A script to benchmark an MQTT broker

## Project description

This is a tool to benchmark the performance of an mqtt broker. It works by launching an arbitrary number of publishers and subscribers in parallel. These workers both publish and subscribe to the same topic and either send or recieve a fixed number of messages.

## Installation

The easiest way to install pymqttbench is use pip:

pip install pymqttbench

Alternatively you can clone the repo with:

git clone https://github.com/mtreinish/pymqttbench.git

then install it using pip:

pip install ./pymqttbench

Alternatively you can run:

cd pymqttbench && python setup.py install

however using pip is recommended.

## Usage

After installing pymqttbench you simply run it with the:

pymqttbench --hostname \$BROKER_HOST

command. The hostname parameter is required to tell pymqttbench the hostname of the broker. This is the only required field, but there are several other options exposed for how to connect to the broker. --port is used to specify the port if you’re not connecting on the standard port, 1883. --username and --password are used to specify user authentication if needed. Similiarly --cacert can be used to specify a trusted CA certificate to verify the TLS connection on the broker. There is also the --topic parameter which is used to specify a topic to use for the benchmark, by default pybench is used. Note that all of these settings are used for both the publishers and subscribers.

Outside of mqtt connection options there is also the --brief flag which is used to print a colon separated list of benchmark results instead of the default human readable formatted output. The format for this output is:

Subscriber Count:Publisher Count:Subscriber Mean Duration:Subscriber Duration Std. Deviation:Subscriber Avg. Throughput:Subscriber Total Throughput:Publisher Mean Duration:Publisher Duration Std. Deviation:Publisher Avg. Throughput:Publisher Total Throughput

### Tuning the benchmark

After pymqttbench knows how to connect to the broker you can tune the benchmark to your specific needs. There are several axes you can adjust the bechmark on, the first being the number of publishers and subscribers. This is configurable with the --pub-clients and --sub-clients flags. By default each is set to 10. The next option you can tune is the number of messages that the publishers will send, with --pub-count, and the number of messages the subscribers will listen for, with --sub-count. It’s worth pointing out that the subscribers do not pair with an individual worker like in some other benchmarking tools, but instead listen to the same topic that all the publishers publish to. In addition with adjusting these options you’ll likely want to change the publisher timeout, --pub-timeout, and the subscriber timeout, sub-timeout, which describe how long the benchmark will wait for the worker to either publish or recieve the specified message count.

You can also set how large the message payload is with --msg-size which takes the number of bytes to use. By default it uses a 1024 byte payload. The last tuning option is --qos which is used to specify the qos level to use for benchmarking. By default qos 0 is used.

## Project details

Uploaded source