JConf Mexico 2020

October 3, 2020 All day

Continuous Stupidities – did DevOps fail?

Heavy motivation to automate everything, even the automation itself, is the common understanding of the most DevOps teams. A strong dominance of administrators in those teams (Operations), transform ideas of increasing the development process to a personal playground. Far away from the necessities of the DEV team, which have to create the application. Often Developers have to fight heavily with the provided infrastructure, because this break productivity massively down. In this talk we resume the lessons learned after more than one decade as configuration manager in different enterprise projects, to discuss nonsense DevOps workflows.

Modern Times (for Configuration Manager)

Heavy motivation to automate everything, even the automation itself, is the common understanding of the most DevOps teams. There seems to be a dire necessity to automate everything – even automation itself. This is common understanding and therefore motivation for most DevOps teams. Let’s have a look on typical Continuous Stupidities during a transformation from a pure Configuration Management to DevOps Engineer.

In my role as Configuration and Release Manager, I saw in close to every project I joined, gaps in the build structure or in the software architecture, I had to fix by optimizing the build jobs. But often you can’t fix symptoms like long running build scripts with just a few clicks. In his post I will give brief introduction about common problems in software projects, you need to overcome before you really think about implementing a DevOps culture.

  1. Build logic can’t fix a broken architecture. A huge amount of SCM merging conflicts occur, because of missing encapsulation of business logic. A function which is spread through many modules or services have a high likelihood that a file will be touched by more than one developer.
  2. The necessity of orchestrated builds is a hint of architectural problems.Transitive dependencies, missing encapsulation and a heavy dependency chain are typical reasons to run into the chicken and egg problem. Design your artifacts as much as possible independent.
  3. Build logic have developed by Developers, not by Administrators. Persons which focused in Operations have different concepts to maintain artifact builds, than a software developer. A good anti pattern example of a build structure is webMethofs of Software AG. They don‘ t provide a repository server like Sonatype Nexus to share dependencies. The build always point to the dependencies inside a webMethods installation. This practice violate the basic idea of build automation, which mentioned in the book book ‚Practices of an Agile Developer‘ from 2006.
  4. Not everything at once. Split up the build jobs to specific goals, like create artifact, run acceptance tests, create API documentation and generate reports. If one of the last steps fail you don’t need to repeat everything. The execution time of the build get dramatically reduced and it is easier to maintain the build infrastructure.
  5. Don’t give to much flexibility to your build infrastructure. This point is strongly related to the first topic I explains. When a build manager have less discipline he will create extremely complex scripts nobody is able to understand. The JavaScript task runner Grunt is a example how a build logic can get messy and unreadable. This is one of the reason, why my favorite build tool for Java projects is always decided to Maven, because it takes governance of understandable builds.
  6. There is no requirement to automate the automation. By definition have complex automation levels higher costs than simple tasks. Always think before, about the benefits you get of your automation activities to see if it make sens to spend time and money for it.
  7. We do what we can, but can we what we do? Or in the words by Gardy Bloch „A fool with a tool is still a fool“. Understand the requirements of your project and decide based on that which tool you choose. If you don’t have the resources even the most professional solution can not support you. If you understood your problem you are be able to learn new professional advanced processes.
  8. Build logic have run first on the local development environment. If your build runs not on your local development machine than don’t call it build logic. It is just a hack. Build logic have to be platform and IDE independent.
  9. Don’t mix up source repositories. The organization of the sources into several folders inside a huge directory, creates just a complex build whiteout any flexibility. Sources should structured by technology or separate independent modules.

Many of the point I mentioned can understood by comparing the current Situation in almost every project. The solution to fix the things in a healthy manner is in the most cases not that complicated. It needs just a bit of attention and well planning. The most important advice I can give is follow the KISS principle. Keep it simple, stupid. This means follow as much as possible the standard process without modifications. You don’t need to reinvent the wheel. There are reasons why a standard becomes to a standard. Here is a short plan you can follow.

  • First: understand the problem.
  • Second: investigate about a standard solution for the process.
  • Third: develop a plan to apply the solution in the existing process landscape. This implies to kick out tools which not support standard processes.

If you follow step by step you own pan, without jumping to more far ten the ext point, you can see quite fast positive results.

By the way. If you think you like to have a guiding to reach a success DevOps process, don’t hesitate to contact me. I offer hands on Consulting and also training to build up a powerful DevOps team.


InPerson (remote) Training’s (Courses) Deutsch / English

Automation options in software configuration management

Software development offers some extremely efficient ways to simplify recurring tasks through automation. The elimination of tedious, repetitive, monotonous tasks and a resulting reduction in the frequency of errors in the development process are by no means all facets of this topic.

(c) 2011 Marco Schulz, Materna Monitor, Ausgabe 1, S.32-34
Original article translated from Deutsch

The motivation for establishing automation in the IT landscape is largely the same. Recurring tasks are to be simplified and solved by machine without human intervention. The advantages are fewer errors in the use of IT systems, which in turn reduces costs. As simple and advantageous as the idea of processes running independently sounds, implementation is less trivial. It quickly becomes clear that for every identified possibility of automation, implementation is not always feasible. Here, too, the principle applies: the more complex a problem is, the more complex its solution will be.

In order to weigh up whether the economic effort to introduce certain automatisms is worthwhile, the costs of a manual solution must be multiplied by the factor of the frequency of this work to be repeated. These costs must be set against the expenses for the development and operation of the automated solution. On the basis of this comparison, it quickly becomes clear whether a company should implement the envisaged improvement.

Tools support the development process

Especially in the development of software projects, there is considerable potential for optimization through automatic processes. Here, developers are supported by a wide range of tools that need to be skillfully orchestrated. Configuration and release management in particular deals in great detail with the practical use of a wide variety of tools for automating the software development process.

The existence of a separate build logic, for example in the form of a simple shell script, is already a good approach, but it does not always lead to the desired results. Platform-independent solutions are necessary for such cases, since development is very likely to take place in a heterogeneous environment. An isolated solution always means increased adaptation and maintenance effort. Finally, automation efforts should simplify existing processes. Current build tools such as Maven and Ant take advantage of this platform independence. Both tools encapsulate the entire build logic in separate XML files. Since XML has already established itself as a standard in software development, the learning curve is steeper than with rudimentary solutions.

Die Nutzung zentraler Build-Logiken bildet die Grundlage für weitere Automatismen während der Entwicklungsarbeit. Einen Aspekt bilden dabei automatisierte Tests in Form von UnitTests in einer Continuous-Integration-(CI)-Umgebung. Eine CI-Lösung fügt alle Teile einer Software zu einem Ganzen zusammen und arbeitet alle definierten Testfälle ab. Konnte die Software nicht gebaut werden oder ist ein Test fehlgeschlagen, wird der Entwickler per E-Mail benachrichtigt, um den Fehler schnell zu beheben. Moderne CI-Server werden gegen ein Versionsverwaltungssystem, wie beispielsweise Subversion oder Git, konfiguriert. Das bewirkt, dass der Server ein Build erst dann beginnt, wenn auch tatsächlich Änderungen im Sourcecode gemacht wurden.

Complex software systems usually use dependencies on external components (libraries) that cannot be influenced by the project itself. The efficient management of the artifacts used in the project is the main strength of the build tool Maven, which has contributed to its strong distribution. When used properly, this eliminates the need to archive binary program parts within the version control system, resulting in smaller repositories and shorter commit times (successful completion of a transaction). New versions of the libraries used can be incorporated and tried out more quickly without error-prone manual copy actions. Libraries developed in-house can be easily distributed in a protected manner in the company network with the use of an own repository server (Apache Nexus) in the sense of reuse.

When evaluating a build tool, the possibility of reporting should not be neglected. Automated monitoring of code quality using metrics, for example through the Checkstyle tool, is an excellent tool for project management to realistically assess the current status of the project.

Not too much new technology

With all the possibilities for automating processes, several paths can be taken. It is not uncommon for development teams to have long discussions about which tool is particularly suitable for the current project. This question is difficult to answer in general terms, since every project is unique and the advantages and disadvantages of different tools must be compared with the project requirements.

In practical use, the limitation to a maximum of two novel technologies in the project has proven successful. Whether a tool is suitable is also determined by whether people with the appropriate know-how are available in the company. A good solution is a list approved by management with recommendations of the tools that are already in use or can be integrated into the existing system landscape. This ensures that the tools used remain clear and manageable.

Projects that run for many years must undergo modernization of the technologies used at longer intervals. In this context, suitable times must be found to migrate to the new technology with as little effort as possible. Sensible dates to switch to a newer technology are, for example, a change to a new major release of the own project. This procedure allows a clean separation without having to migrate old project statuses to the new technology. In many cases, this is not easily possible.

Conclusion

The use of automatisms for software development can, if used wisely, energetically support the achievement of the project goal. As with all things, excessive use is associated with some risks. The infrastructure used must remain understandable and controllable despite all the technologization, so that project work does not come to a standstill in the event of system failures.