Google Cloud Storage has an officially supported fuse client! This is something I have always wanted and would have expected for Google Drive, but 🤷. The only thing better than a fuse client is a fuse directory that gets mounted automatically when you log in, which you can do fairly simply using systemd --user, which is just systemd, but everthing runs as you. Here’s a gist of how I set this up on my machine:

I never realized the detail that the Google SRE book provided on the production environment at Google, including such things as D, Colossus, Jupiter, BNS, and GSLB. Also there’s now The Site Reliability Workbook, a hands-on implementation guide.

Meritocracy A system in which venture capitalists give large sums of money to 20-something male Stanford graduates in computer science.

Alerting in Prometheus is great, and easy, but one of the gotchas is that there’s no warnings or errors if you write an alert rule and there’s no data for that alert. Now Prometheus does have the absent() function, and you could tack OR absent(some_metric_name) == 1 to the end of every alert you write, but that’s tedious and error prone. So I wrote promk-absent a quick little tool on Go to create a set of absence alerts based on an existing set of alerts.

With Spectre and Meltdown we can see the inevitable results of trying to squeeze more and more performance out of a small number of cores. I believe the future is in processors with a high number of cores, such as RISC-V. With hundreds or thousands of cores on a single chip you can stop doing time-slicing and start doing core-slicing. That is, instead of doing preemtive multitasking, each process will be scheduled with a subset of the total number of core.

This Yale Law Journal Note from Lina M. Khan on Amazon’s Antitrust Paradox is an amazing read on several levels. First, the case it makes against Amazon is impressive, shining a light on quite a few unsavory things Amazon has done over the years. Second, she then expands that case to the rest of the tech giants. Third, for a Note in the Yale Law Journal, the entire thing is very readable.

Crowdsourcing, the idea of slapping together an algortihm that quietly works alone in the background on a large corpus of data to generate recommendations, most popular lists, etc. is slowly dying. I pointed this out in my 2018 Predictions post, but didn’t offer a concise explaination as to why that would occur, but as I’ve watched the year progress, and added more examples to that predicitions post, I’ve come up with a fairly simple model.

The kubectl command combined with jq is the Swiss Army Chainsaw of kubernetes management. Some of my more useful kubectl commands: watch 'kubectl get pods | column' # Delete all failed pods. kubectl get pods --field-selector=status.phase=Failed -o json | jq -r '.items[] .metadata.name' | xargs kubectl delete pod # All images running that have the word 'dirty' in their name. kubectl get pods -o json | jq -r '..|.containerStatuses?|select(.!=null)|.[].image' | sort | uniq | grep dirty

Firebase Hosting is amazing, so fast and easy to setup and they even take care of the SSL cert for you. I work for Google and didn’t even know we had this as a product, so just posting it here in case anyone else missed it.

Here are the slides for my OSCON presentation Machine Learning for Continuous Integration. As usual my slides are pretty worthless on their own, except for the links. Update (24-Sep-2018) Recording is now available:

As usual, XKCD nails it, our society is built under the basic premise that people aren’t sociopaths. This is the root of Facebook and Twitter’s issues, which is that they have built platforms that are being exploited by sociopaths, but they seem unwilling to act, always concerned about free speech. It is a particularly Silicon Valley conceit to not understand the difference. While Zuckerberg is busy enabling holocaust deniers:

A simple demo of k-means clustering. The little squares are the observations and the cirles are the centroids. Press the 'Step:' button to step through the algorithm. Restart const K = 3; const WIDTH = 500; const HEIGHT = 500; const RADIUS = 0.1; const STEP1 = 'Step 1: Assign observations to clusters.'; const STEP2 = 'Step 2: Update centroids.'; const COLORS = ['darkgreen', 'blue', 'darkorange', 'red', 'purple']; let ctx = document.

Both Pulito and VanillaJS have updated to webpack 4. Additionally both have moved from yarn to npm. While npm still isn’t as fast as yarn, the support for package-lock.json files and increased attention to security and speed are worth the tradeoff.

Remember the last time silicon valley was growing like crazy without any supervision? We’re still dealing with that mess.

Facebook’s Role in Data Misuse Sets Off Storms on Two Continents On March 18th, 2018 Libertarianism, aka the #Just-Trust-In-The-Giant-Corporations movement, died of a self-inflicted gun wound to the head. In lieu of flowers, please send regulators to look after your citizens privacy.

The End of Crowdsourcing With Russian bots on Twitter, the attack on Rotten Tomatoes scores, the deluge of fake comments sent to the FCC, and even attacks on a DARPA Network Challenge, I think 2018 is the beginning of the end for crowdsourcing. There are two problems to solve, the first is that some of these systems are trying to extract a signal from noisy data, and it only takes a small number of bad actors to intentionally inject a strong signal into that data stream and overwhelm the real signal.

Via Joel Spolsky, I learned that Facebook was pivoting away from engagement. Meanwhile Twitter continues to be Twitter, and Uber continues to be an ethical cesspool.

It was over three years ago that I wrote No more JS frameworks, at which time I was roundly criticized for not understanding that data binding could only be done via JS framework, the two were inextricably linked, and only 2-way data binding would do, as one way data binding was for weak-minded fools who weren’t building real applications. You can find the comments on HN yourself, I don’t link to that cesspool.

This article from Technology Review, How AI Can Keep Accelerating After Moore’s Law is a good follow-on from a previous article Moore’s Law Is Dead. Now What?. From the second article: Engineers have kept GPUs getting more powerful because they can be more specialized to the particular math they need to perform for graphics or machine learning, he says. In order to continue to squeeze more performance out of the same number of transistors and/or watts, we are going to need to get closer to the metal, and the metal is going to have to become more and more specialized, or at the very least, the metal has to stop looking like monolithic CPUs with a small number of cores.

The article Will Tech-Driven Deflation Export Japan’s Economic Woes to the World?reminded me of this video, "Evolution of the Desk": The origin for the above video is http://bestreviews.com/electronics#evolution-of-the-desk. Each time an object gets digitized that's one less object to manufacture, one less object to ship, one less object consuming raw materials. It's actually more surprising that this hasn't had an even larger impact on the U.

We just finished migrating all of our monitoring from InfluxDB to Prometheus and I thought I'd write up our reasons for the change. Please note that these are my own personal observations and relate to a specific project, these issue may not apply to you and you should evaluate each product for your own uses. Update: To clarify, the versions of InfluxDB and Prometheus that I am talking about are InfluxDB 1.