100% uptime is impossible. Modern architectures are designed around failure but what does that mean for the human aspect of incident management? David's talk considers how to prepare for outages, how to structure the response, and how those experiences and techniques differ for small and large companies.
As developers we tend to neglect the importance of the academic research on distributed systems. Sam reminds us to spend some time into diving into the details and the formal theory behind many of our distributed data stores, grasp the trade-offs they make, and avoid getting fooled by their marketing.
Paul delves into some of the challenges involved in building a distributed database specifically for time series data. High write throughput, even higher read throughput, large range scans of data to answer queries, and large range deletes all conspire to make time series data both the worst and best use case to build for in distributed databases.
Convergent Replicated Data Types offer a principled approach to eventually consistent data modelling, with defined data types always converging to single correct values. CRDTs are one of the key building blocks of a distributed system, enabling strong eventual consistency and highly available, low latency applications.