What is Layercake?
https://github.com/kraken-hpc/kraken-layercake
Kraken “Layercake” is a kraken-based tool for provisioning and managing clusters of physical (or virtual) computer systems, such as those used in high-performance compute clusters.
The layer design
“Layercake” refers to the model used to manage systemm images. As opposed to traditional HPC management systems, Layercake provisions nodes in layers:
- Layer0: aka the MinimalOS or MinOS - A light-weight, persistent OS image that runs services needed to automate and orchestrate loading higher layers. This layer runs a kraken-based layercake agent and the ImageAPI service that loads higher-level functionality (Layer1), called “ImageSets” on top of the Layer0 as Linux namespaced containers.
- Layer1: aka ImageSets are sets of container images (potentially full, systemd-based system images) that are loaded on top of Layer0. They represent higher-level functions of the node in the system, such as the ability to run job schedulers and traditional system management tools.
- Layer2: the intended layer for jobs to run. Layer2 is an additional layer typically containing user-provided software, e.g. scientific software.
This layered model provides many advantages, including the ability to dynamic “roll” system images extremely rapidly and even transparently to the end-user.
Automation
Because Layercake is kraken-based, it inherits all of the distributed automation features of the kraken framework. For instance, all of the orchestration of the Layercake layers is automated and self-healing.
Examples
To try out the Layercake system, take a look at the Examples (see their README files for more details) contained in /examples in the repository.
Status
The Layercake stack is functional. Owing to the recent split of this project from the kraken framework, documentation and examples are in active development. Check back soon.