Continuous System Test - Test Automation

Written by Kotaro Ogino and Francois Picalausa

Background of this entry:

We reported on our experience with automated system testing for improving development process of large and distributed search platform (システムテスト自動化による大規模分散検索プラットフォームの開発行程改善) at Japan Symposium on Software 'Tokyo 2014.

This entry is the English version of 継続的システムテストについて - kokotatata's blog which explains the background of continuous system testing presented at the symposium.

【JaSST'14 Tokyo】システムテストの自動化による大規模分散検索プラットフォームの開発工程改善 from Kotaro Ogino

Note: Currently the slides are only in Japanese. We hope we will have the opportunity to hold the presentation in English in the near future.

Background: Why continuous testing is important for Lean and DevOps.

Lean is a software development style which focuses on fast delivery of features of software to customers[1].To this end, one small requirement is implemented in each development cycle. Hence requirements analysis and development cycles are short, and value can be delivered rapidly to customers. Lead time is also decreased. Compared to waterfall, in which analysis, design, implementation and quality assurance are each tackled in one go, Lean allows to deliver value to customers more quickly.

On the other hand, a key factor in sucessful implementations of Lean and Agile development is software quality. For ensuring software quality in a Lean and Agile development, quality assurance activities should be conducted from earlier phases of a project[2]. In particular continuous testing helps verifying new features and preventing regressions.

DevOps is a culture where quick and fast feedback from customers to developers and analysts is valued. The realization of DevOps requires making a continuous improvement framework based on Agile practices[3]. Such practices include continuous integration (CI), continuous delivery (CD), and continuous monitoring. Continuous testing is not as well kown as CI and CD, but it can also help creating a safer continuous improvement cycle.

Objective: Why should we automate system tests?

The test automation pyramid is a representation of ROI and priority of automation for different test categories[4]. In general, unit tests and component tests lie at the bottom of the pyramid as they have high ROI. Integration tests, System tests, and acceptance tests which are a few levels above, should have lower ROI in theory, as quality is accumulated from lower levels. However, the effectiveness of test automation strongly depends on the context of the project, such as the culture of the team, and the characteristics of the software[5]. For instance, Ito Hiroyuki reports on a project which whose prioritisation of automation of acceptance tests led to high ROI [6].

Big Data related applications analyze large volumes of data with complex algorithm. In general, these applications are designed as distributed and collaborative. In addition, values for customers are not only depending on functionality of software but also on quality and size of the data.

Therefore, verification of cloud platforms characteristics including availability and operability and evaluation of both algorithm and data are important to assure customer's values. System testing is one of the places where those verification and evaluation can be conducted. Providing a continuous verification and evaluation environment for those test categories accelerate the continuous improvement cycle for both developers and analysts.

Benefits of continuous system test: reducing the time before a bug is fixed.

From our experience, continuous system tests allow to greatly reduce the time before bugs are fixed.

We identified two reasons for this. 1. We avoid big bang merges resulting from fixing system wide issues. 2. Developers can conduct system testing in their own timing.
We detail these two points next.

1. System tests can detect integration issues including specification and interface mismatch, incompatibility of versions and unexpected external specification changes. Continuous system testing aims to avoid encountering these issues at the very end of a project. Indeed, fixing these issues often implies having big bang merges that only makes the situation worse as the bugs are becoming more complicated and bigger in scope, which in turn means that investigation and fix take longer time.

Continuous system tests aim to reduce this problem by exercising the whole system in a continuous way. In particular, quick feedback is provided to developper by means of smoke tests which are quick, but usual test scenarios that the system should handle.

Using this technique, along with continuously merging source code changes allow developers to find out the integration issues as soon as possible, when their resolution is easier and quicker. This observation was also made by Henrik Kniberg in Lean from the Trenches (in its Japanese translation リーン開発の現場).

2. In addition, the cost of system test execution was very high so we needed to minimize the number of system tests that are run. For this reason, we run system tests only after a set number of bug fix and we verify them together. Since we automated these tests, our developers can verify the impact of their changes on their own schedule. The time frame for fixing bugs, from their discovery to the verification of the fix by automated system test, is shorter and the process becomes lean.

Evaluation: A reason for using "number of days for bug fix" for evaluation

For evaluating Software development practices and processes, evaluation metrics must allow comparing different practices and processes[7]. For evaluating test automation, coverage improvement, cost reduction, and number of days for bug fix can be used as metrics.

We have two important believes. First, tests themselves do not increase the quality of software, it can only be increased through analysis and development. Our second belief is that the objective of test automation is to accelerate the continuous improvement cycle of analyst and developers.

Of course, we calculated coverage improvement and cost reduction for internal project management purpose. However, to verify our assumption that is that continuous system testing accelerates the continuous improvement cycle, we chose the number of days to bug fix as our main evaluation metric.

Conclusion: Continuous system tests initiate continuous improvement.

In this report, we showed that continuous system tests is a first step towards improving development agility. In particular, it accelerates continuous improvement cycle of developers and analyst. Developers and analyst can aggressively push changes and improvments to the software safely, by verifying that their changes meet both functional and non-functional requirements using automated system test. In addition, automation of system test allows us to monitor and collect metrics for service improvement.

We hope to see more and more techniques to improve quality assurance in an agile way. Our journey to continuous improvement starts from now.