7

We are in the healthcare industry, where bugs can have an impact on people health. As a result, we started having a code freeze on the staging environment to have the QA test each build. In addition:

  • We don't deploy on Fridays so developers will not have to fix urgent bugs over the weekend
  • We do 2 weeks sprints
  • The QA asked to have the code on Thursday morning on staging for testing

As a result, code that wasn't completed by Thursday morning is not being released in that release. In other words, all the code developer code over Thursday-Friday isn't being delivered in the current sprint.

Because of that, I proposed to the stakeholders to consider separating the engineering sprint and the code delivery, setting the expected sprint delivery to a few days after the sprint ends. As expected, I got a lot of pushback from the stakeholders, telling me that they never heard of anything like it and that I break the Scrum process. Trouble is that I agree with them, but I can't think of any other way, apart from committing to less each sprint and having developers work on other things (support, next sprint tickets) in the last couple of days.

Anyone with a similar setup - how did you resolve it?

4

Having worked in two regulated environments (aerospace and pharmaceutical/healthcare), I've seen the problem of needing independent quality assurance in a deliverable.

A few things to consider:

Build QA time into your Sprint. My current organization currently runs two week Sprints. We begin on a Wednesday afternoon (some teams begin on Thursday morning) with Sprint Planning and have development through the next week's Friday. Code freeze is Friday afternoon. QA has at least 2.5 days (Monday after code freeze until Wednesday afternoon) of testing exclusively on a release candidate.

Involve QA in every step of the process. When your product management team is starting to develop ideas and requirements (stories, backlog items), get QA involved so they can provide feedback on testability. When you refine these requirements, get QA involved to review. If you are using acceptance criteria, have QA help to write them to make sure they are complete. When you are doing estimation, including QA effort in the estimation. When you plan your work, include QA to ensure that they are confident that the planned work can be done (developed, tested appropriately by development, and go through an appropriate QA process) within the Sprint.

Involve developers in the quality. In an environment where developers cannot test their work, they can still do valuable work after a code freeze. Bugs (or questions) may come up - the developers need to work with QA to fix issues and address concerns. Otherwise, they can improve automated test coverage to prevent regressions in future changes or improve developer tooling and infrastructure. If there's not much in improving quality, the developers can focus on risk reduction for upcoming work by learning and improving the skills or prototyping upcoming functionality.

You do have to make changes to Scrum to make it work in a regulated environment, but I would not break Scrum. Ensure that your increment is done (by development and quality assurance) at the end of the Sprint. Ensure that your Development Team includes all of the skills needed to deliver the work and everyone's estimation and effort is being considered.

  • In line with the idea of involving QA throughout the process, an interesting approach I've come across recently is to use conditions of satisfaction to break down PBIs into smaller chunks, on which the dev and the QA person can work in parallel. That way you can start testing (on parts of the PBI) before the item is completed. It would be a challenge to do this with independent QA, but it might still be feasible – Balala May 4 '18 at 0:15
  • Makes sense to me. Out QA and devs are working closely together and 99% of the bugs are found on dev boxed while running feature testing. The QA are helping whiting tickets and plan the testing as part of the backlog grooming. The main issue with our setup is that out 2 weeks sprints are Monday to Monday, if we'll change it to Wednesday to Wednesday it will all make sense. I think I'll go with this recommendation. Thanks! – Kuf May 4 '18 at 12:29
3

The way I see it your problem comes down to the definition of done. Does your DoD contain the requirement that your code has passed QA?

If so, then however long your QA needs to vet the latest build automatically becomes slack time for your devs at the end of the sprint. That's not necessarily a bad thing. You can do backlog refinement, research, learning, refactoring, etc in that time (if no bugs turn up). You just can't push out new builds. You can still deploy on the friday in the middle of the sprint. Bugs would simply get fixed when the devs are back in on monday. If this is how you want to roll you should however do minimize the time QA needs for their work. Your feedback cycle should be as short as possible.

If your DoD does not contain the requirement for QA testing then your proposed solution is the correct one: Your team will push out new builds whenever they have them. The last build occuring in the last days of the sprint whenever you are done. Whatever issues QA finds are then treated like new backlog items. They will be included into sprints according to their priority just like any other task. If QA finds a particularly critical bug then the PO can renegotiate the scope of your current sprint to include that issue (generally at the expense of something else).

  • While I agree this approach makes the most sense from scrum planning side, I will reduce the capacity of each sprint by 20% (2 days out of 10). I do like how it keeps the scrum notion intact and will try to sell this to the stakeholders. – Kuf May 4 '18 at 12:19
1

Analysis

When it comes to managing the release cadence, a business really only has three choices, regardless of industry or framework:

  1. Fully integrate QA with development.
  2. Accept that either developers or testers will be idle some part of each Sprint when practicing Scrummerfall.
  3. Decouple releases from delivery or deployment (more on that below).

While high-compliance environments aren't orthogonal to agility, the demands of the compliance requirements can sometimes force a business to face the fact that you can't have disparate processes that move at different speeds run in lockstep. High-compliance environments also tend to fall prey to the 100% utilization fallacy, so the idea of having sufficient slack in the process to allow disparate processes to deploy on the same cadence is unrealistic.

Decoupling, when supported by the right engineering and process controls, is usually the sanity-preserving way out.

Differentiate "Delivery" from Deployments and Releases

While it's generally best to integrate QA directly into your Sprints, a lot of businesses overlook the value of decoupling delivery and deployment from release. While Scrum mandates that the output of a Sprint is a potentially-shippable increment, it doesn't actually require that the increment be released to end-users.

In a regulated environment, agile teams are still responsible for delivering increments that meet the Definition of Done. However, downstream processes such as third-party testing or production releases do not have to move in lock-step with the development team.

The main issues you are likely to face with decoupling deliveries from your other processes are: the need to negotiate bug-fixing with the business, and the assignment of responsibility for releases outside of the Scrum team.

In canonical Scrum, bugs found outside the development cycle must re-enter the Product Backlog. The business contracts with the Scrum team not to bypass the backlog and estimation process, although early terminations (costly as they are) certainly remain an option.

Furthermore, a truly decoupled release plan means QA or some other team will own the release. Whether the business tries (and likely fails) to have every delivered increment tested and released, or whether they more sensibly give the release teams the authority to batch up deliveries and set their own cadence for tagging and deploying releases is strictly a management decision.

There are certainly variations on these themes, but the core tenet is that each Sprint must maintain its integrity against unplanned work, and that any work moved outside the team must abide by that. So long as that's true, decoupling releases is the approach I advocate in healthcare and government settings.

0

Can't QA be integrated in your scrum team? This not only means you integrate one person that does QA, but you spread the QA knowledge in your team.

The main goal building a scrum team, is that the team is able to build (or even better deploy) a product increment on its own. Further there are no other roles in the team than developer. So your team can have a QA specialist, but all other developers should be able to support in QA. The other way around your QA specialist should be able to support development. Perhaps, to accomplish this, you have to manage the knowledge transfer.

This way you could integrate QA tasks completely in your sprint planning. Anyone in your team could handle or at least support any type of task, development and QA. At any time in the sprint you can do QA for finished features. At any time in the sprint you can react on QAs feedback on development.

Perhaps its also worthy to define what you mean by QA. Is it really quality assurance or more like user/stakeholder acceptance testing?

  • 2
    Further there are no other roles in the team than developer. So your team can have a QA specialist, but all other developers should be able to support in QA. In regulated environments, such as healthcare, this is difficult. The formal testing of a deliverable must be done by someone who did not participate in the design & development of the feature. Having independent QA assess the quality of requirements and/or acceptance criteria, write test cases against those requirements and acceptance criteria, implement automated acceptance tests, and run manual test cases is important for compliance. – Thomas Owens May 2 '18 at 9:26
  • I don't think I'll adopt this approach. developers are doing unit testing and test their code on the dev box, but I need the QA, who are much better in breaking stuff, to do the second round but also test the new 'build' in integration and regression tests. while developers can do that too, I feel that will be time-consuming for the biggest bottleneck on the eng team, so if I can offload some of this to a QA I rather do that. – Kuf May 4 '18 at 12:24
  • What happens if QA finds something that prevents the integration? It goes back to the DEV team and they fix it and then it goes back to the QA Team? Does the whole increment go back? I don't know these highly regulated environments. As I wrote you would integrate QA knowledge and manpower to the team. This way you probably could open the bottleneck a little. – Seb May 4 '18 at 13:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.