subscribe

Log4j, Faker and Black Swan Events

In December log4j, a library that’s used by a massive amount of projects had a major vulnerability, and in January the author of ‘Faker’ and ‘Color’ went nuclear, released a intentionally buggy version that broke a lot of projects (temporarily).

A lot has been said about both of these, but I wanted to add my own (very late) take to this.

I think the general theme of what people are writing about this is is: this is a major problem with open source. Critical projects aren’t getting enough funding.

In the case of Faker.js/Color.js, a lot of people in the Node.js community suggested to no longer specify ‘version ranges’ for dependencies but lock everything to a specific version in package.json.

I however found myself disagreeing with both of those conclusions.

For both of these major issues, I looked at the events unfolding and I have a hard time reaching the conclusion that there’s a big problem that needs to be solved. It was quite the opposite actually.

It’s hard to find exact statistics, but the sources I could find suggests that there’s at least 100,000s of open source developer as many open source packages. If we have couple of major issues like Log4j and Faker.js every year, that feels like a number that’s not really worth optimizing for.

Proprietary or well-funded software can also have major security bugs, and in each of these cases workarounds popped up fast and the issues were fully resolved within 2 weeks for most users.

Yes, the issues themselves were problematic and I have sympathy for the developers involved and how they were treated, but the conclusion I got from these events is that things are actually pretty robust and correct themselves in the face of major issues.

A counter example of the open source world, a report recently came out that Linux developers generally fix bugs faster compared to Apple, Google or Microsoft by a wide margin.

I don’t want to deny that that open source needs more funding. It’s a complex issue that I don’t know enough about, but If there are indeed major structural issues related to how open source is developed, we should look at widespread problems and not ‘black swan events’.

I’ve often seen organizations trying to find and fix root causes for any issue that pops up, but all these process changes can lead to a situation where it’s hard to get anything done. An important part of any post-mortem should be ‘how can we prevent this in the future’, but also ‘is this issue common enough to prevent it from happening again’.

Web mentions