Need clarification on "client_id"

[briantully]

Hey there. First of all I wanted to say thank you for your amazing work and efforts. I am hoping I can leverage locust to load test my first IoT application.

I was hoping you could clarify the comment within mqtt_locust.py where a random client_id is created by Paho. I have 2500 devices/things that I want to simultaneously test against, each with its own client_id/name. When Paho randomly creates a random client_id, is it using one of the existing/known client_ids or is it creating a unique/new one? If it's a new client_id, how valid is the load test if it's not connecting to an existing thing/client_id?

Any clarification would be greatly appreciated! :)

[concurrencylabs]

Since the main goal of this code is to load test the AWS infrastructure behind an AWS IoT application, in most cases it's OK to test with an empty client_id.

You will still generate load that will help you measure how many shards you'll need in a Kinesis stream, how many EC2 instances you'll need, how fast your Lambda functions will run, or the capacity units consumed in a DynamoDB table, etc.

The way the Paho client works is that when client_id is not provided, Paho creates a completely random client_id to connect to the IoT broker. For simplicity, this code was implemented assuming a random client_id would suffice.

I understand this is not ideal in some use cases, but there is a way around it. You would have to:

• Modify the Locust test (in this case aws-iot-publish.py), so it gets a known client_id each time "class MyThing(MQTTLocust)" is instantiated. Locust instantiates this class each time a user is hatched. This can be accomplished by adding client_id as a member variable and populating its value from a text file or something similar.
• Modify MQTTLocust(Locust) in mqtt_locust.py, so it initializes MQTTClient in the following way: "self.client = MQTTClient(self.client_id)"
• Modify MQTTClient in mqtt_locust.py, so it contains a self.client_id variable, which will be used by the Paho client. This way Paho will be using using a real client_id that exists in your application.

I hope this helps.

[briantully]

It helps immensely!!! Wow, thank you so much for your detailed and thoughtful response. I really appreciate it!

Best regards, Brian

On Feb 15, 2017, at 10:34 PM, Concurrency Labs notifications@github.com wrote:

Since the main goal of this code is to load test the AWS infrastructure behind an AWS IoT application, in most cases it's OK to test with an empty client_id.

You will still generate load that will help you measure how many shards you'll need in a Kinesis stream, how many EC2 instances you'll need, how fast your Lambda functions will run, or the capacity units consumed in a DynamoDB table, etc.

The way the Paho client works is that when client_id is not provided, Paho creates a completely random client_id to connect to the IoT broker. For simplicity, this code was implemented assuming a random client_id would suffice.

I understand this is not ideal in some use cases, but there is a way around it. You would have to:

Modify the Locust test (in this case aws-iot-publish.py), so it gets a known client_id each time "class MyThing(MQTTLocust)" is instantiated. Locust instantiates this class each time a user is hatched. This can be accomplished by adding client_id as a member variable and populating its value from a text file or something similar. Modify MQTTLocust(Locust) in mqtt_locust.py, so it initializes MQTTClient in the following way: "self.client = MQTTClient(self.client_id)" Modify MQTTClient in mqtt_locust.py, so it contains a self.client_id variable, which will be used by the Paho client. This way Paho will be using using a real client_id that exists in your application. I hope this helps.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.