 在只写模式下的表现
0 2 配置,只写模式下,mysql5.7.22的tps 抖动比较大;mysql5.7.22 的qps比mysql8.0.15好1/3左右
5、结论
整体来看,mysql5.7.22在读写模式、只读模式、只写模式下的表现是优于mysql8.0.15的
- 随着并行数的增加,性能表现不会也跟着增加,还会出现下降
- 本次测试结果是在配置很低的情况下进行的,不代表绝对
:::tips
注意sysbench 需要设置
--db-ps-mode=disable
禁用预编译语句,不然并发测试线程多时会报下面的错误 FATAL: mysql_stmt_prepare() failed FATAL: MySQL error: 1461 “Can’t create more than max_prepared_stmt_count statements (current value: 16382)” FATAL: mysql_stmt_prepare() failed FATAL: MySQL error: 1461 “Can’t create more than max_prepared_stmt_count statements (current value: 16382)” FATAL: thread_init’ function failed: /usr/local/share/sysbench/oltp_common.lua:288: SQL API error FATAL: mysql_stmt_prepare() failed FATAL: MySQL error: 1461 “Can’t create more than max_prepared_stmt_count statements (current value: 16382)” FATAL:thread_init’ function failed: /usr/local/share/sysbench/oltp_common.lua:288: SQL API error FATAL: mysql_stmt_prepare() failed ::: 使用脚本 ```bash cat sysbench_test_mysql5.7_8.0_tps_qps.sh!/bin/bash
用于sysbench 测试在读写模式、只读模式、只写模式下 mysql5.7和mysql8.0 的tps,qps
nohup bash $0 >/tmp/sysbench_test 2>& 1 &
#
user=admin passwd=admin ports=”8015 57222” host=127.0.0.1 sysbench_test_mode=”oltp_read_write oltp_read_only oltp_write_only” sysbench_test_info_path=/tmp/sysbench-test
function red_echo () {
local what="$*"
echo -e "$(date +%F-%T) \e[1;31m ${what} \e[0m"
}
function check_las_comm(){ if [ $1 -ne 0 ];then red_echo $2 exit 1 fi }
function restart_mysqld(){ service mysqld${1} restart sleep 2 }
function purge_binlog(){ port=$1 mysql -u$user -p$passwd -P$port -h$host<<EOF purge binary logs before now(); EOF }
function clean_os_cache(){ echo 3 > /proc/sys/vm/drop_caches }
function sysbench_with_diff_thread(){
thread_num=$1 port=$2 order=$3 test_mode=$4 sysbench /usr/local/share/sysbench/${test_mode}.lua —mysql_storage_engine=innodb —table-size=100000 —tables=20 —mysql-db=test_1 —mysql-user=$user —mysql-password=$passwd —mysql-port=$port —mysql-host=$host —threads=$thread_num —time=60 —report-interval=2 —db-ps-mode=disable —events=0 —db-driver=mysql $order
}
function main(){ for test_mode in $sysbench_test_mode;do
for port in $ports;do for thread_num in {5,10,20,30,40,80,120,200};do restart_mysqld “$port” check_las_comm “$?” “restart mysqld${port} failed “ clean_os_cache purge_binlog “$port”
red_echo "sysbench $thread_num threads cleanup mysqld${port}"
sysbench_with_diff_thread "$thread_num" "$port" "cleanup" "$test_mode">/dev/null
red_echo "sysbench $thread_num threads prepare mysqld${port}"
sysbench_with_diff_thread "$thread_num" "$port" "prepare" "$test_mode">/dev/null
mkdir -p $sysbench_test_info_path
red_echo "sysbench $thread_num threads run mysqld${port} $test_mode"
sysbench_with_diff_thread "$thread_num" "$port" "run" "$test_mode" > $sysbench_test_info_path/${test_mode}_${thread_num}_$port
# service mysqld{port} stop
done
done done
}
main ```