Hi, welcome to another weekly update! It’s been a quiet week, I have been busy finishing off the VPC chapter of my video course Production-Ready Serverless, which should be available to you next week. In the meantime I do have some new posts that I can share with you :-)
We can do better than percentile latencies. I outlined a number of problems with percentile latencies as a latency metric — they are often averaged by providers and that they don’t tell you how bad things are when they go wrong. I proposed some ideas of what we can do instead, including monitoring percentage of requests over SLO instead, and alerting on the absence of successful signal instead of trying to alert on every possible error.
Pricing pitfalls in AWS Lambda. I drilled into the common mistakes people make when it comes to estimating the total cost of their serverless applications. Mistakes such as not considering the 100ms charge blocks, that you’re still paying for idle, and the additional cost of event sources and peripheral services.
With all the media coverage, I don’t think you could have missed the news about the China hardware hack as well as the software side of the hack. Both Amazon and Apple has denied the story, but as a casual observer it just makes it the more believable! As the saying goes “Never believe anything in politics until it has been officially denied.” :-P
Netflix published an awesome post on the design for their edge load balancer, Zuul 2.
Martin Kleppmann’s talk on CRDTs and distributed consistency is finally available on youtube.
Sidney Dekker published a thought provoking article on why we need to better understand “why do things go right?” so we can do more of the right things as much as avoid doing the wrong things.
Jaana Dogan published an equally thought provoking article on Critical Path Driven Development (CPDD), which is a brand new concept to me.