c54185a4f5 | ||
---|---|---|
.. | ||
Azure_Login.md | ||
Deploy_Load_Test_VM.md | ||
README.md | ||
azuredeploy.parameters.loadtest.defaults.json | ||
loadtest.sh | ||
moodle-on-azure-test-course-1.mbz | ||
simple-test-1.jmx | ||
simple-test-2.jmx |
README.md
Load-Testing Deployed Moodle Cluster
This directory currently contains utility scripts, a Moodle test course, and an Apache jMeter test plan that can be used to load-test a Moodle cluster on Azure using the Azure Resource Manager templates in this repository.
Prerequisites
To run load tests using the resources in this directory, you'll want to spin up a VM to manage the Load Test in your Azure subscription. This VM will generate the traffic, running it in Azure will minimize network charges.
Deploying Moodle using templates and running load test
Once dependencies are installed, you can initiate the load testing process by using included utility scripts. These scripts will:
- deploy a Moodle cluster
- set up the test course in Moodle
- enrol students for the course
- run a synthetic test workload using jMeter
- teardown the test cluster
Use the function deploy_run_test1_teardown
to perform all these
steps. This function takes 18 parameters in the following order:
See the included example run_load_test_example
in
loadtest/loadtest.sh
. At the time of writing this example is
configured as follows:
ssh $ipAddress "az login --username $AZURE_LOGIN --password $AZURE_PASSWORD; az account set --subscription $AZURE_SUBSCRIPTION_ID; run_load_test_example"
Running this example will deploy a cluster with the following configuration:
- Apache web server
- Standard_DS2_v2 Azure VM SKU
- mysql database (with 200 DTU and 125GB DB size)
- NFS file share (with 2 disks and 128GB disk size each)
- uses your SSH pub key in
~/.ssh/id_rsa
[NOTE ON SSH KEYS] Ensure your ~/.ssh/id_rsa
has been added to ssh-agent using eval $(ssh-agent)
and ssh-add
).
Once the Moodle cluster is deployed and configured with course and student data (using moosh it will run the synthetic workload with designated number of concurrent threads (in the example we use 1600 thread) for the designated duration and rampup time (18000 seconds = 5 hours duration, 4800 seconds rampup time in the example).
Test plans
We'd like to offer test plans that are as realistic as possible, so that potential Moodle users on Azure can have better confidence with their Moodle deployment on Azure. We are just starting out in that direction and here are descriptions of currently available test plans.
Simple Scenario simple-test-1.jmx
This test plan is a simple scenario that performs the following operations repeatedly:
- Login to the Moodle site
- View the test course
- View any resource if any
- View a forum in the test course
- View any forum discussion
- Post a new discussion topic
- Take a quiz and submit
The scripts in loadtest.sh are tailored for this test plan.
Moodle Data Stress Testing simple-test-2.jmx
Currently loadtest.sh doesn't have any tailored scripts for this test plan. Therefore, this test plan will need to be executed by issuing the actual jmeter command with properly modified parameters manually, or it'd be greatly appreciated if someone can contribute better support for this test plan in loadtest.hs.
The purpose of this test plan is to try stressing the moodledata directory in a shared file system (either a gluster volume or an NFS share, depending on the choice). Initially attaching a random file in a forum discussion post was tried, but for some reason (probably due to my lack of understanding in PHP/web interaction), files were not attached. I instead tried to upload random files to each test Moodle users's Private Files area, and it did work. This test plan basically performs the following operations repeatedly:
- Login to the Moodle site
- Open the Moodle user's Private Files repository
- Upload a randomly generated file (of a random size within a hard-coded range)
- Save the change
This way, we were able to populate the shared moodledata directory with random files in Moodle users' Private Files repositories. The mechanism to generate random files is not so efficient, so that's currently what slows down the upload speed, and any improvement in that BeanShell preprocessor code would be great. Note that the uploaded files have to be different. Moodle seems so good at deduplicating that a single file uploaded multiple times by different users won't increase the file system usage beyond its single copy.
It'd be also great if we add a download operation step in the test plan, and it's left as a future work item.
Please contribute
It'd be great if we have more test plans, and make other parameters configurable (for example, make the auto-scaling thresholds configurable, which actually requires some changes in the templates as well). The currently available test plans also have hard-coded database type (JDBC connection string) that won't work for Postgres SQL server, so making it work would be also greatly appreciated. Also, if you run this load test with any parameters, it'd be great to share the numeric results so that we can have more performance data on various configurations. Here is a link to an Excel spreadsheet where anyone can share their load testing results.
Acknowledgement
The original test course and the test plan were generously provided by Catalyst as part of this template modernization project. jMeter is a great load testing tool, and also thanks to moosh, the whole process could be automated without too much difficulty, which was really nice.