/invokust

A wrapper for locust to allow running load tests in python or on AWS Lambda

Primary LanguagePythonMIT LicenseMIT

invokust

A tool for running Locust load tests from within Python without the need to use the locust command line. This gives more flexibility for automation such as QA/CI/CD tests and also makes it possible to run locust on AWS Lambda for ultimate scalability.

Installation

Install via pip:

pip3 install invokust

Examples

Running a load test using a locust file:

import invokust

settings = invokust.create_settings(
    locustfile='locustfile_example.py',
    host='http://www.iana.org',
    num_users=1,
    spawn_rate=1,
    run_time='3m'
    )

loadtest = invokust.LocustLoadTest(settings)
loadtest.run()
loadtest.stats()
"{'requests': {'GET_/': {'request_type': 'GET', 'num_requests': 923, 'min_response_time': 113.54585000000128, 'median_response_time': 120.0, 'avg_response_time': 145.68631223510297, 'max_response_time': 331.89674199999786, 'response_times': {270.0: 2, 120.0: 479, 150.0: 17, 200.0: 83, 210.0: 80, 160.0: 20, 190.0: 55, 220.0: 9, 130.0: 30, 170.0: 22, 230.0: 5, 110.0: 69, 140.0: 19, 180.0: 27, 240.0: 2, 320.0: 3, 330.0: 1}, 'response_time_percentiles': {55: 120.0, 65: 150.0, 75: 190.0, 85: 200.0, 95: 210.0}, 'total_rps': 5.136500841568583, 'total_rpm': 308.190050494115}, 'GET_/about': {'request_type': 'GET', 'num_requests': 308, 'min_response_time': 113.23035299999873, 'median_response_time': 120.0, 'avg_response_time': 146.04534828246747, 'max_response_time': 290.40608500000076, 'response_times': {120.0: 147, 200.0: 36, 190.0: 25, 110.0: 27, 160.0: 12, 150.0: 6, 180.0: 13, 210.0: 12, 170.0: 13, 220.0: 2, 130.0: 8, 140.0: 6, 290.0: 1}, 'response_time_percentiles': {55: 120.0, 65: 160.0, 75: 180.0, 85: 200.0, 95: 200.0}, 'total_rps': 1.7140219492991589, 'total_rpm': 102.84131695794953}}, 'failures': {}, 'num_requests': 1231, 'num_requests_fail': 0, 'start_time': 1608207776.312684, 'end_time': 1608207956.070369}"

Running a load test without locust file:

import invokust

from locust import HttpUser, between, task

class WebsiteUser(HttpUser):
    wait_time = between(1, 3)

    @task()
    def get_home_page(self):
        '''
        Gets /
        '''
        self.client.get("/")

settings = invokust.create_settings(
    classes=[WebsiteUser],
    host='http://www.iana.org',
    num_users=1,
    spawn_rate=1,
    run_time='3m'
)

loadtest = invokust.LocustLoadTest(settings)
loadtest.run()
loadtest.stats()
"{'requests': {'GET_/': {'request_type': 'GET', 'num_requests': 71, 'min_response_time': 138.60819600000696, 'median_response_time': 360.0, 'avg_response_time': 327.0060322394364, 'max_response_time': 603.2539320000012, 'response_times': {590.0: 1, 210.0: 3, 370.0: 2, 230.0: 5, 200.0: 2, 490.0: 2, 420.0: 4, 480.0: 1, 190.0: 5, 180.0: 6, 400.0: 3, 270.0: 1, 260.0: 3, 280.0: 2, 360.0: 4, 470.0: 2, 460.0: 3, 350.0: 1, 250.0: 1, 380.0: 4, 410.0: 2, 140.0: 1, 440.0: 1, 600.0: 1, 390.0: 2, 450.0: 1, 430.0: 3, 290.0: 1, 240.0: 2, 340.0: 1, 220.0: 1}, 'response_time_percentiles': {55: 370.0, 65: 390.0, 75: 420.0, 85: 450.0, 95: 490.0}, 'total_rps': 0.4443058717398536, 'total_rpm': 26.658352304391215}}, 'failures': {'GET_/': {'method': 'GET', 'name': '/', 'error': "ConnectionError(ProtocolError('Connection aborted.', RemoteDisconnected('Remote end closed connection without response')))", 'occurrences': 1}}, 'num_requests': 71, 'num_requests_fail': 1, 'start_time': 1608208114.321394, 'end_time': 1608208276.0525749}"

Running Locust on AWS Lambda

Lambda logoLocust logo

AWS Lambda is a great tool for load testing as it is very cheap (or free) and highly scalable.

There are many load testing tools such as ab and wrk. Then there are other cloud based load testing options such as BlazeMeter or Loader and some more DIY solutions that use AWS Lambda too such as Goad or serverless-artillery. But these all have the same drawback: They are too simplistic. They can perform simple GET or POST requests but can't accurately emulate more complex behaviour. e.g. browsing a website, selecting random items, filling a shopping cart and checking out. But with Locust this is possible.

Included is an example function for running Locust on AWS Lambda, lambda_locust.py.

Creating a Lambda function

The process for running a locust test on Lambda involves creating a zip file of the locust load test, creating a Lambda function and then triggering the function.

Install invokust (and its dependencies) python packages locally:

pip3 install invokust --target=python-packages

Or if running on a Mac (python packages need to be compiled for 64 bit Linux) you can use docker:

docker run -it --volume=$PWD:/temp python:3.6 bash -c "pip install /temp --target=/temp/python-packages"

Create the zip file:

zip -q -r lambda_locust.zip lambda_locust.py locustfile_example.py python-packages

Then create the Lambda function using using the AWS CLI:

aws lambda create-function --function-name lambda_locust --timeout 300 --runtime python3.6 --role arn:aws:iam::9999999999:role/lambda_basic_execution --handler lambda_locust.handler --zip-file fileb://lambda_locust.zip

Or Terraform and the example main.tf file:

terraform apply
...

Invoking the function

The Locust settings can be passed to the Lambda function or can be set from environment variables. The environment variables are:

  • LOCUST_LOCUSTFILE: Locust file to use for the load test
  • LOCUST_CLASSES: Names of locust classes to use for the load test (instead of a locustfile). If more than one, separate with comma.
  • LOCUST_HOST: The host to run the load test against
  • LOCUST_NUM_CLIENTS: Number of clients to simulate
  • LOCUST_HATCH_RATE: Number of clients per second to start
  • LOCUST_RUN_TIME: The time the test should run for
  • LOCUST_LOGLEVEL: Level of logging

AWS CLI example with Locust settings in a payload:

aws lambda invoke --function-name lambda_locust --invocation-type RequestResponse --payload '{"locustfile": "locustfile_example.py", "host":"http://www.iana.org", "num_users": "1", "spawn_rate": "1", "run_time":"3m"}' --cli-binary-format raw-in-base64-out output.txt
{
    "StatusCode": 200
}
cat output.txt
"{\"success\": {\"GET_/\": {\"request_type\": \"GET\", \"num_requests\": 20, \"min_response_time\": 86, \"median_response_time\": 93 ...

Python boto3 example:

import json
from boto3.session import Session
from botocore.client import Config

session = Session()
config = Config(connect_timeout=10, read_timeout=310)
client = session.client('lambda', config=config)

lambda_payload = {
    'locustfile': 'locustfile_example.py',
    'host': 'https://example.com',
    'num_users': '1',
    'spawn_rate': 1,
    'run_time':'3m'
}

response = client.invoke(FunctionName='lambda_locust', Payload=json.dumps(lambda_payload))
json.loads(response['Payload'].read())
'{"success": {"GET_/": {"request_type": "GET", "num_requests": 20, "min_response_time": 87, "median_response_time": 99, "avg_response_time": 97.35 ...

Running a real load test

Lambda function execution time is limited to a maximum of 5 minutes. To run a real load test the function will need to be invoked repeatedly and likely in parallel to generate enough load. To manage this there is a class called LambdaLoadTest that can manage invoking the function in parallel loops and collecting the statistics.

import logging
from invokust.aws_lambda import LambdaLoadTest

logging.basicConfig(level=logging.INFO)

lambda_payload = {
    'locustfile': 'locustfile_example.py',
    'host': 'https://example.com',
    'num_users': 1,
    'spawn_rate': 1,
    'run_time':'3m'
}

load_test = LambdaLoadTest(
  lambda_function_name='lambda_locust',
  threads=2,
  ramp_time=0,
  time_limit=30,
  lambda_payload=lambda_payload
)

load_test.run()
print(load_test.get_summary_stats())

The output:

INFO:root:
Starting load test...
Function name: lambda_locust
Ramp time: 0s
Threads: 2
Lambda payload: {'locustfile': 'locustfile_example.py', 'host': 'https://example.com', 'num_users': '1', 'spawn_rate': 1, 'run_time': '3m'}
Start ramping down after: 30s
INFO:root:thread started
INFO:root:Invoking lambda...
INFO:root:threads: 1, rpm: 0, time elapsed: 0s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 1, rpm: 0, time elapsed: 3s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:thread started
INFO:root:Invoking lambda...
INFO:root:threads: 2, rpm: 0, time elapsed: 6s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 9s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 12s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 15s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 18s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 21s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 24s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 27s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 30s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:threads: 2, rpm: 0, time elapsed: 33s, total requests from finished threads: 0, request fail ratio: 0, invocation error ratio: 0
INFO:root:Time limit reached. Starting ramp down...
INFO:root:Waiting for all Lambdas to return. This may take up to 3m.
INFO:invokust.aws_lambda.lambda_load_test:Lambda invocation complete. Requests (errors): 1867 (0), execution time: 180066ms, sleeping: 0s
INFO:root:thread finished
INFO:invokust.aws_lambda.lambda_load_test:Lambda invocation complete. Requests (errors): 1884 (0), execution time: 180065ms, sleeping: 0s
INFO:root:thread finished
{'lambda_invocation_count': 2, 'total_lambda_execution_time': 360131, 'requests_total': 3751, 'request_fail_ratio': 0.0, 'invocation_error_ratio': 0.0}

There is also an example CLI tool for running a load test, invokr.py:

$ ./invokr.py --function_name=lambda_locust --locust_file=locustfile_example.py --locust_host=https://example.com --threads=1 --time_limit=15 --locust_users=2
2017-05-22 20:16:22,432 INFO   MainThread
Starting load test
Function: lambda_locust
Ramp time: 0
Threads: 1
Lambda payload: {'locustfile': 'locustfile_example.py', 'host': 'https://example.com', 'num_users': 2, 'spawn_rate': 10, 'run_time': '15s'}

[2020-06-28 19:58:22,103] pudli/INFO/root: thread started
[2020-06-28 19:58:22,107] pudli/INFO/root: threads: 1, rpm: 0, run_time: 0, requests_total: 0, request_fail_ratio: 0, invocation_error_ratio: 0
[2020-06-28 19:58:25,108] pudli/INFO/root: threads: 1, rpm: 0, run_time: 3, requests_total: 0, request_fail_ratio: 0, invocation_error_ratio: 0
[2020-06-28 19:58:28,109] pudli/INFO/root: threads: 1, rpm: 0, run_time: 6, requests_total: 0, request_fail_ratio: 0, invocation_error_ratio: 0
[2020-06-28 19:58:31,110] pudli/INFO/root: threads: 1, rpm: 0, run_time: 9, requests_total: 0, request_fail_ratio: 0, invocation_error_ratio: 0
[2020-06-28 19:58:34,112] pudli/INFO/root: threads: 1, rpm: 0, run_time: 12, requests_total: 0, request_fail_ratio: 0, invocation_error_ratio: 0
[2020-06-28 19:58:37,113] pudli/INFO/root: threads: 1, rpm: 0, run_time: 15, requests_total: 0, request_fail_ratio: 0, invocation_error_ratio: 0
[2020-06-28 19:58:39,001] pudli/INFO/invokust.aws_lambda.lambda_load_test: Invocation complete. Requests (errors): 224 (120), execution time: 15066, sleeping: 0
[2020-06-28 19:58:40,116] pudli/INFO/root: threads: 1, rpm: 795, run_time: 18, requests_total: 224, request_fail_ratio: 0.5357142857142857, invocation_error_ratio: 0.0
[2020-06-28 19:58:40,117] pudli/ERROR/root: Error limit reached, invocation error ratio: 0.0, request fail ratio: 0.5357142857142857
[2020-06-28 19:58:40,117] pudli/INFO/root: Waiting for threads to exit...
[2020-06-28 19:58:54,086] pudli/INFO/invokust.aws_lambda.lambda_load_test: Invocation complete. Requests (errors): 242 (131), execution time: 15052, sleeping: 0
[2020-06-28 19:58:54,086] pudli/INFO/root: thread finished
[2020-06-28 19:58:54,142] pudli/INFO/root: Aggregated results: {"requests": {"GET_/": {"median_response_time": 92.0, "total_rps": 7.18569301694931, "avg_response_time": 91.08271769409947, "max_response_time": 114.66264724731445, "min_response_time": 84.4886302947998, "response_times": {"histogram": [85, 45, 4, 6, 7, 47, 11, 0, 0, 10], "bins": [84.0, 86.6, 89.2, 91.8, 94.4, 97.0, 99.6, 102.2, 104.8, 107.4, 110.0]}, "total_rpm": 431.1415810169586, "num_requests": 215}, "POST_/post": {"median_response_time": 150.0, "total_rps": 8.38878329746517, "avg_response_time": 157.73737294831653, "max_response_time": 1087.4686241149902, "min_response_time": 142.15636253356934, "response_times": {"histogram": [247, 0, 0, 1, 2, 0, 0, 0, 0, 1], "bins": [140.0, 236.0, 332.0, 428.0, 524.0, 620.0, 716.0, 812.0, 908.0, 1004.0, 1100.0]}, "total_rpm": 503.32699784791026, "num_requests": 251}}, "failures": {"POST_/post": {"method": "POST", "name": "/post", "error": "HTTPError('404 Client Error: Not Found for url: https://example.com/post',)", "occurrences": 251}}, "num_requests": 466, "num_requests_fail": 251, "total_lambda_execution_time": 30118, "lambda_invocations": 2, "approximate_cost": 6.3008e-05, "request_fail_ratio": 0.5386266094420601, "invocation_error_ratio": 0.0, "locust_settings": {"locustfile": "locustfile_example.py", "host": "https://example.com", "num_users": 2, "spawn_rate": 10, "run_time": "15s"}, "lambda_function_name": "lambda_locust", "threads": 1, "ramp_time": 0, "time_limit": 15}
[2020-06-28 19:58:54,142] pudli/INFO/root: ===========================================================================================================================
[2020-06-28 19:58:54,143] pudli/INFO/root: TYPE    NAME                                                #REQUESTS    MEDIAN   AVERAGE       MIN       MAX  #REQS/SEC
Scratch
[2020-06-28 19:58:54,143] pudli/INFO/root: ===========================================================================================================================
[2020-06-28 19:58:54,143] pudli/INFO/root: GET     /                                                         215      92.0     91.08     84.49    114.66       7.19
[2020-06-28 19:58:54,144] pudli/INFO/root: POST    /post                                                     251     150.0    157.74    142.16   1087.47       8.39
[2020-06-28 19:58:54,144] pudli/INFO/root: Exiting...

Occasional errors

  • ERROR : xxxxx-3f19-11e7-a1d1-xxxxxxx Process exited before completing request"
    • SOLUTION: Double the size of the memory for the function.