juice

A tool to test the performance of MariaDB, Galera and CockroachDB with OpenStack using enoslib

Installation

On a Grid'5000 frontend:

git clone https://github.com/Marie-Donnie/juice.git
cd juice

virtualenv -p python3 venv 
source venv/bin/activate
pip install -U pip
pip install -r requirements.txt

Configuration

There is a sample configuration you can use cp conf.yaml.sample conf.yaml You can then change conf.yaml according to your needs. For instance, change the value of database from cockroachdb to mariadb or galera to go with the mariadb or a galera replicated database.

Launch

Launch with ./juice.py deploy.

Once it has been launched, you can destroy the containers using ./juice destroy and then restart them with ./juice deploy.

Full experiment

In the experiments folder you'll find different scenarios you can tweak to accomodate to your needs. For example, if you want to test the impact of latency on the cluster:

cd experiments
python latency-impact.py

Getting information about the used environment

Use ./juice.py info to get informations about current environment. You can format the results into json, yaml and pickle using ./juice.py info --out <format>.

Benchmarking

Sysbench

You can launch sysbench tests afterwards using ./juice stress.

Rally (CockroachDB only for now)

1. Launch devstack (only Keystone service) using ./juice openstack. This will deploy Keystone on every database nodes. (- Note from 02/22/18 - this is a bit unstable for now, you may have to use it several times until it succeeds. You can check the logs using the command ansible will provide you before launching stack.sh)
2. You can use either ./juice.py rally --files <service-name>/<scenario> to launch one or several specific scenarios or ./juice.py rally --directory <service-name> to launch every scenarios for the given service.
   • tip1: you can use ./juice.py rally --directory . to launch every scenarios
   • tip2: you can use ./juice.py rally to execute every keystone scenarios

Grafana

There are two data sources for Grafana, using InfluxDB: collectd (only for MariaDB currently) and cAdvisor.

To monitor activity on your databases:

1. Open a ssh tunnel in a shell usingssh -NL 8080:<control-node>:3000 <location-of-the-node>.g5k (assuming you have followed Grid'5000 tutorial). You can check which node is used using ./juice.py info and find control node.
2. Go to http://localhost:8080/ in your browser
3. You can check your data sources in Configuration -> Data Sources
4. There are several dashboard you can import for cadvisor and collectd:
   • Service - MySQL InnoDB Metrics provides performance metrics about MariaDB Innodb engine
   • Service - MySQL Metrics provides performance metrics for MariaDB
   • For cAdvisor, you have to make your own dashboards using whatever metrics you need because the one made for cAdvisor/InfluxDB does not work with Juice

Emulate

You can emulate different constraints on the network with emulate by changing the juice.py (or directly in the experiment file if you use one). The constraints look like that:

tc = {
    "enable": True,
    "default_delay": "0ms",
    "default_rate": "10gbit",
    "constraints": [{
        "src": "database",
        "dst": "database",
        "delay": "0ms",
        "rate": "10gbit",
        "loss": "0",
        "network": "database_network",
    }],
    "groups": ['database'],
}

The constraints are applicable to a specific network (network) for different roles (src and dst). You can change the delay, rate and packet loss of the network. There are also global choices for the network, with a default delay and rate, applicable if need be on specific roles.

Backup

Backup important metrics using ./juice.py backup.

Destroy

The destroy tasks, called with ./juice.py destroy removes all dockers and unmount volumes.

Help

Tool to test the performances of MariaDB, Galera and CockroachdB with OpenStack on Grid'5000 using Enoslib

Usage:
    juice [-h | --help] [-v | --version] <command> [<args>...]

Options:
    -h --help      Show this help
    -v --version   Show version number

Commands:
    deploy         Claim resources from g5k and configure them
    openstack      Add OpenStack Keystone to the deployment
    rally          Benchmark the Openstack
    stress         Launch sysbench tests (after a deployment)
	  emulate        Emulate network using tc
    backup         Backup the environment
    destroy        Destroy all the running dockers (not the resources)
    info           Show information of the actual deployment

Run 'juice COMMAND --help' for more information on a command