Application Development Blog Posts
Learn and share on deeper, cross technology development topics such as integration and connectivity, automation, cloud extensibility, developing at scale, and security.
cancel
Showing results for 
Search instead for 
Did you mean: 
larshp
Active Contributor
7,570

Introduction


At the recent #DSAGTT20, SAP showed a few slides regarding in the area of continuous integration and ABAP:

https://twitter.com/PanzerDominik/status/1227274686750478338/photo/1

https://twitter.com/WachterSascha/status/1227254813542559745/photo/1

I did not attend the event, so the below might be out of context.

Continuous Integration


Continuous Integration(CI) is a word that can mean a lot of different things to different people, to me its something about providing feedback to the developer for each change, the feedback must be reliable and provided within a timely manner.

A change could be whenever the development is saved, or a commit, or something else, its really up to the development environment, which is a different topic. The developer needs the information to determine if its okay to push the change to a wider audience.

Also exactly how the development is deployed to the CI system is a deployment topic. The scenarios below can be set up using the classic transport system.
Elements of a CI pipeline might include:


  • Code Inspector / ATC

  • Coverage results

  • Unit tests

  • Integration tests

  • eCATT

  • UI tests

  • Performance tests

  • Metrics


Scenario 1 - Baseline


Let's take a common setup with development, test and production. The developers do development in the development system, the test system is used for manual testing.

Assumption: All organizations have a requirement for doing manual exploratory testing, even if all testing is automated, the QAS system cannot be scrapped.

Scenario 2 - Use QAS for CI


For each change done by the developers, the change is deployed to QAS and CI is run. When the results have been reported the change is reverted.

For ABAP systems its difficult to do a proper rollback of the objects. The unit and integration tests might change data in the system(they should not), so a full database rollback is required to have fully reliable results.

This means that the system cannot be used for manual testing at the same time, for each change the developers make the system is changed. Alternatively, specific time-slots are allocated for CI and some other for manual testing.

All-in-all this ends up with an automated process for breaking the QAS system, instead of avoiding errors via automation.

Scenario 3 - Add new CI system


So, to not disturb manual testing in the QAS system, the CI run can be moved to a different server. CI runs in CIS system, and manual testing is done in QAS,

Scenario 4 - Scalable CI


In scenario 3, assuming CI is run for each change run one by one.

And assuming each CI run takes 10 minutes, and each developer does 30 changes per day, with 2 developers = 60 changes = 600 minutes = 10 hours.

It should be possible to determine the feedback for the change while developers go get coffee, so more CIS systems are required. Typically hardware is cheaper than wages, so one CIS system per developer is an easy solution, but expensive?

Scenario 5 - Containerization


Instead of having a lot of idle systems, spawn a new CIS system using containerization and deploy the changes, run CI.

This way there will be a common baseline for everything, and no need to do any reverts, just spawning systems instead.

This can give reliable results in a timely manner, but I still have not seen anyone doing this setup yet, as its difficult to get everything running, and tough decisions have to be made to determine which baseline to use. If its a large container snapshot, hot standbys can be setup to respond fast.

Scenario 6 - Run inside DEV


Going in the other direction, setting up new infrastructure is time-consuming and expensive. How about running the CI inside the existing DEV system? This will sacrifice some reliability as the changes are not tested in isolation, but partner solutions exists that can help reduce this risk.

Scenario 7 - Multiple active versions


According to the slides, SAP has support for development and maintenance within the same system on the road-map.

This is possible today, but with multiple systems, so I assume the new innovation will be supporting this in a single system, and at the same time. So the ABAP kernel will have to support multiple active versions at a time. When(if?) this is implemented, the CI can run within the DEV system, just on the version with the changes done by the developer.

This, however, raises a lot of other questions regarding the maintenance of data, database isolation etc.


Guess this will never happen, instead what the slides refer to is probably a possibility to change between active versions. If there is more than one developer working in the area switched to development or maintenance on the same system, they will have to coordinate when to do one thing or another. I.e. it still makes sense to have separate development and maintenance systems.

Scenario 8 - One system per developer


As suggested by db48526ccb71414183a64aa46e31e784 back in 2016:

http://searchsap.techtarget.com/answer/How-can-I-use-Git-and-GitHub-for-SAP-software-development

http://searchsap.techtarget.com/tip/Implementing-modern-practices-in-an-ABAP-development-shop

Give one development system to each ABAP developer, this will help to do the development and unit tests in isolation. But still, CI should be run centrally on aligned infrastructure and configuration.

Scenario 9 - Front-load outside ABAP AS


Performing CI on ABAP application servers is required but might be an extensive setup.

For tasks like checking whitespace, a complete ABAP AS is not really required, multiple steps can be added in the CI pipeline, and once initial checks have been made, the ABAP CI run can be started.

abaplint is one free option that exists, it can run on Linux, so it hooks easily into GitHub Actions / GitLab Pipelines / whatever. If doing development in other languages than ABAP this is a typical setup for development, so it might be able to run on already existing infrastructure. Also note that this does not require any commercial licenses to run, so it can be used for open source development.

Results can be unreliable, but the setup is easy and cheap.

Conclusions


CI in ABAP is difficult, choose wisely among cost, reliability, speed, complexity etc. they have large consequences on the development process and infrastructure.

Fully supporting continuous integration is not just getting some results to the pipeline, like getting code into git is not devops.

ABAP containerization is required to fully support CI, without it it's not possible to fully support continuous integration.

17 Comments
Labels in this area