Loadtesting VerneMQ

Loadtesting VerneMQ with vmq_mzbench

You can loadtest VerneMQ with our vmq_mzbench tool. It is based on Machinezone's very powerful MZBench system and lets you narrow down what hardware specs are needed to meet your performance goals. You can state your requirements for latency percentiles (and much more) in a formal way, and let vmq_mzbench automatically fail, if it can't meet the requirements.

If you have an AWS account, vmq_mzbench can automagically provision worker nodes for you. You can also run it locally, of course.

1. Install MZBench

Please follow the MZBench installation guide

2. Install vmq_mzbench

Actually, you don't even have to install vmq_mzbench, if you don't want to. Your scenario file will automatically fetch vmq_mzbench for any test you do. vmq_mzbench runs every test independently, so it has a provisioning step for any test, even if you only run it on a local worker.

In case you still want to have `vmq_mzbench on your local machine, go through the following steps:

git clone git://github.com/vernemq/vmq_mzbench.git
cd vmq_mzbench
./rebar get-deps
./rebar compile

To provision your tests from this local repository, you'll have to tell the scenario scripts to use rsync. Add this to the scenario file:

{make_install, [
{rsync, "/path/to/your/installation/vmq_mzbench/"},
{exclude, "deps"}]},

If you'd just like the script itself fetch vmq_mzbench, then you can direct it to github:

{make_install, [
{git, "git://github.com/vernemq/vmq_mzbench.git"}]},

3. Write vmq_mzbench scenario files

MZBench recently switched from an Erlang-styled Scenario DSL to a more python-like DSL dubbed BDL (Benchmark Definition Language). Have a look at the BDL examples on Github.

You can familiarize yourself quickly with MZBench's guide on writing loadtest scenarios.

There's not much to learn, just make sure you understand how pools and loops work. Then you can add the vmq_mzbench statement functions to the mix and define actual loadtest scenarios.

Here's a list of the most important vmq_mzbench statement functions you can use in MQTT scenario files:

  • random_client_id(State, Meta, I): Create a random client Id of length I

  • fixed_client_id(State, Meta, Name, Id): Create a deterministic client Id with schema Name ++ "-" ++ Id

  • worker_id(State, Meta): Get the internal, sequential worker Id

  • client(State, Meta): Get the client Id you set yourself during connection setup with the option {t, client, "client"}

  • connect(State, Meta, ConnectOpts): Connect to the broker with the options given in ConnectOpts

  • disconnect(State, Meta): Disconnect normally

  • subscribe(State, Meta, Topic, QoS): Subscribe to Topic with Quality of Service QoS

  • subscribe_to_self(State, _Meta, TopicPrefix, Qos): Subscribe to an exclusive topic, for 1:1 testing

  • unsubscribe(State, Meta, Topic): Unubscribe from Topic

  • publish(State, Meta, Topic, Payload, QoS): Publish a message with binary Payload to Topic with QoS

  • publish(State, Meta, Topic, Payload, QoS, RetainFlag): Publish a message with binary Payload to Topic with QoS and RetainFlag

  • publish_to_self(State, Meta, TopicPrefix, Payload, Qos): -> Publish a payload to an exclusive topic, for 1:1 testing

It's easy to add more statement functions to the MQTT worker if needed. For a full list of the exported statement functions, we encourage you to have a look at the MQTT worker code directly.

Last updated