Recreation of Containers for High Availability Architecture and Container-Based Applications

Over the past few years containers have become a very popular solution for virtualization purposes. They have all major advantages of virtual machines and, additionally, reduced hardware and time requirements and built-in support for detecting faults in containers and recreating the ones which are not working. Docker is an open platform which enables running applications in a loosely connected environment called a container. A Docker Swarm framework may be used on a Docker platform to cluster and schedule containers. The time needed for detecting and recreating failed containers was tested in this paper. The time needed to recreate a given container depends on the size of the container and the reason of its failure. Docker Swarm needs from 8 to 31 s to recreate a failed container, however, for some purposes, e.g. e-commerce systems, it is too long. In such systems the administrators attempt to minimize the time of website unavailability. To mitigate this problem, we propose the architecture of a lightweight system for fast monitoring of containers and scheduling their recreation. The prototype based on the proposed architecture is insensitive to a container image size, and recreation times depend only on the reason of a container failure. The recreation times range from 3 to 10 s. If we omit the time needed to create containers, which depends solely on internal Docker Engine implementation, our solution detects a container failure and schedules the recreation of a container instance almost instantaneously. This paper presents a part of the holistic solution of fast and lightweight container creation, monitoring and responding to failures.