Xpand-Locust
This is a reference implementing of custom locust client for Xpand/MariaDB performance testing. It includes (but not limited to):
- MySQL getenv friendly framework
- Simplified locusfile creation
- Load balancer (during connection time) for Xpand
- Ability to redefine functions weights from configuration file
Standalone install
If you want to run locust standalone or master - workers inside the same host please follow next steps:
# First we need to install git
curl -s -LJO https://raw.githubusercontent.com/mariadb-corporation/xpand-locust/main/bin/git_setup.sh | bash -s
# Checkout xpand-locust
mkdir -p $HOME/tools && cd $HOME/tools && rm -rf xpand-locust && git clone https://github.com/mariadb-corporation/xpand-locust.git
$HOME/tools/xpand-locust/bin/pyenv_setup.sh
# You have to re-login before continue
# Install python
$HOME/tools/xpand-locust/bin/python3_setup.sh
pip install -r $HOME/tools/xpand-locust/requirements.txt
echo 'export XPAND_LOCUST_HOME="$HOME/tools/xpand-locust"' >>~/.bashrcBefore running your locustfile please setup PATH and PYTHONPATH as follow:
export XPAND_LOCUST_HOME=$HOME/tools/xpand-locust
export PYTHONPATH=$XPAND_LOCUST_HOME
export PATH=$XPAND_LOCUST_HOME/bin:$PATHQuick start
Setup you sample database using sales.sql and update params.yaml accordantly.
If you would like to rely on default transaction management, you very fist example can be as simple as this (make sure you you have autocommit: True in params.yaml):
@task(1)
def count_by_product(self):
_ = self.client.query_all(
"select count(*) from orders where product_name=%s",
(next(self.products_iterator),),
)time for your query will be measured and posted in the final stats as:
Name # reqs # fails | Avg Min Max Median | req/s failures/s
----------------------------------------------------------------------------------------------------------------------------------------------------------------
CUSTOM count_by_product 16 0(0.00%) | 19 15 25 17 | 0.72 0.00
CUSTOM insert_order 55 0(0.00%) | 19 15 33 18 | 2.48 0.00
----------------------------------------------------------------------------------------------------------------------------------------------------------------
Aggregated 71 0(0.00%) | 19 15 33 18 | 3.20 0.00More complex example
Before run this example make sure you set autocommit: False in params.yaml:
from xpand_locust import custom_timer
@task(1)
@custom_timer # Now I am going to measure time by myself
def my_complex_transaction(self):
self.client.trx_begin() # Transaction begin
# Insert order
_ = self.client._execute(
"insert into orders (product_name, amount) values (%s, %s)",
(
next(self.products_iterator),
10,
),
row = self.client._query ('SELECT LAST_INSERT_ID();')
# Update last inserted order
_ = self.client._execute(
"update orders set amount = %s where order_no=%s",
(
20,
row[0],
),
self.client.trx_commit()SSL support
You should be able to pass any of the following keys: 'ca', 'key', 'cert' as shown below.
ssl:
ca: sky.pemRunning Xpand-Locust
To be able to make use of more than one core running python, because of the python GIL, separate python instances have to be run. So to utilize the full potential of a CPU with 4 cores, 4 python instances running locust would be used.
It is highly recommend (and required for distributed mode) to have all required files in the same directory
cd example
ls -la
-rw-r--r-- 1 dmitryvolkov staff 216 Dec 8 14:12 params.yaml
-rw-r--r-- 1 dmitryvolkov staff 412 Dec 9 15:24 sales.sql
drwxr-xr-x 3 dmitryvolkov staff 96 Oct 4 15:43 seed_values
-rw-r--r-- 1 dmitryvolkov staff 1761 Dec 8 14:12 swarm_config.yamlRunning as standalone
Standalone means you will be running both master and workers as a single process. You will be able to utilize only one processor core.
cd examples
swarm_runner.py --swarm-config ./swarm_config.yaml --log-level DEBUG -f locustfile_simple run_standalone --run-time 100 --users 10 --spawn-rate 10 --csv mysql --params ./params.yamlRunning master and workers on the same machine as separate processes
This time you will be able to utilize multiple cores on the same machine. This could be useful to check a connectivity before running in the full distributed mode
First you should start workers:
cd examples
swarm_runner.py --swarm-config ./swarm_config.yaml --log-level DEBUG -f ./locustfile_simple run_workers --num-workers 2 --master-host=127.0.0.1 --params ./params.yamlSecondly, start master:
cd examples
swarm_runner.py --swarm-config swarm_config.yaml --log-level DEBUG -f locustfile_simple run_master --run-time 100 --users 10 --spawn-rate 10 --csv mysql --params params.yaml --expected-workers 2Note: if you run manually workers and master on the different hosts you will need to clean up workers (using kill command).
You can run workers and master using single run command as shown below. In this case workers will be killed on your behalf.
cd examples
swarm_runner.py --swarm-config ./swarm_config.yaml --log-level DEBUG -f ./locustfile_simple run --run-time 100 --users 10 --spawn-rate 10 --csv mysql --params ./params.yaml --num-workers 2 --drivers 127.0.0.1Running in truly distributed mode
In order to run in the distributed mode swarm config, params and locust files (as well as all other required files, like seed_values directory) should be in the same directory. This directory will distribute to all drivers.
Now you can run workload:
cd examples
swarm_runner.py --swarm-config ./swarm_config.yaml --log-level DEBUG -f ./locustfile_simple run --run-time 100 --users 10 --spawn-rate 10 --csv mysql --params ./params.yaml --num-workers 2 --drivers yin01a,yin01bOther features
Histograms
It is possible to get histogram per task at the end of the workload. This is approximate a histogram as locust rounds response time in order to sve amount of data sent between worker and master.
In order to enable histograms you need to add a corresponding keyword to the extra master_options to the swarm_config.yaml:
locust_master_options:
extra_options: --headless --csv-full-history --print-stats --reset-stats --histogram # SSH tunnel
Workers running on drivers machines need to send latest stats to master machine using master-bind-port. If for certain reason this port is not open you could request to use ssh tunnel for that communication:
use_ssh_tunnel: True # use ssh tunnel between master and workers
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent