Ymir

Ymir is designed as a distributed script tool running on kubernetes.

Design

master - manage test task agent - get/run test work from master, agent's life cycle is same with a single task dashboard - ymir frontend, on which user can manage and start task task, watch test metric and node load.

What is Ymir

Ymir is designed as distributed script tool which can be used as user load testing tool like locust.

• Ymir is running on kubernetes, it can test service target in or out kubernetes.
• Ymir is first designed for user load test, but as a general script tool, it can be used to run any distributed script
• Ymir use golang as test script, you can edit and rerun online.
• Ymir job defined with ymir script and node you selected to run, after a test job finish, it collect running status and runing metrics such as status code, avg/max/min latency.

Ymir Script

Ymir Script is user's golang script which implented interface task, here are examples.

• A Ymir Script may contain one or more tasksets, taskset show be registered and ymir will run parallelly. Load will be distribute to tasksets by taskset's Weight
• A taskset may contain one or more tasks and ymir will run tasks in a taskset serially
• Ymir Script can import and use any package in golang std library

// TTaskSet ...
type TTaskSet interface {
	Name() string  // taskset name
	Weight() int   // default 1
	Tasks() []TTask // tasks
}

//TTask ....
type TTask interface {
	Name() string  // task name
	Run() int     // entrypoint
}

// Register your taskset with a NewFunc ...
func Register(newfunc NewFunc) {
}

// Set Run Time for your script ...
func SetRunTime(t time.Duration) {
}

// Set Routine Count for your script ...
func SetRoutineCount(c int) {
}

Demo

Deploy / Develop

deploy

cd deploy && kubectl create -f server_deploy.yaml

develop

# change ImageDestBase and Tag in makefile
make build  # build dashboard and binary
make publish_docker # build and push server docker image
make publish_docker_agent # build and push agent docker image

RoadMap

• datamodel: done
• task management: done
• task flow control: 90%
• node/pod monitor: get node monitor metrics during a task