|Ned Batchelder : Blog | Code | Text | Site|
Painless bug triaging
» Home : Text
Created 3 May 2006
Triaging bugs is an important part of any development process. It's the simple but treacherous process of deciding what bugs should get fixed when. Simple because there's only one thing to decide: when should we fix this bug? Treacherous because it is tempting to turn the triage process into a long drawn-out affair with many people in a conference room. I've often seen triages like that, but it doesn't have to be.
The important thing about triaging is to do it efficiently. Too often I've seen triaging become a weekly half-day process involving more than five people. That's too much time spent by too many people, and it's too infrequent. If you can get triaging down to a quick decision process, it can be done daily by a few people, rather than be an all-team meeting that begins to feel like the opposite of progress.
Triaging doesn't have to be a complicated process. Yes, a simple process may be less accurate, but the expense of additional accuracy simply isn't worth it. Here's the simple triaging process.
Identify a number of milestones in your development. A milestone might be shipping 1.0, or releasing a public beta, or having something to demo to the board of directors. Maybe it's simply What We Want To Finish This Week. It doesn't really matter what the milestone is, so long as you name it and define it.
The definition of a milestone is a description of the milestone, with some tangible description of its goals. The goals will guide the triaging process, so try to make them concrete. Here are some examples of goals that help with triaging:
Set up your bug tracking system to allow you to assign bugs to milestones. If it doesn't have a field like this, you can hijack some other field, so long as everyone agrees what it means.
Create a milestone in your bug tracker for all of your foreseeable milestones. Make it so that the milestone values sort according to their date order.
Make the default for this field be "unknown". This is important: "unknown" milestone means that the bug has to be triaged, and a milestone chosen for it. People often want to default to the next milestone. This is bad, because it clogs the next milestone with all the most-recently found bugs. It also means that you can't tell the difference between bugs you decided to fix next, and bugs that were simply found recently. To keep the triage process efficient, you need to clearly identify which bugs need triaging. That's what the "unknown" milestone is for.
The default value for Milestone must be "unknown".
If you want developers to pounce on newly-found bugs, just instruct them to keep an eye on the "unknown" milestone category to get a jump on the triage process.
Similarly, it's a good idea to have a "someday" milestone. There's always bugs in the tracker that are going to be hard to get to. A "someday" milestone lets you keep those bugs, but not distract you from the near-term work at hand.
Getting the right mix of people to triage is important. A common mistake is to get everyone involved. That's too many! The simplest triage team is one person, and often this is enough. One intelligent person who clearly understands the goals of the milestones can make a pass through a list of bugs, and triage at least some of them. This can be a good way to make the process more efficient. Some bugs simply don't need more than one pair of eyes to make a decision.
Adding more people is helpful because different people bring different expertise, and simply because a pair of people will make it possible to discuss matters, and come to a decision that one person couldn't make by themselves. If you have two people triaging, the best pair are a customer-focused person, and a technology-focused person. This allows for each to advocate for their own constituents.
It's tempting to add more people. Everyone in the organization has a unique insight into the bugs, and everyone cares about the direction the product is taking. That doesn't mean they should all be present during triaging. Have a few people do the triaging. If they need information, they can ask for it. If they triage some bugs wrong, it's OK, it'll get fixed.
Adding people to the team is the single surest way to spend too many resources on bug triaging.
Once you have your milestones, and a few people with differing viewpoints on the triage team, triaging is now a simple process: Look at all the bugs in the "unknown" milestone, and pick a different milestone for them. It's not always an easy task, but if you remember that all you have to do is choose a milestone for each bug, it's much easier.
One common mistake that slows down triaging: trying to decide if the bug is really a bug. Here's a familiar conversation at a triage:
This is silly. What's the harm in triaging it? Alice is probably trying not to burden Joe with another bug in his queue, but her kindness has resulted in a distraction for Joe, and the rest of the triage team had to sit and wait while Joe came down and thought about it. What she should have done was put the bug in Joe's queue and let him figure out if it was already fixed when he got around to looking at the bug.
The triage team doesn't have to decide if bugs are really bugs. Another common tangent is doubting that the bug really happens, or under what circumstances. None of this can be determined efficiently in a triage meeting. If you have enough people to discuss bugs at this depth, then you have too many people, and if you have the right number of people, then you can't get to the bottom of questions like this.
The triage team should be answering this question:
Assuming this really is a bug, when do we want it fixed?
All the other side conversations about the nature of the bug, its current status, and so on, can be skipped and left to the developer when the bug comes to the front of the queue. It is necessary to discuss the details of the bug, for instance to determine how severe it is, or to assess how likely it is to affect the milestone's goals. But even some of those discussions can be avoided.
Rather than digging for an answer to a question during the triage, and then triaging based on it, you can make assumptions, and state them explicitly in the bug. Triage the bug pessimistically, explain the decision in the bug report, and let the developer give you the good news that you were wrong later.
For example, you could add this paragraph to a bug report:
The decision is made quickly, the bug is in a developers queue, he knows when he needs to look at it, and he has instructions about what to do. Move on.