CloysterHPC is a software that guides the user to set up an HPC clustered environment. It asks for specific questions regarding the system to get an HPC cluster up and running as quick as possible. That's CloysterHPC.
Its goal is to enhance the installation and maintenance experience, making it user-friendly, creating an easy-to-use questionnaire built with a familiar user interface, for gathering and verifying the required cluster information. For experienced users, an unattended mode will be available in future releases with a classic configuration file.
Documentation about usage is not yet available, but the software should be self-explanatory since help information is baked directly into it.
CloysterHPC is written in modern C++ with some C code for compatibility with newt, that's used as the terminal user interface. The software follows the Object-Oriented Paradigm and implements a simplified architectural pattern based on MVP (Model-View-Presenter) where the components can be easily swapped or expanded in the future if needed.
- Localization support.
- Always running daemon to ease cluster maintenance.
- Static binaries.
- CloysterHPC requires boost C++ libraries.
- External fmt library is required while
std::formatisn't available in major compilers. - Logging is available through spdlog.
- magic_enum is used for static
enumreflections. - Adherence with best practices is done with gsl-lite.
- newt for Terminal UI.
- Perl for the libxcrypt, one of our subdependencies. The lib (and the one who depend on it) will be removed because this dependency caused a lot of issues
Only newt must be pre-installed for compilation. We don't ship it. Everything else should be found and installed by Conan during CMake. Our shipped build script should handle this automatic.
- CloysterHPC uses C++23 as the programming languages. We are not looking for memory and performance optimization here, we strive for correctness. The features that we want are a flexible software that can interact with the Operating System without much hassle. We don't prematurely optimize for resources, but we avoid unnecessary dynamic memory allocations even if this costs some system resources, which are pretty much abundant anyway on an HPC system manager (the headnode).
- The newt library is chosen since it seems to be the default library to do terminal composing on Linux systems, which is the goal here. We are basically developing this with Red Hat Enterprise Linux, and it's clones in mind, and newt is basically a Red Hat (and Debian) sponsored library. Finally, newt is easier to work than ncurses, so it seems like a good fit.
- We avoid OOP techniques that may add complexity without any visible benefits, such as multiple inheritance. Because in this context they are usually a bad idea. It's not forbidden to use, but it should be used with caution.
- At the current state there's no need for a database since the software is stateless, and in the future configuration will be stored in a single INI style or equivalent configuration file.
- Build system is based on CMake and Conan for package management. It was heavily based on cpp_starter_project with some modifications.
- Expansion and modularity is desired and should be achieved. A daemon
compatible with
systemdis planned to enable more powerful features like: keeping state, maintenance interface, common housekeeping routines, backups and upgrades. - We consume a lot of existing software to avoid recreating everything from scratch, most evidently: xCAT and OpenHPC. We also push changes on consumed projects to enhance them, so we directly benefit from those changes.
- CloysterHPC is not made to run inside a container. It needs bare metal access.
- Why this isn't written in the whatever language since it's more modern? -- Simply because I tend to write better C++ and C code than other languages, and the libraries and bindings that I need, and willing to use in the future, were promptly available in C++ or C. Also C++ is very good at managing an OS directly. We're aware that Python can do it too, but Python would create more issues than C++ in this case.
CloysterHPC is alpha quality software. Some features are still missing, but they are on the roadmap. Production use should be done with caution and with preliminar testing.
We recommend to build and run this software in a virtual machine due to its
nature and root execution requirements.
- EL8.9 or EL9.4 and higher based system with minimal package selection
- Half of system CPU cores as vCPU
- At least 4GB of RAM
- 50GB of Disk
- 2 Network Interfaces
- External Access
- Internal Management Network
- Disable side channel mitigations (for performance reasons)
- UEFI mode
To compile and run the software, clone this repository and execute the
setupDevEnvironment.sh shell
script.
All dependencies should be installed after its execution and then follow
with standard CMake procedure.
All default CMake targets are supported. By default, it builds a Release target that modifies the running OS, so be advised to run the code on a development system only.
Finally, to build the software just run the commands:
$ cmake -S . -B ./build
$ cmake --build ./build
Resulting binary will be available on /build/src directory in the root
directory of the project.
As a final warning: running the software without DUMMY option will probably damage the running OS if they run as root. Be advised.
You can run it by calling ./binary_name -t (TUI - Terminal User Interface) or
by using a answerfile and CLI (Command Line Interface) with
./binary_name -a /path/to/answerfile.
If you require any further help, please use ./binary_name -h (-h, --help).
To run tests you need to run ctest inside the build directory.
CloysterHPC is made available under the Apache License, Version 2.0.
Want to help? Feel free to report bugs, issues or open a discussion thread, so we can discuss features and enhancements. Pull requests are greatly appreciated.
Any feedback is welcome. Feel free to use it, report issues and complain about the software.