edenhill / muckrake

Confluent Platform system tests

muckrake

Confluent Platform system tests System Integration & Performance Testing

This repository contains system tests for the Confluent Platform as well as a few scripts to help set up a Vagrant cluster provisioned with all of the repositories in the Confluent Platform. The system tests should be run using ducktape (see https://github.com/confluentinc/ducktape/ for installation instructions).

Muckrake set up

In order to run the muckrake system tests using ducktape, a few preparatory steps are necessary.

• Get muckrake and ducktape (the base directory doesn't matter):

    git clone [e-mail]:confluentinc/muckrake.git
    git clone [e-mail]:confluentinc/ducktape.git
    cd ducktape && python setup.py install # Now you can run ducktape!

• Download and install the Confluent projects within muckrake directory. This provides a mechanism to help provision the machines in your Vagrant cluster with Confluent binaries, scripts etc. (If you already initialized the repositories and want to update, instead run ./build.sh --update && vagrant rsync)

    cd muckrake
    ./build.sh # check out and build Confluent projects against specified version

• Configure Vagrant by creating the file Vagrantfile.local. num_workers = <enough workers to run your tests>.

    # example Vagrantfile.local
    num_workers = 7  # This should be set high enough

• Ensure your Vagrant installation has the necessary plugins.

    # Check that vagrant-hostmanager is in your list of installed plugins
    vagrant plugin list | grep vagrant-hostmanager
    
    # If it's not there, you'll need to install:
    vagrant plugin install vagrant-hostmanager

• Bring up the Vagrant cluster

    # Run this command if you're testing on your local machine
    vagrant up 
    
    # Run this command if you're on aws
    # vagrant provision is only necessary the first time you bring up your Vagrant cluster
    vagrant up --provider=aws --no-parallel && vagrant provision
    

Run tests

Run one or more tests with ducktape.

    # Run tests in a file
    ducktape muckrake/tests/test_my_system_test.py
    
    # Run tests in a directory (search recursively)
    ducktape muckrake/tests
    
    # See which tests will be run without actually running any
    ducktape muckrake/tests --collect-only

Summary results are printed to screen and output into ./results. The most recent test results are sym-linked from ./results/latest.

If you find that tests are never getting loaded, you may be experiencing confluentinc/ducktape#32. Some users have trouble with the test loader due to the muckrake directory not being in the module search path (sys.path). As a temporary work-around, add this line to your bashrc: export PYTHONPATH=$PYTHONPATH:<path_to_muckrake>

Adding New Repositories

If your new service requires a new code base to be included, you may also need to modify the build.sh script. Ideally you just need to add a new line like

build_maven_project "kafka-rest" "[e-mail]:confluentinc/kafka-rest.git" "package"

at the bottom of the script. The generic build_maven_project function uses the given directory name, git repository, and Maven action to check out and build the code. Any code under the repository is automatically pushed to worker nodes and made available under /vagrant. (You'll need to use the "install" action if the repository is for a shared library since subsequent packages will need to be able to find it in the local Maven repository. You probably also want to update the .gitignore file to ignore the new subdirectory and vagrant/base.sh to provide a symlink under /opt in addition to the code under /vagrant to get Vagrant-agnostic naming.

EC2 Quickstart

• Start jump server instance (manually through the AWS management console)

• If you haven't already, setup a keypair to use for SSH access.
• Nothing too big, t2.small or t2.medium is easily enough since this machine is just a driver. t2.micro is even usable, but you do have to do builds locally so you might want something a bit bigger.
• Ubuntu is recommended (only because the initial setup has been documented and tested below)
• Most defaults are fine. Make sure you get an auto-assigned public IP (normally on for the default settings) and set IAM role -> ducktape-master, which gives permissions to launch/kill additional machines. Tagging the instance with a useful name, e.g. "-ducktape-master" is recommended. Set the security group to 'ducktape-insecure', which leaves the machine open on a variety of ports for traffic coming from other machines in the security group and enables SSH access from anywhere. This is less secure than a production config, but makes it so we generally don't have to worry about adding ports whenever we add new services to ducktape.
• Once started, grab the public hostname/IP and SSH into the host using ssh -i /path/to/keypair.pem [e-mail]. All remaining steps will be run on the jump server. Highly recommended: use tmux so any connectivity issues don't kill your session (installation - mac: brew install tmux; ubuntu/debian: sudo apt-get install tmux).
• Put a copy of the SSH key you're using (the pem file) on the jump server using scp -i /path/to/confluent.pem /path/to/confluent.pem [e-mail]:confluent.pem

• Start by making sure you're up to date and installing a few dependencies, getting ducktape, and building:

    sudo apt-get update && sudo apt-get -y upgrade && sudo apt-get install -y git
    git clone https://github.com/confluentinc/ducktape.git
    cd muckrake
    aws/aws-init.sh
    . ~/.bashrc
  

  Now is a good time to install any extra stuff you might need, e.g. your preferred text editor.

• An example Vagrantfile.local has been created by aws-init.sh which looks something like:

    ec2_instance_type = "..." # Pick something appropriate for your
                              # test. Note that the default m3.medium has
                              # a small disk.
    enable_dns = true
    num_workers = 4
    ec2_keypair_name = 'confluent'
    ec2_keypair_file = '/home/ubuntu/confluent.pem'
    ec2_security_groups = ['ducktape-insecure']
    ec2_region = 'us-west-2'
    ec2_ami = "ami-29ebb519"
  

  These settings work for the default AWS setup you get with Confluent's account (of course num_workers, ec2_keypair_name and ec2_keypair_file should all be customized for your setup).

• Start up the instances:

    vagrant up --provider=aws --no-provision --no-parallel && vagrant provision
  

• Now you should be able to run tests:

    python -m ducktape.tests.native_vs_rest_performance
  

• Once configured, you can just shutdown your machines until you need them again. To do so, use

    vagrant halt
  

  to stop your cluster and manually stop your jump server from the management console. To bring them back up, restart your jump server manually, then log in and run

     vagrant up --provider=aws --no-provision && vagrant provision
  

  to restore the cluster. You need to reprovision to fix up some permissions for ephemeral drives (under /mnt) which are lost when you halt the VMs, but it should be very fast since most of the provisioning work is maintained across reboots.

  Updating AMI

  To speed up the provisioning process when bringing up a new node, we bring up nodes using pre-provisioned amazon machine instances (AMI). From time to time as dependencies change, it may be necessary to update the AMI. To do so manually:

  • Bring up an ec2 cluster with Vagrant (one node is enough) and provision it with vagrant up --provider=aws --no-provision && vagrant provision.
  • When the machine is running, sign on to AWS, navigate to EC2 dashboard, and click on Running Instances link.
  • Right-click one of the provisioned machines and select images->Create image
  • Once the image (AMI) is created, it will have an id that looks something like ami-fdc2f6cd
  • Replace the id with the newly generated one in your Vagrantfile.local, and in muckrake/aws/aws-example-Vagrantfile.local