The 9th installment of the Friday Blast series. A slight focus on physics for this week.

What is a field really? (2011) - a Physics StackExchange question. My physics education leaves much to be desired. And I’ve always been confused by the concept of field. Turns out it’s quite simple - it is just a way of describing some physical phenomenon which occurs over every point in space and varies with time. So basically a function which can be scalar, vectorial, tensorial etc. The examples with temperature and wind from the second answer are especially instructive, since they’re easier to relate to than electromagnetic fields.

Intro to SDR and RF signal analysis (2017) - this article prompted me to explore the previous question about fields actually. The most interesting parts the reverse-engineering of various wireless devices’ communication with GNU radio tools.

The fundamental limits of knowledge (2015) - the discussion here was way over my head. I had expected it to be about things that can be proved or not, or things like that. But it turns out that there’s a bunch of stuff about the Big Bang and the very earliest moments of our Universe, which we can’t ever know, because that information was permanently lost. Interesting stuff.

The microservices cargo cult (2015) - when microservices became a thing 2 or 3 years ago, everybody and their grandmother seemed to be caught in the hype. Things have calmed down a bit in the meantime[1]. This article was written during the height of the hype period and reminded people that there’s some pretty hefty operational costs to running a microservices system and people shouldn’t employ this architecture lightly. Things have improved a little bit, in the sense that there are a bunch of mature monitoring, logging etc. providers right now which can ease building such systems from the get-go rather than when you reach operational maturity. But still, sage advice.

Lessons learned writing highly available code (2015) - we’ve had these sorts of tips here on Friday Blast before, but they’re always welcome. In short: have timeouts for RPCs, have exponential backoffs for retries, have multiple levels of health checks and watchdogs and generally limit everything that can come from a user (data requests, open connections, timeouts etc). Perhaps the sagest advice is to use battle tested tools rather than the new hotness.

Using atomic transactions to power an idempotent API (2017) - an example of building a small user database for a webapp and ensuring emails are unique via transactions alone. It’s a nice example of the benefits of using the SERIALIZABLE isolation level of a database. Unfortunately not many people are aware of isolation levels and the ability to tune-them by transaction. A small nitpick - they’re using isolation rather than atomicity from ACID here. Isolation ensures that transactions execute without stepping on each other’s toes - as if each was executing alone on the DB. Atomicity just ensures that the transaction either succeeds fully or fails fully, but doesn’t leave the DB in an inconsistent state. There’s also a nice discussion about getting events data out of the DB to other systems via a periodic task rather than a perilous double write.

The unreasonable effectiveness of Random Forests (2015) - there’s more to machine learning than Neural Networks. Shocking right? After linear models, a workhorse of the field is tree models and their Random Forest realisation. There isn’t anything super technical here, but there’s a good overview of their pros and cons. They’re not the best choice when you’re dealing with signal data, but for other modalities I think they’re often overlooked for data hungry and resource intensive neural networks. And that’s a shame, because they’re quite nice models and they have a good theoretical background.

[1] We now have serverless though to take the heat.