@clusterworks/cdocker v1.0.2

cdocker

An easy way to run commands across a cluster of Docker Engines without needing a Swarm.

Example

cdocker run -d -e "constraint:node==enode{1..10}" --name "nginx_{1..10}" -p 80:80 nginx

It looks like a regular Docker run command but with a few smart additions:

• Prefix your command with cdocker instead of docker
• Use classic Swarm mode constraints
• Use brace expansion in constraint and name parameters to create arrays of containers

This command will create an array of 10 containers, 1 on each node, named nginx_1 on node enode1, nginx_2 on node enode2 and so on.

In addition to brace expansions, you can use globbing:

cdocker kill "nginx_*"

will kill containers on all nodes with names that match the nginx_* pattern.

You can narrow this down by specifying which nodes, again similar to how they are represented in classic Swarm:

cdocker kill "enode{1..5}/nginx_*"

will kill the matching containers only on enode1 to enode5.

Installation

It is written in Node.js, so install using npm:

sudo npm install -g cdocker

and add your nodes like so

cdocker node add 192.168.0.2

The default port we will try and find your Docker Engines on is 2375 but that can be changed with the --port option.

Of course, you can also use brace expansion here

cdocker node add "192.168.0.{11..20}"

When you add or remove nodes, this list will be stored in the ~/.cdocker.json file.

Scheduling

Scheduling arrays of containers to nodes is achieved by first finding nodes that match the specified constraints.

At the moment we only support name and label constraints

Then, one of 3 situations will be reached:

• There are less nodes matched than there are containers you want to start (nodes < containers)
• There are exactly the same number of nodes to containers you want to start (nodes == containers)
• There are more nodes matched than there are containers you want to start (nodes > containers)

Less nodes than containers you want to start

In this instance, cdocker will use a round robin strategy to distribute the containers among the matched nodes.

For example, if you want to start 10 containers on 5 nodes, each node will have 2 containers on it.

Exactly the same number of nodes as containers you want to start

In this instance, cdocker will create a straight up mapping of 1 container to 1 node.

More nodes than containers you want to start

In this instance, you can decide between different strategies on how to utilise the available nodes that matched your constraints.

• unique-random will schedule the containers to random nodes, never using the same node twice
• least-running will order the nodes based on how many containers are already running on them and use those that are the least loaded

You can specify the strategy in the run command as an environment variable, just like the constraints:

-e strategy:unique-random

If you don't specify a strategy, least-running will be used by default. However, this requires enumeration of all matched nodes which may degrade performance on very large clusters.