Hacker News Reader

Comment by Sebb767 | Hacker News Reader

I think you're taking the 5 whys to literally. The plan is not to ask why five times and replace the issue text with the fifth answer, but to take a moment to inspect what actually caused the issue instead of directly trying to fix the symptoms.

Sure, there's a chance that there is no root cause (although, at least in my experience, there usually is a major contributor) or that it's something that's not easy to fix right now, but at least is shows you what issues are brewing below and what you need to put on your roadmap. And, in the best case, you might actually be able to fix a bug higher up the chain directly and resolve future issues.

View on HN

When we used 5Ys at Uber SRE I always like to say that you are looking for a tree of depth 5 but if it's width 1 you've not tried hard enough. "Root cause" or at least the idea there was a single factor that was the defining feature of an issue is hardly ever true.

There is quite often one symptom that is pivotal and cause a critically failure, but generally that symptom itself is a result of multiple inputs.

I'd simply say 5Ys should be a rule of thumb to get folks to be sufficiently inquisitive to keep asking follow-up questions about whatever they are investigating until they've hit around 5 questions per area.

There is also the concept of ownership, that you are really in total ownership of any situation.

I think the 5 why's should arrive to what part in your circle of ownership does the issue lie.

Hard disagree, I would rather have multiple clear and singular RCAs than any 5Y exercise that is >1 width. Too much context to follow to accurately drill into each cause per an RCA, from my experience.

You end up looking for random causes to increase your tree width, instead of the actual analysis at hand.

As long as you prune branches that don't have action items, what's the problem?

5 why's presents a list of action items, and a justification for them. One action item could depend on two leaf nodes of the RCA, or solve multiple leaf nodes.

Reply on news.ycombinator.com