| 摘要: | 隨著雲端運算技術的進展，虛擬化技術被廣泛應用於各種運算環境之中。而在近幾年來的IT產業裡，作業系統容器化(OS Containerization)的概念正逐漸被重視，這種概念就是在作業系統層級開啟虛擬而且獨立的作業環境，亦即是軟體容器(Software Container) ，並讓應用程式在軟體容器執行。這種方式與虛擬機器相較而言，具有較輕量的額外效能負擔，因此逐漸受到企業所青睞。然而，對於企業而言，服務的高可用性(High Availability, HA)永遠是其所注重的問題，因此需要一套機制來支援軟體容器的高可用性需求。在軟體容器平台的管理與建置工具中，Docker是目前最被廣泛重視與應用的開源專案。Docker Swarm則是一個用來建立Docker容器運算叢集的開源專案。而在我們的研究中發現，目前Docker Swarm的版本對於Docker容器的HA功能並未非常完善，例如對於以多個容器互相連結提供服務的容器叢集發生錯誤時，Docker Swarm無法確實回復服務。針對這些HA的需求，我們提出一個輕量型的HA架構，並於Docker Swarm上實作高可用性Docker容器叢集之功能。這個HA架構利用Docker容器共享數據卷(shared Docker volume)來確保資料不會在軟體容器復原後遺失。這個架構也使用Docker層疊網路(Overlay Network)動態配置容器位址讓軟體容器間可以跨主機彼此溝通。HA架構中最重要的是本研究所提出的容器錯誤監測與回復機制，讓軟體容器可以在發生錯誤後快速重啟，並且彌補了Docker Swarm 在容錯回復機制的缺陷。我們的實驗結果指出，本研究所提出的HA機制可以大幅降低Docker Container的服務停止時間(Downtime)，並提升以容器運行的應用程式或服務之可用性。;Hardware virtualization fulfills users’ computing demands by providing users software compute instances namely virtual machines, which emulate how physical machines work. It is widely used in various computing platforms in the cloud era due to its manageability, flexibility and elasticity. In recent years, a similar light-weight technology, usually called OS containerization (or OS-level virtualization), has gradually become a hot issue in IT industry. The OS containerization technology provides a logical compute instance, known as the software container, as the fundamental compute unit to host applications or services. A software container is emulated by the host OS. Therefore, it has light-weight overheads in terms of instance creation/deletion time and consumed memory space, when compared with hardware virtualization. Docker is one of the popular open-source projects for OS containerization. It allows people to create and to manage software containers on a host OS. Docker Swarm is another open-source project for building a compute pool of Docker containers. Based on our study, Docker Swarm does not well support the high availability (HA) feature for its Docker containers in the current version. For example, the failover mechanism of Docker Swarm cannot handle failures on linked containers, which can be viewed as a virtual container cluster where each member can directly or indirectly communicate with each other. To this end, this research focuses on the HA issue of Docker Swarm. In this paper, we present a light-weight HA architecture as well as its implementation based on Docker Swarm. The HA architecture uses shared Docker volume to prevent data loss. It also uses Docker overlay networks to enable communication among software containers on different hosts. We also implement a light-weight monitoring mechanism to detect whether a software container fails, as well as a recovery mechanism to recover a failed software container in a second. With the proposed mechanisms, we can reduce the downtime of any failed software container on Docker Swarm, and improve the ability of the applications/services running on the software containers consequently. |
