Earlier this month Monica Lent published a list of Tech talks that’ll change how you think. Every one on this list is awesome and some have influenced me pretty heavily. Especially Simple made Easy by Rich Hickey. To these I’d like to add a few from my list of favorite talks.
Full Stack Awareness Link to heading
In this, Arthur Bergman rants about how everything is broken. We, as engineers, are so used to just connect a few systems and the new feature is working. We take so many things for granted that we forget how underlying systems down to the operating system work.
Stop Rate Limiting! Capacity Management Done Right Link to heading
A very nice introduction into Little’s Law and how often times rate limiting your backend is actually a bad idea. When protecting your backend system usually you limit the RPS per customer. Initially you negotiate the rate limits during onboarding and then forget about the exact numbers. In normal operations you then discover that client A needs more RPS while client B needs less. So you limit requests for client A even though you have spare capacity.
There is also a nice implementation of the algorithm in Elasticsearch called Adaptive Replica Selection that we are using in production for a while now.
How NOT to Measure Latency Link to heading
Measuring and reporting latency is commonly based on percentiles. The problem here is that we are happy with reporting this one data point for our system. In today’s Microservices architectures this is problematic though. A real user’s request usually triggers multiple calls to our backend systems. But that means the chances of a user experiencing the p99 latency or even the maximum latency is very high.
Turning the database inside out with Apache Samza Link to heading
To my knowledge the first talk describing a event streaming system based on Change Data Capture. It shaped the way I’m thinking about how data flows through an organization.