dowding

Docker cluster scheduler

The name is a tribute to the real master planner

install

$ npm install dowding

usage

var dowding = require('dowding')

var scheduler = dowding({
    // pass a function that will list our inventory
    inventory:function(done){

        // the inventory is an array of objects
        // each object has 'hostname' and 'docker' properties
        // servers can also have arbitrary meta data

        var servers = [{
            hostname:'host1',
            docker:'192.168.8.120:2375',
            tags:'apple,peach'
        },{
            hostname:'host2',
            docker:'192.168.8.121:2375',
            tags:'orange'
        }]

        done(null, servers)
    },

    leastBusy:function(servers, done){

        // this is a naive function that always returns the first server
        // you can have any logic you want here
        done(null, servers[0])
    }

})

To allocate a new container:

scheduler.allocate({
    name:'myjob'
}, function(err, server){
    // server is an entry from our inventory	
})

Allocate a server that does not run another 'myjob' container (mutual exclusion):

scheduler.allocate({
    name:'myjob.2'
}, function(err, server){
    // server will not be where myjob.1 is running
})

Allocate the server that is running the basevolume container:

scheduler.allocate({
    name:'mydatabase',
    parent:'basevolume'
}, function(err, server){
    // server will be the same as where basevolume is running
    // it will pass an error if basevolume is not found
})

mutual exclusion

If a container has a name of the following format:

<jobname>.<pid>

For example:

auth.abc

and

auth.xyz

These 2 auth containers will not be allocated onto the same server.

Here is an example of running the allocation:

scheduler.allocate({
    name:'auth.abc'
}, function(err, server){
    // server is one object from the inventory
})

If you just want a server you can omit the filter information:

scheduler.allocate(function(err, server){
    // server is one object from the inventory
})

parent

You can instruct dowding to allocate a container onto a server that another container is running.

This is useful in scenarios like --volumes-from where one container has a local dependency on another.

least-busy

When there are multiple servers that could be choosen - the default behaviour is to pick the least busy server.

Least busy is determined by the number of containers running on each host.

You can pass a function to override this decision.

The function will be passed an object where the keys are the server hostname and the values are objects with the following properties:

• hostname
• docker
• jobs

You can use this information to decide which server should be allocated.

api

var scheduler = dowding(opts)

Create a new scheduler passing the following options:

• inventory - a function that will return a list of servers on our network
• leastBusy - a function that defines the logic for picking the least busy of our servers

scheduler.allocate([opts], function(err, server){})

Allocate a docker server for a job

Opts is an optional object used to filter the allocation - it has these keys:

• name - the name of the container, possibly in . format for mutual exclusion
• parent - the name of a parent container - the job will be routed to the server it is running on

scheduler.find(name, function(err, server){})

Find which server that a container is running - server can be null if the container is not found.

license

MIT