Skip to content

Application Installation Guide

Administrators can install and manage applications in TKFE in the following ways:

Install Spack applications

The recommended method of application installation is via Spack. Spack, an established package management system for HPC, contains build recipes of the most widely used open-source HPC applications. This method is completed automated. Spack installation is performed as a Slurm job. Simply choose a Slurm partition to run Spack.

Advanced users may also customise the installation by specifying a Spack spec string.

Install custom applications

For applications not yet covered by the Spack package repository, e.g., codes developed in-house, or those failed to build by Spack, use custom installations by specifying custom scripts containing steps to build the applications.

Register manually installed applications

Complex packages, such as some commercial applications that may require special steps to set up, can be installed manually on the cluster's shared filesystem. Once done, they can be registered with the FrontEnd so that future job submissions can be automated through the FrontEnd.

Application status

Clicking the Applications item in the main menu leads to the application list page which displays all existing application installations. Applications can be in different states and their Actions menus adapt to this information to show different actions:

  • n - Application is being newly configured by an admin user through the web interface. At this stage, only a database record exists in the system. The user is free to edit this application, although in the case of a Spack application , most information is automatically populated. When ready, clicking Spack Install from the Actions menu to initiate the installation process.
  • p - Application is being prepared. In this state, application build is triggered from the web interface and information is being passed to the cluster.
  • q - In this state the Slurm job for building this application is queueing on the target cluster. Note that all application installations are performed on a compute node. This leaves the relatively lightweight controller and login nodes to handle management tasks only, and also ensures the maximum possible compatibility between the generated binary and the hardware to run it in the future.
  • i - In this state, the Slurm job for building this application is running on the target cluster. Spack is fully responsible for building this application and managing its dependencies.
  • r - Spack build has completed successfully, and the application is ready to run by authorised users on the target cluster.
  • e - Spack has somehow failed to build this application. Refer to the debugging section of this document on how to debug a failed installation.
  • x - If a cluster has been destroyed, all applications on this cluster will be marked in this status. Destroying a cluster won’t affect the application and job records stored in the database.

A visual indication is shown on the website for any application installation in progress. Also, the relevant web pages will refresh every 15 seconds to pick status changes.

Install a Spack application

A typical workflow for installing a new Spack application is as follows:

  1. From the application list page, press the New application button. In the next form, select the target cluster and choose Spack installation.
  2. In the Create a new Spack application form, type a keyword in the Create a new Spack application form, and use the auto-completion function to choose the Spack package to install. The Name and Version fields are populated automatically. If Spack supports multiple versions of the application, click the drop-down list there to select the desired version.
  3. Spack supports variants - applications built with customised compile-time options. These may be special compiler flags or optional features that must be switched on manually. Advanced users may supply additional specs using the optional Spack spec field. - For a guide to the Spack spec syntax see the Spack documentation - By default, the GCC 11.2 compiler is used for building all applications. - Other compilers may be specified with the % compiler specifier and an optional version number using the @ version specifier (e.g., %intel@19.1.1.217). Obviously, admin users are responsible for installing and configuring those additional compilers and, if applicable, arrange their licenses. - Spack is configured in this system to use Intel MPI to build application. - Other MPI libraries may be specified with the ^ dependency specifier and an optional version number.
  4. The Description field is populated automatically from the information found in the Spack repository.
  5. Choose an Slurm partition from the drop-down list to run the Slurm job for application installation. This, typically, should be the same partitions to run the application in the future.
  6. Click the Save button. A database record is then created for this application in the system. On the next page, click the Edit button to modify the application settings; click the Delete button to delete this record if desired; click the Spack install button to actually start building this application on the cluster. The last step can take quite a while to complete depending on the application. A visual indication is given on the related web pages until the installation Slurm job is completed.
  7. A successfully installed application will have its status updated to ‘ready’. A New Job button becomes available from the Actions menu on the application list page, or from the application detail page. The User Guide contains additional information on how jobs can be prepared and submitted.

Application problems

Spack installation is fairly reliable. However, there are thousands of packages in the Spack repository and packages are not always tested on all systems. If a Spack installation returns an error, first locate the Spack logs by clicking the View Logs button from the application detail page. Then identify from the Installation Error Log the root cause of the problem.

Spack installation problems can happen with not only the package installed, but also its dependencies. There is no general way to debug Spack compilation problems. It may be helpful submit an interactive job to the cluster and debug Spack problems there manually. It is recommended to not build applications from the controller or login nodes, as the underlying processor may differ to the compute nodes.

Complex bugs should be reported to Spack. If an easy fix can be found, note the procedure. This can be then used in a custom installation.