Build and Deployment Process for Web Applications


and any build failures were immediately resolved. The build and deployment was then performed on the QA server. Once the failure was resolved, the system was not always clean-rebuilt immediately.

Refinement 1—Basics:
The Lserve moved to an XML based script using Ant as the build tool. The script deleted the staging and deployment directories on the target server, ported the code over to the build area, produced the binaries and finally copied the binaries and the utilities to the deployment area. Any server configuration and properties were manually modified. The target server was “hard-coded” in the XML file and thus had to be changed for each server. The developers still manually built and deployed the new code and debugged it.

Refinement 2—Automation:
While Mserve was still using the manual process, the LServe started to automate the build process. First the script was modified to handle any changes in the server configuration and the properties. The second enhancement to the script was to specify the server name on the command line. At this point the build process was fairly stable however, the developers were still manually building and testing their code changes.

Refinement 3—We became diligent:
At this point NServer was actively contributing to the application. The focus shifted to more automation and providing a universal build script for the developers and the build miester. The plan to achieve it is illustrated below.

Each developer was provided a build template which she could configure to her PC environment. Both LServe and NServe had e-Service build files and produced jar files. The final application was built and deployed by the build script on the target server.

This provided a reliable build process to build the e-Services and the applications.

Build Problems and Remedies:

Multiple 3rd Party Utility Jar Versions:
This was a major and ongoing problem. The developers freely downloaded jars often causing version inconsistency among the services and the application. There were multiple occurances of the same component in the directory structure. This causes build problems, especially when there are multiple versions of the same component. To remedy this problem, the teams emphasized use of a single version. Domain experts were assigned to each utility and were responsible to determine the adequate version for use. The problems caused by incorrect sequencing of utility jars were corrected by proper education and communication.

System Configuration and Property Files:
The development team made changes to the development server configurations via .profile file. In the beginning deleting a property file was considered unsafe. Instead, it was decided to manually update the property files. At times these changes were not communicated to the build miester resulting in build failures. This problem was overcome by establishing better communication between the build miester and the developers and later by progressive build automation.

Human Errors and Limitations:
In rare instances the developers forgot to verify their changes. Also, some directories were moved around on the development server. The target server was not updated accordingly with the changes causing build failures. The team had very limited experience with the build tool. This was overcome by improved communication. As the build miester gained more experience, the situation improved.

Hard-coded String References:
At times the developers used hard-coded strings to specify a particular server name. Because the target server was different, it caused build failures. This was not a build problem and was overcome by modifying configuration and properties files. Several copies of the configuration were maintained, one for each deployment environment. Eventually, the strings which were moved to configuration and properties files were created as part of the build process. The

About the author

Bhushan Gupta's picture Bhushan Gupta

Bhushan Gupta has nineteen years of experience in software engineering, nine of which have been in the software industry. Currently a Process Engineering Architect at Hewlett-Packard, he joined the company as a software quality engineer in 1997 where he was responsible for identifying key process areas to reduce rework. Based upon Hewlett-Packard's quality parameters, he has developed quality goals, quality plans, and software validation strategies for several products. In his current position he has promoted the iterative lifecycle, contributed to software validation planning and execution processes, carried out project retrospectives, and software process improvements. He has worked on use cases and their adoption for development of requirements and test cases. From 1995-97 he worked as a Systems Analyst at Consolidated Freightways where he contributed to the design and development of a Windows based logistic management system. During his ten years tenure (1985-1995) at the Oregon Institute of Technology Bhushan Gupta developed, planned, and taught numerous software engineering courses. He served as the curriculum coordinator for five years and resigned from his duties as an associate professor.

AgileConnection is one of the growing communities of the TechWell network.

Featuring fresh, insightful stories, is the place to go for what is happening in software development and delivery.  Join the conversation now!