Skip to main content
Version: 1.0.2

Stress Test

This page will show you how to stress test our data pipeline

1. NATS built-in testing

NATS provide some commands to test/benchmark the data pipeline (nats bench). 💡 Notes:

  • To use it, you have to install the nats_cli first.

  • Like all other nats command, you need to put any authentication you have. For example, if you are using TOKEN as your authentication, your command will look like nats -s nats://TOKEN@HOST [commands] Commands:

  • Without Jetstream

    • nats bench [NAME_OF_TEST] --pub [N] --sub [M] --size [S] --msgs [X]
    • N: number of publisher
    • M: number of subscriber
    • S: size of each message
    • X: number of total messages
  • With Jetstream

    • nats bench --js [NAME_OF_TEST] ...

The result will look like the attached screenshot. You can see how many messages per second for each publisher (mean & standard deviation).

Untitled

More details about it can be found here.

2. Write your own stress testing program

If you want more metrics or test the functionality of the whole pipeline, you probably need to write your own testing program. You can refer to this file change in our PR for more infomation!

2.1 Setup

The stress testing program works similar to a client, except it will use mock data instead of real-world data from dump1090.

  1. We use a shared queue to generate mock data at a fixed speed (10 messages/second to 1000 messages/second)
  2. Next, we have a consumer to pop data from the queue. The consumer will then publish data to the data pipeline. You can define how your messages will be look like now.
  3. You need some mechanism to collect metrics. You can find some examples (latency & throughput) from section 2.2 and 2.3.

2.2 Latency

In order to test latency, we will collect both the time when a client sent it and the time when the Elastic Search see it.

  1. You need to put those two time field into your data.
    1. This can be done easily by using the datetime module in Python for both client and elastic search.
  2. After publishing messages with your stress test script, query the elastic search DB to collect all the data from stress testing.
    1. This can be done by using the elasticsearch module in Python. You can refer to the Elastic Search Notes (link todo) for more information
    2. You might need to wait for all the messages to be stored in the DB since there will be delay when message/second is large
  3. After collecting all the data, do the math whatever you want!
    1. We used mean and standard deviation to show the result, the plot we get looks like this.

      latency.png

2.3 Throughput

In order to test throughput, it is similar to 2.2 latency and even simpler.

  1. After publishing messages with your stress test script, query the elastic search DB to collect all the data from stress testing.
    1. This can be done by using the elasticsearch module in Python. You can refer to the Elastic Search Notes (link todo) for more information
    2. You might need to set a waiting time (X) in your script so that you can # messages client sent vs # messages elastic search received after X seconds.
  2. After collecting all the data, do the math whatever you want!
    1. We used mean and standard deviation to show the result, the plot we get looks like this.

      throughput.png