A Kubernetes 'Bridge' Operator between Cloud and External Resources

Abstract

Many scientific workflows require dedicated compute resources, including HPC clusters with optimized software, quantum resources, and dedicated hardware cluster systems like Ray, for example. At the same time, many scientific workflows today are built on Kubernetes leveraging growing support for workflow and support tools. To address the growing demand to support workflows on both cloud and dedicated compute resources we present the Bridge Operator, a software extension for container orchestration in Kubernetes which facilitates the submission and monitoring of long running processes on external systems which have their own cluster resources manager (SLURM, LSF, quantum services and Ray). The Bridge Operator consists of a custom Kubernetes controller that employs a Kubernetes Custom Resource Definition to manage applications. We present controller logic to manage the cloud container orchestration and external resource workload manager interface, a resource definition to submit HTTP/HTTPS requests to the external resource, and a controller pod communicating with the external resource manager to submit and manage job execution. The implementation allows us to mirror the external resource in Kubernetes pods, which allows the operator to use these pods as proxies to control the external system. The implementation is agnostic to the choice of resource manager but assumes the system exposes a HTTP/HTTPS API for its control/management. The Bridge Operator automates the role of a human operator running jobs on a black box external resource as part of a complex hybrid workflow on the Cloud.