PolarDB:Sysbench tests

PolarDB-X instance specifications

Benchmark executor

Test dataset

Test scenarios

All tests use the standard Sysbench OLTP workloads: point_select, read_only, read_write, update_index, update_non_index, and write_only. Results are measured in QPS (queries per second).

Step 1: Create an ECS instance

Create an ECS instance with 32 vCPUs and 128 GB of memory. This capacity prevents the benchmark executor from becoming a bottleneck when testing high-specification PolarDB-X instances.

Deploy the ECS instance in a virtual private cloud (VPC). Record the VPC name and ID. All database instances created in subsequent steps must reside in the same VPC.

Step 2: Create a PolarDB-X instance

1. Create a PolarDB-X instance with the MySQL 5.7 or MySQL 8.0 engine. See Create a PolarDB-X instance.

2. Create a database named sbtest in the instance. See Create a database.

   CREATE DATABASE sbtest MODE = 'auto';

Step 3: Configure instance parameters

Modify these parameters on the compute nodes to achieve optimal stress test performance.

1. Set the following parameters via the console. See Parameter settings.

   Parameter	Value
   ENABLE_COROUTINE	true
   XPROTO_MAX_DN_CONCURRENT	4000
   XPROTO_MAX_DN_WAIT_CONNECTION	4000

2. Connect to the instance with a command-line client and run the following SQL statements in the same session to disable logging and CPU statistical sampling:

   set global RECORD_SQL=false;
   set global MPP_METRIC_LEVEL=0;
   set global ENABLE_CPU_PROFILE=false;
   set global ENABLE_TRANS_LOG=false;

Step 4: Install Sysbench

1. Download and extract the Sysbench package:

   tar xzvf sysbench.tar.gz
   cd sysbench/

   Download link: sysbench.tar.gz Alternatively, build from source at sysbench on GitHub.

2. Install dependencies, then compile and install:

   yum -y install make automake libtool pkgconfig libaio-devel mysql-devel
   ./autogen.sh
   ./configure
   make -j
   make install

3. Verify the installation:

   sysbench --version

   Expected output: sysbench 1.1.0

Step 5: Configure Sysbench

Create a configuration file named sysb.conf with your PolarDB-X connection details:

mysql-host=<HOST>
mysql-port=<PORT>
mysql-user=<USER>
mysql-password=<PASSWORD>
mysql-db=sbtest
db-driver=mysql
percentile=95
histogram=on
report-interval=1
time=60
rand-type=uniform

Replace the connection placeholders with your actual values:

The remaining parameters control test behavior:

Step 6: Load test data

sysbench --config-file=sysb.conf \
  --create-table-options='single locality="balance_single_table=on"' \
  --tables=16 \
  --table-size=10000000 \
  --threads=16 \
  --auto_inc=off \
  oltp_point_select prepare

Step 7: Run the tests

Use the following command template. Replace <WORKLOAD> with the scenario name and <THREADS> with the number of concurrent threads.

sysbench --config-file=sysb.conf \
  --db-ps-mode=disable \
  --mysql-ignore-errors=all \
  --tables=16 \
  --table-size=10000000 \
  --threads=<THREADS> \
  <WORKLOAD> run

For read_only and read_write, add --range-size=5:

sysbench --config-file=sysb.conf \
  --db-ps-mode=disable \
  --mysql-ignore-errors=all \
  --tables=16 \
  --table-size=10000000 \
  --range-size=5 \
  --threads=<THREADS> \
  <WORKLOAD> run

Supported workloads: oltp_point_select, oltp_read_only, oltp_read_write, oltp_update_index, oltp_update_non_index, oltp_write_only.

Run the first stress test with an extended duration to let the database reach a stable state before recording results.

Performance is lower than expected

If your test results are significantly lower than the reference QPS values in this document, check the following:

• Parameter tuning not applied: Confirm that ENABLE_COROUTINE, XPROTO_MAX_DN_CONCURRENT, and XPROTO_MAX_DN_WAIT_CONNECTION are set to the values in Step 3. Run the set global SQL statements in the same session after connecting.

• ECS bottleneck: If your ECS instance does not meet the recommended specification of 32 vCPUs and 128 GB of memory, the benchmark executor itself may be the limiting factor, not the database.

• Database not warmed up: Run the first stress test with an extended duration to let the database reach a stable state. Results from a cold start are not representative.

• Network latency: Confirm that the ECS instance and the PolarDB-X instance are in the same VPC.

The sysbench --version command returns an error

The version check command is sysbench --version, not synben --version. If the command is not found, the installation did not complete successfully. Rerun the dependency installation and compilation steps.