{"id":10213,"date":"2022-12-19T10:45:30","date_gmt":"2022-12-19T15:45:30","guid":{"rendered":"https:\/\/www.redline13.com\/blog\/?p=10213"},"modified":"2023-06-22T14:19:36","modified_gmt":"2023-06-22T18:19:36","slug":"load-testing-with-locust-and-jmeter","status":"publish","type":"post","link":"https:\/\/www.redline13.com\/blog\/2022\/12\/load-testing-with-locust-and-jmeter\/","title":{"rendered":"Guest Post: Load Testing with Locust and JMeter on RedLine13"},"content":{"rendered":"\n

Eldad Uzman, Automation Architect at Merck, is the guest author. You can read more from Eldad at Medium<\/a>.<\/p>\n\n\n\n

Abstract<\/h2>\n\n\n\n

In this article, I\u2019ll present to you a case study of how you can integrate Locust into your Redline13 JMeter tests to get the maximum out of your testing environment, increase your load testing coverage while reducing the usage of infrastructure and thus reducing the costs.<\/p>\n\n\n\n

JMeter in a nutshell<\/h2>\n\n\n\n

JMeter is a java-based load testing tool and one of the most long-standing tools on the market.<\/p>\n\n\n\n

For the sake of this article, it\u2019s main characteristic is it\u2019s thread-based architecture, which means that for each virtual user (VU) JMeter will spawn a new thread in the operating system and each thread will perform the tasks related to the test flow of each virtual user.<\/p>\n\n\n\n

For the most part, JMeter flows are scripted via its GUI, but it also allows some coding, namely by using groovy language.<\/p>\n\n\n\n

Locust in a nutshell<\/h2>\n\n\n\n

Locust is a python-based load testing tool, and it is relatively new on the market.<\/p>\n\n\n\n

Unlike JMeter, locust is powered by asynchronous io, which means that all virtual users run on a single thread, and concurrency is achieved not by spawning new threads in the operating system but rather by switching between IO contexts using a greenlet.<\/p>\n\n\n\n

Hands on with locust<\/h2>\n\n\n\n

Since JMeter is already heavily covered in RedLine13 blog posts<\/a>, in this section I\u2019ll provide a short introduction to the various options available in locust.<\/p>\n\n\n\n

Prerequisites<\/h3>\n\n\n\n

First<\/strong>, lets make sure we have python installed on our pc:<\/p>\n\n\n\n

root@eldaduzman-lap:~# python  --version\nPython 3.10.4\nroot@eldaduzman-lap:~#<\/code><\/pre>\n\n\n\n
Next<\/strong>, we can install locust using pip<\/p>\n\n\n\n
root@eldaduzman-lap:~# pip install locust<\/code><\/pre>\n\n\n\n
Simple example<\/h3>\n\n\n\n
from itertools import count\nfrom locust import HttpUser, between, task\nUSER_NUMBERS = count(1)\n\nclass WebsiteUser(HttpUser):\n\twait_time = between(10, 12)\n\thost = \"https:\/\/postman-echo.com\"\n\n\tdef __init__(self, *args, **kwargs):\n\t\tself._user_number = next(USER_NUMBERS)\n\t\tsuper().__init__(*args, **kwargs)\n\n\t@task\n\tdef index(self):\n\t\tresponse = self.client.get(f\"\/get?var={self._user_number}\", name=\"get_usr_num\")\n\t\tvar = response.json()['args']['var']\n\t\tresponse = self.client.get(f\"\/get?var={var}\", name=\"get_var\")<\/code><\/pre>\n\n\n\n
In this example, each virtual user is initiated with a unique sequence number (similar to the __threadNum variable in JMeter) and then it executes its task every 10-12 seconds<\/p>\n\n\n\n
The task is to send a one get request with the sequence number, extract the number from the response, and then to send it as a different request.<\/p>\n\n\n\n
To execute the locust test script, we can run the following command:<\/p>\n\n\n\n
root@eldaduzman-lap:~# locust -f locustfile.py<\/code><\/pre>\n\n\n\n
This will create a web interface that you can open in your browser at `http:\/\/localhost:8089\/\u2019<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
We will spawn 20 virtual user and leave them to run for a while<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n
Now with the test completed, we see that the average request per second was 3.7, which makes sense, as each user would send 2 requests every ~11 seconds.<\/p>\n\n\n\n
Which means that 10.9 requests per minute will be sent from each user.<\/p>\n\n\n\n
If we multiply that by 20, the total request per minute count should be 218.18, which would make ~3.63 request per second.<\/p>\n\n\n\n
A more complex example<\/h3>\n\n\n\n
Let\u2019s try a different scenario, instead of sending http requests, lets send a message to AWS SQS queue<\/p>\n\n\n\n
Lets install first boto3 from pip:<\/p>\n\n\n\n
root@eldaduzman-lap:~# pip install boto3 <\/code><\/pre>\n\n\n\n
import os\nimport time\nimport sys\nimport json\nfrom itertools import count\nfrom locust import User, between, task\n\nimport boto3\nfrom boto3.exceptions import Boto3Error\n\nREGION = os.environ.get(\"REGION\")\nACCESS_KEY = os.environ.get(\"ACCESS_KEY\")\nSECRET_KEY = os.environ.get(\"SECRET_KEY\")\nQUEUE_NAME = os.environ.get(\"QUEUE_NAME\")\n\nassert (\n    REGION and ACCESS_KEY and SECRET_KEY and QUEUE_NAME\n), \"One or more of the following environment variables is missing [REGION | ACCESS_KEY | SECRET_KEY | QUEUE_NAME]\"\n\nUSER_NUMBERS = count(1)\n\ndef get_queue():\n    \"\"\"get boto3 sqs queue object\"\"\"\n    return boto3.resource(\n        \"sqs\",\n        region_name=REGION,\n        aws_access_key_id=ACCESS_KEY,\n        aws_secret_access_key=SECRET_KEY,\n    ).get_queue_by_name(QueueName=QUEUE_NAME)\n\nclass SQSUser(User):\n    wait_time = between(120, 130)\n    host = \"https:\/\/postman-echo.com\"\n\n    def __init__(self, *args, **kwargs):\n        self._user_number = next(USER_NUMBERS)\n        self._queue = get_queue()\n        super().__init__(*args, **kwargs)\n\n    @task\n    def send_sqs_message(self):\n        \"\"\"sends a message to SQS\"\"\"\n        time_start = time.time() * 1000\n\n        try:\n            aws_response = self._queue.send_message(\n                MessageBody=json.dumps({\"user_id\": self._user_number})\n            )\n            time_end = time.time() * 1000\n            self.environment.events.request_success.fire(\n                request_type=\"SQS\",\n                name=\"queue\",\n                response_time=(time_end - time_start),\n                response_length=sys.getsizeof(aws_response),\n            )\n        except Boto3Error as ex:\n            time_end = time.time() * 1000\n            self.environment.events.request_failure.fire(\n                request_type=\"SQS\",\n                name=\"queue\",\n                response_time=(time_end - time_start),\n                exception=ex,\n                response_length=0,\n            )<\/code><\/pre>\n\n\n\n
In this example, the script assumes that the details for the SQS queue are stored in the environment variables.<\/p>\n\n\n\n
Then it will use the details to generate connections to the sqs, and it will send messages every 120-130 seconds.<\/p>\n\n\n\n
Now we can run this script with 50 users<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
And the results would be as follow:<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n
Locust as library<\/h3>\n\n\n\n
In the examples above, we would be required to execute locust manually, now let\u2019s run it as a python script<\/p>\n\n\n\n
Let\u2019s install another dependency, click, for parsing command line arguments.<\/p>\n\n\n\n
root@eldaduzman-lap:~# pip install click<\/code><\/pre>\n\n\n\n
import os\n\nimport click\nimport gevent\n\nfrom locustfile_sqs import SQSUser\nfrom locust.env import Environment\nfrom locust.html import get_html_report\nfrom locust.env import Environment\nfrom locust.stats import stats_printer, stats_history\n\n@click.option(\n    \"--number-of-users\",\n    help=\"how many locust users to simulate\",\n    type=click.INT,\n    required=True,\n)\n@click.option(\n    \"--duration-in-seconds\",\n    help=\"Duration of execution - if not provided, execution will run infinitely\",\n    type=click.INT,\n    default=-1,\n    required=False,\n)\n@click.option(\n    \"--output-path\",\n    help=\"path to output report\",\n    type=click.STRING,\n    default=\"report.html\",\n    required=False,\n)\ndef main(\n    number_of_users,\n    duration_in_seconds,\n    output_path: str = \"report.html\",\n):\n    env = Environment(user_classes=[SQSUser])\n    env.create_local_runner()\n    assert env.runner, \"must have a runner\"\n    try:\n\n        gevent.spawn(stats_printer(env.stats))\n        gevent.spawn(stats_history, env.runner)\n\n        env.runner.start(number_of_users, spawn_rate=1)\n        if duration_in_seconds > -1:\n            gevent.spawn_later(duration_in_seconds, env.runner.quit)\n        env.runner.greenlet.join()\n    except KeyboardInterrupt:\n        pass\n    finally:\n\n        dir_path = os.path.dirname(os.path.abspath(output_path))\n        if not os.path.exists(dir_path):\n            os.makedirs(dir_path)\n        with open(output_path, \"w\", encoding=\"utf-8\") as html_file:\n            html_file.write(get_html_report(environment=env))\n\ndispatch = click.command()(main)  # make pylint happy\nif __name__ == \"__main__\":\n    dispatch()<\/code><\/pre>\n\n\n\n
In this script, we use click to collect 3 command line argument:<\/p>\n\n\n\n