Zero bug policy
Bugs are a tricky subject. We all write code with bugs and they have a considerable impact on our productivity and on the value added by our deliverables. So we come up with several strategies to handle the bug stream and have a ongoing effort to balance bug house keeping and new features.
I’m a big advocate for a zero bug policy. This means that we should usually have 0 registered bugs. Whenever I present this idea I’m met with skepticism. This is seen as an utopia and not possible. My impression is that developers interpret in a way that would generate punishment when new bugs are added.
But that is not the point. The point is to embrace that we’ll always have bugs, but also aim for a process that will minimize as much as possible the amount of bugs we produce. This will make us leave our comfort zone and question our beliefs.
It’s not about “writing code that never has bugs”. It’s more about “what do we need to change in our day to day work to minimize bugs”.
First we need to decide that we’re going to do something about the bugs. Then we can plan for it. But if we don’t have the team committed to this, it will be harder. Step by step we can make small changes that will help us in controlling the bug stream.
1. Always address the root cause
This is mostly common with bugs related with bad data. There was a bug that somehow created invalid data on the database. So the bug report focuses on fixing that data. We can go to the database and simply correct the data and the bug is solved.
But that won’t fix the root cause. Meaning that we’ll have more bugs related to the issue. We need to find and address the root cause first. This means time investment. And sometimes it’s not easy to find it and fix it, specially when we’re dealing with concurrency issues. But fixing the bug source once and for all should be our main objective.
2. Can’t solve it? Instrument
Sometimes we can’t find the root cause, because we lack context or additional information. We can try to contact the customers that had the bug and try to understand what they were doing at that time.
Another approach is simply to identify where the bug is happening and instrument the code. We can add more logs or enhance the error messages. So next time we have that bug we’ll have more material to help us fix it.
3. Really fix the bug
All patches that fix a bug should have a test that covers the bug. This regression test will make sure this doesn’t happen ever again, even if somehow we mess up somewhere else. These tests can be very hard to write. It’s an investment. It can be just a unit test or we may need to create a flow that launches the app and with side processes does some heavy loading looking for concurrency issues.
We should also add proper documentation to the test, explaining what was happening and what was our approach.
“I can’t recall ever getting a call to help a department or program get its application back on the rails and discovering a nice, healthy unit test suite.” - Interview with Erik Dietrich
4. Bugs on design phase are easier to solve
If we can detect bugs before starting development, that would be ideal. This is part of the process of aiming for zero bugs. We try to find strategies to prevent bugs and we try them out.
One strategy is to have a RFC development process and have it peer reviewed. Bonus points if the QA squad contributes to that review.
5. Transform bugs in features
We need to properly categorize bugs. If a bug fix is something big, or if the fix requires complex changes on the product, we should close the bug and create a proper feature to address it.
Going for zero bugs, means zero bugs impacting the users. It doesn’t mean solving all the product problems, and we need to separate those concepts. We’ll always have improvement points and problems to solve. But we can aim for zero bugs.
6. Retrospectives for bugs
If we are having a flow of specific bugs, if we delivered a new feature and the bugs spiked, we need to do a retrospective on what happened and how it could have been prevented. It’s an iterative process.
We may fail several times at it and it will question our processes. But it’s very important to learn about what we’re doing wrong and make it better.
7. Everyone is responsible for the quality
In some teams we have a QA squad that is responsible for testing the application. I don’t believe this is the best approach. I believe that all developers are responsible for delivering bug free code, and for testing it. QAs should be responsible for making sure that everything is okay, and that is not just tests. It’s much more.
Also, having the QAs doing tests usually creates a ping pong that is not productive:
- Developers create a feature and change code, submits it for testing
- The tester finds bugs, send the feature back
- Go back to 1)
This is not ideal and we need to be careful and try to minimize as much as possible this flow.
8. Elect a bug groomer
We should have a designated bug groomer. Someone that is responsible for reviewing all new bugs. The groomer will properly categorize them and maybe even convert them to features of improvements if that makes sense.
Usually, I advocate for the QA team to be encharged of the bug grooming. They have a wider context and ultimately they are responsible for making sure that the applicant is without bugs.
9. Beware of the dark support vibe
Doing support and being responsible for fixing bugs can be daunting. We may experience the frustration of our customers. Day after day dealing with problems and feeling bad by the impact we’re having on our customers can be a heavy burden. It will impact the developer’s satisfaction.
This is a reason to rotate the support developers. But ideally we’d be able to switch the focus of the support squad from fixing bugs to making improvements. Having zero bugs, will mean that all the time is allocated in making us perform better.
10. We should fix bugs right away
Sometimes we don’t have anyone allocated to bugs. We may receive bug reports and schedule them to be fixed on the next sprint. This will add a huge delay to the solution and can be troublesome. We should aim to fix the bugs right away and don’t let them stack.
11. Consider other testing methodologies
Having confidence in our test suite is very important. But there is always some bug on an edge case we’re not considering yet. We can be reactive and fix it with a regression test when it is registered. But we can also try to find those scenarios with other techniques.
We can see if we can take value from generative testing or mutation testing.
12. Take care of the test suite healthiness
If we have long builds it will impact our confidence and our productivity. We should aim for a maximum build time. This will depend on several factors. For some teams 20 minutes is too long. For other, 2 hours is acceptable.
We can create mechanisms to improve this, for example to detect what modules were changed and just test them and their dependencies. Having a 2 hour build running because we changed some markdown documentation is not that useful.
Flaky tests are also very problematic. We need to invest on fixing them, and even consider them bugs.
13. Embrace immutability and pure functions
Functional programming has several techniques that harden our code and make it more robust. Embracing immutability means that we’re removing the possibility of a full class of bugs.
Using pure functions allow us to create code that is easier to test. If the code is easier to test and to maintain, we may argue that we’ll have less bugs.
14. Beware of integrations
When we depend on a service from another squad, the integration is almost always troublesome. We have assumptions that are put to the test and fail almost every time. We should manage this risk and try to integrate as soon as possible, and maintain a healthy communication channel with the other squads.
We can focus on contract based testing, or start to plan right away integration tests.
15. Track runtime exceptions
Usually bugs are reported by the users. But we an be more proactive regarding that and register every runtime exception/error. There are several tools for that. Because many times users don’t complain at all and we end up not knowing all of our problems.
A great user experience related to bugs, is when a customer experiences a bug, having someone from our side contacting them: “We noticed that you experienced some problems with the application. We’ve adddressed and fixed them. Could you try again?”.
Let’s try to do this before they file a bug.
I was able to implement this policy a couple of times already. It’s something that usually takes some time. We need to change our habits and work towards that goal. It’s not easy and requires discipline and a plan.
Typical SaaS project
I worked on a project that had many quality issues. That resulted in several bugs being registered every day. The project also had an awesome support squad and mindset. Meaning that all bugs would somehow affect the support squad and the development team. At some time we had 2 developers doing full time support work. This was very frustrating work and no one wanted to do it.
So we rotated every two weeks. That gave us a nice balance between fixing things and building new things.
Some day we decided that we’d need to stop this. It made us reconsider all our quality practices. Just by deciding that we would have a controlled bug intake, we started to change our processes and way of work. Week by week we managed to decrease the bug flow to a point where we didn’t require a full time developer.
I eventually left this team, but when I met them for lunch a year later I was joking around with one of them and said: “Damn, for that you should go to the support duty another month”. And they replied: “Oh, we don’t have that anymore. We do have bugs from time to time but it’s very calm now”.
Deploys and downtime
On another project, it was common to the have application down. Deploys added downtime and we had some issues that made the app unresponsive and forced us to perform new deploys. This was impacting our delivery speed. The first step was starting to track the amount of times we were down and the amount of time. The data showed us that we had a problem.
We started to question our process and consider improvements to it. Sometimes it only took simple patches to make improvements. Other times we actually had to invest in new features and improvements to improve the healthiness of the application. All this was after we settled for a process with incident reports and postmortems, where we discussed what would be necessary to fix issue by issue for good.
After this hardening period, we managed to have 3 sprints without any downtime. But that was not it. When we started to have no downtime issues, we focused on the next level: SLAs for bugs.
The work is never done, but we can always aim higher.
Long lived branches
On another team I had, it was usual to have a very well defined quarter. We’d know exactly what we’d need to deliver on the next months. Usually the features would take at least one month to implement. In our flow, we would create a long lived branch per feature and each squad would work on their feature in isolation.
When the features were finished, we’d merge them upstream, and the QA squad would do their tests and quality flows.
The problem was that the upstream merge was always problematic. The first ones to merge were okay. But the other ones would have to deal with tons of conflicts. We’d actually need to invest days in doing the merge. And then more days to fix the bugs that were introduced by the merges.
This is actually not a success story.
I was unable to motivate the team to change our process. Sometimes we’re doing things for years and it’s hard to question what we do and aim to change. Sometimes we believe that what we’re doing is the best possible, even if it has flaws. And so we settle for it.
There are several changes we could do to minimize this specific problem. Maybe they wouldn’t solve everything, or maybe they would introduce new problems. But without that iteration process we wouldn’t be able to be better.
I don’t think that a zero bug policy is a silver bullet and that everything starts to work great and we start having bug-free code. But I do believe that if we decide to have no bugs, that our processes will be much better and that we will in fact experience less bugs. And that will give us time to build new things and to make improvements to our code base.
It’s a roller coaster of good practices that starts with a simple decision.
Discussion and references
Author Pedro Pereira Santos
License CC BY-NC-ND 4.0