Why Does Semi‑Synchronous MySQL Replication Match Asynchronous Speed in Real Tests?

MySQL replication mode comparison: many claim semi‑synchronous is much slower than asynchronous, but current tests show the performance gap is tiny.

Performance Test Report

Results:

Semi‑synchronous replication: TPS 760.35, QPS 15207.04
Asynchronous replication: TPS 837.05, QPS 16740.90

Observations:

The difference is slight; semi‑synchronous is only a bit slower.

Why is the gap so small? Because only one slave is used and the master waits for a single acknowledgment (rpl_semi_sync_master_wait_for_slave_count = 1). In a one‑master‑two‑slave setup the gap would be larger.

During the test the slave had little load, so binlog ACKs were fast; heavier slave load would increase latency.

Which is better to use, semi‑synchronous or asynchronous replication?

Example deployment: in a one‑master‑three‑slave architecture you can set one slave to semi‑synchronous and the other two to asynchronous, balancing data safety and consistency.

Semi‑Synchronous Replication Stress Test

1. Build GTID semi‑synchronous replication

# Check GTID status on master
show variables like '%gtid%';
# Install semi‑sync plugin on master
install plugin rpl_semi_sync_master soname 'semisync_master.so';
set global rpl_semi_sync_master_enabled = 1;
# Install plugin on slave
install plugin rpl_semi_sync_slave soname 'semisync_slave.so';
set global rpl_semi_sync_slave_enabled = 1;
# Create replication user
grant replication slave on *.* to repl@'%' identified by '123456';
# Configure master
CHANGE MASTER TO MASTER_HOST='10.186.61.30', MASTER_USER='repl', MASTER_PASSWORD='123456', MASTER_PORT=3306, MASTER_CONNECT_RETRY=10, MASTER_AUTO_POSITION=1;
START SLAVE;

2. Pressure test semi‑synchronous mode

# Create test database
create database testdb;
grant all on *.* to root@'%' identified by '123456';
# Load data with sysbench (200k rows, 8 tables, 4 threads, 60 s)
sysbench /usr/share/sysbench/tests/include/oltp_legacy/oltp.lua --mysql-host=10.186.61.30 --mysql-port=3306 --mysql-user=root --mysql-password=123456 --mysql-db=testdb --oltp-table-size=200000 --oltp-tables-count=8 --threads=4 --time=60 --report-interval=3 run;

Sample sysbench output (selected lines):

[3s] thds: 4 tps: 599.32 qps: 15207.04 (r/w/o: 8401.09/2397.26/1199.96) lat (ms,95%): 11.04 err/s: 0.00 reconn/s: 0.00
[6s] thds: 4 tps: 740.06 qps: 14808.55 (r/w/o: 10364.19/2964.24/1480.12) lat (ms,95%): 8.58 err/s: 0.00 reconn/s: 0.00
[9s] thds: 4 tps: 761.70 qps: 15239.62 (r/w/o: 10668.43/3047.32/1523.39) lat (ms,95%): 8.74 err/s: 0.00 reconn/s: 0.00
[12s] thds: 4 tps: 766.66 qps: 15327.62 (r/w/o: 10730.97/3063.32/1533.33) lat (ms,95%): 8.43 err/s: 0.00 reconn/s: 0.00
[15s] thds: 4 tps: 770.31 qps: 15407.89 (r/w/o: 10784.03/3083.25/1540.62) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00
[18s] thds: 4 tps: 775.37 qps: 15504.64 (r/w/o: 10854.78/3099.13/1550.73) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00
[21s] thds: 4 tps: 796.65 qps: 15931.60 (r/w/o: 11151.39/3187.25/1592.96) lat (ms,95%): 7.98 err/s: 0.00 reconn/s: 0.00
[24s] thds: 4 tps: 766.31 qps: 15325.15 (r/w/o: 10726.31/3065.90/1532.95) lat (ms,95%): 8.43 err/s: 0.00 reconn/s: 0.00
[27s] thds: 4 tps: 750.69 qps: 15016.21 (r/w/o: 10512.05/3002.78/1501.39) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00
[30s] thds: 4 tps: 767.35 qps: 15349.98 (r/w/o: 10745.89/3069.40/1534.70) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00
[33s] thds: 4 tps: 744.96 qps: 14894.10 (r/w/o: 10425.70/2978.49/1489.91) lat (ms,95%): 9.39 err/s: 0.00 reconn/s: 0.00
[36s] thds: 4 tps: 752.64 qps: 15050.90 (r/w/o: 10535.03/3010.58/1505.29) lat (ms,95%): 8.13 err/s: 0.00 reconn/s: 0.00
[39s] thds: 4 tps: 786.04 qps: 15721.48 (r/w/o: 11006.57/3142.83/1572.08) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00
[42s] thds: 4 tps: 782.34 qps: 15649.87 (r/w/o: 10955.81/3129.37/1564.69) lat (ms,95%): 8.43 err/s: 0.00 reconn/s: 0.00
[45s] thds: 4 tps: 790.66 qps: 15810.29 (r/w/o: 11063.97/3164.99/1581.33) lat (ms,95%): 8.13 err/s: 0.00 reconn/s: 0.00
[48s] thds: 4 tps: 741.67 qps: 14829.09 (r/w/o: 10381.40/2964.35/1483.34) lat (ms,95%): 9.06 err/s: 0.00 reconn/s: 0.00
[51s] thds: 4 tps: 773.67 qps: 15474.64 (r/w/o: 10831.31/3095.99/1547.33) lat (ms,95%): 8.74 err/s: 0.00 reconn/s: 0.00
[54s] thds: 4 tps: 778.59 qps: 15580.14 (r/w/o: 10908.26/3114.69/1557.18) lat (ms,95%): 8.13 err/s: 0.00 reconn/s: 0.00
[57s] thds: 4 tps: 801.73 qps: 16031.28 (r/w/o: 11219.90/3207.92/1603.46) lat (ms,95%): 8.13 err/s: 0.00 reconn/s: 0.00
[60s] thds: 4 tps: 761.29 qps: 15228.56 (r/w/o: 10660.79/3045.18/1522.59) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00

Asynchronous Replication Stress Test

1. Switch to GTID asynchronous replication

# Disable semi‑sync on master
set global rpl_semi_sync_master_enabled = 0;
# Disable on slave
set global rpl_semi_sync_slave_enabled = 0;
STOP SLAVE IO_THREAD;
START SLAVE IO_THREAD;

2. Pressure test asynchronous mode

sysbench /usr/share/sysbench/tests/include/oltp_legacy/oltp.lua --mysql-host=10.186.61.30 --mysql-port=3306 --mysql-user=root --mysql-password=123456 --mysql-db=testdb --oltp-table-size=200000 --oltp-tables-count=8 --threads=4 --time=60 --report-interval=3 run;

Sample async sysbench output (selected lines):

[3s] thds: 4 tps: 805.14 qps: 16122.80 (r/w/o: 11286.63/3224.56/1611.61) lat (ms,95%): 7.70 err/s: 0.00 reconn/s: 0.00
[6s] thds: 4 tps: 891.19 qps: 17819.06 (r/w/o: 12475.28/3561.41/1782.37) lat (ms,95%): 7.43 err/s: 0.00 reconn/s: 0.00
[9s] thds: 4 tps: 838.69 qps: 16775.78 (r/w/o: 11742.98/3355.42/1677.38) lat (ms,95%): 7.98 err/s: 0.00 reconn/s: 0.00
[12s] thds: 4 tps: 855.24 qps: 17100.48 (r/w/o: 11969.37/3420.63/1710.48) lat (ms,95%): 7.84 err/s: 0.00 reconn/s: 0.00
[15s] thds: 4 tps: 856.08 qps: 17122.62 (r/w/o: 11987.13/3423.32/1712.16) lat (ms,95%): 8.13 err/s: 0.00 reconn/s: 0.00
[18s] thds: 4 tps: 834.35 qps: 16683.96 (r/w/o: 11678.20/3337.39/1668.36) lat (ms,95%): 8.13 err/s: 0.00 reconn/s: 0.00
[21s] thds: 4 tps: 836.93 qps: 16733.92 (r/w/o: 11712.34/3347.72/1673.86) lat (ms,95%): 7.98 err/s: 0.00 reconn/s: 0.00
[24s] thds: 4 tps: 849.73 qps: 17006.66 (r/w/o: 11906.93/3399.93/1699.80) lat (ms,95%): 8.13 err/s: 0.00 reconn/s: 0.00
[27s] thds: 4 tps: 847.67 qps: 16953.77 (r/w/o: 11865.41/3393.02/1695.34) lat (ms,95%): 7.43 err/s: 0.00 reconn/s: 0.00
[30s] thds: 4 tps: 841.00 qps: 16808.31 (r/w/o: 11765.98/3360.66/1681.66) lat (ms,95%): 8.58 err/s: 0.00 reconn/s: 0.00
[33s] thds: 4 tps: 811.29 qps: 16238.80 (r/w/o: 11366.73/3249.16/1622.91) lat (ms,95%): 8.58 err/s: 0.00 reconn/s: 0.00
[36s] thds: 4 tps: 809.04 qps: 16175.85 (r/w/o: 11324.93/3232.84/1618.08) lat (ms,95%): 7.84 err/s: 0.00 reconn/s: 0.00
[39s] thds: 4 tps: 843.62 qps: 16870.73 (r/w/o: 11809.68/3373.81/1687.24) lat (ms,95%): 7.98 err/s: 0.00 reconn/s: 0.00
[42s] thds: 4 tps: 841.00 qps: 16826.31 (r/w/o: 11778.65/3365.66/1682.00) lat (ms,95%): 7.56 err/s: 0.00 reconn/s: 0.00
[45s] thds: 4 tps: 820.69 qps: 16411.52 (r/w/o: 11487.36/3282.77/1641.39) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00
[48s] thds: 4 tps: 830.99 qps: 16623.45 (r/w/o: 11635.18/3326.29/1661.98) lat (ms,95%): 7.98 err/s: 0.00 reconn/s: 0.00
[51s] thds: 4 tps: 838.67 qps: 16763.33 (r/w/o: 11734.33/3352.00/1677.00) lat (ms,95%): 7.70 err/s: 0.00 reconn/s: 0.00
[54s] thds: 4 tps: 822.67 qps: 16456.73 (r/w/o: 11521.37/3289.68/1645.67) lat (ms,95%): 8.28 err/s: 0.00 reconn/s: 0.00
[57s] thds: 4 tps: 834.70 qps: 16696.98 (r/w/o: 11686.78/3340.80/1669.40) lat (ms,95%): 7.98 err/s: 0.00 reconn/s: 0.00
[60s] thds: 4 tps: 834.58 qps: 16684.62 (r/w/o: 11679.47/3335.99/1669.16) lat (ms,95%): 7.98 err/s: 0.00 reconn/s: 0.00

Both modes achieve similar throughput; the semi‑synchronous mode incurs only a modest overhead, making it a viable option when stronger data consistency is required.