multi-container environment manager
What is bulker?
Bulker builds multi-container computing environments that are both modular and interactive. A bulker environment consists of an individual container image for each command. Bulker environments are portable, interactive, and independent of any specific workflow. Bulker simplifies both interactive analysis and workflow development by building drop-in replacements to command-line tools that act like native tools, but run in containers. Think of bulker as a lightweight wrapper for docker/singularity to simplify sharing complete, containerized environments.
What makes bulker useful?
It simplifies using containers. If you load pandoc with bulker, you'll invoke it in a shell by typing just
pandoc; bulker wraps the details like "
docker run -it --volume ...". It also spans container engines (docker and singularity), reducing to a single interface.You'll be using containers without knowing it.
It distributes multiple tools all at once. You can distribute a set of images, like all the tools needed to run a workflow, with one command:
bulker load namespace/toolset. Bulker loads multiple independent tools as easily as installing just one.
It makes workflows automatically containerized. Since bulker commands act like native commands, a bulker environment immediately makes native workflows containerized. Workflow authors need only provide a list of commands. Bulker automatically containerizes any workflow built with any workflow system.
It makes environments portable and interactive. You can move a bulker environment around from host to host easily because bulker decouples environment settings from tool settings. Bulker environments can be moved independently and activated interactively.
Is bulker a package manager? How is it different?
Bulker is not a package manager, but it can replace some package manager tasks. For example, instead of installing software natively using pip or conda, you could load it using bulker. The differences are:
Bulker uses containers. Package managers install tools natively. Bulker installs containerized software.
Bulker is 'collection first'. Most package managers operate on tools (e.g.
pip install tool,
conda install tool, or
apt install tool). Bulker operates on collections:
bulker load toolset.
With these two features,
bulker makes it easier than a traditional package manager to distribute and use an entire version-controlled computing environment.
Is bulker a workflow manager? How is it different?
Bulker is not a workflow manager, either. But bulker can take over the container-management aspect of an existing workflow. The key difference is:
- Bulker environments are independent. Bulker environments are not tied to workflows. In fact, bulker doesn't know anything about your workflow. It only cares about the computing environment.
Just write your workflow without containers and use a bulker environment instead, and you'll get portability, interactivity, and efficiency for free!
1 Install bulker
pip install --user bulker
2 Load a crate
A bulker crate is a collection of executables that run inside containers. Load the cowsay fortune example:
bulker load demo
3 Activate your new crate:
Activate a crate with
bulker activate demo
Activating a crate will give you drop-in replacement command-line executables for any commands in the crate.
4 Run some commands:
Now run any executables in the crate as if they were installed natively. The first time you run a command, the actual container image will be downloaded automatically by docker:
$ cowsay Hello world!
______________ < Hello world! > -------------- \ ^__^ \ (oo)\_______ (__)\ )\/\ ||----w | || ||
So far, seems similar to a vanilla container system, right? But wait -- the fortune command is also included in the demo crate, so we can pipe that output to cowsay:
fortune | cowsay
_________________________________________ / You are deeply attached to your friends \ \ and acquaintances. / ----------------------------------------- \ ^__^ \ (oo)\_______ (__)\ )\/\ ||----w | || ||
What just happened? The user-issued command behaves as if it's running in an environment that includes both
cowsay; but what's actually happening is each command is running in its own container, as there is no container that contains both fortune and cowsay. That's an interactive, modular, multi-container computing environment. You can mix-and-match individual images efficiently and distribute complete environments in re-usable components.