Abstract:

A Dockerfile builds a Docker image by specifying how to construct a software system by downloading, installing and configuring software packages and their dependencies. However, manually writing a Dockerfile can be error-prone because system dependency resolution requires a lot of domain knowledge. Therefore, an approach to automating Dockerfile generation based on domain knowledge was proposed. The approach automatically parses Dockerfiles and extracts knowledge of building Docker images and stores the knowledge in a graph database. When generating new Dockerfiles, the system dependencies and their installation operations for the designated software based on the knowledge base were inferred. Experiments indicate that it is viable to automate Dockerfile generation for diversified software by inferring system dependencies and software package installations with the domain knowledge.

Key words: Docker, Dockerfile, knowledge graph, software package, system dependency