| • Discussion on Lars Wikman's unchanged operational setup since June 2021 | |
| • Recap of past episode "Why Kubernetes" and the follow-up with Gerhard Lazu | |
| • Lars' nuanced feelings about Kubernetes, considering its complexity and mystery | |
| • Current take on Kubernetes landscape and cloud-native technologies | |
| • Comparison of various tools and platforms, including k3s, ArgoCD, and Fly | |
| • Exploration of production setup, including operating system, packages, CI/CD, and server choices | |
| • Lars' experience with client projects using Fly and GitLab for platform engineering | |
| • Discussion on the trade-offs between layering and packaging deployment aspects into the app itself | |
| • The simplicity of certain programming languages like Go and Elixir in handling load and scaling | |
| • Challenges with Node.js in terms of scaling and CPU-bound loads | |
| • Comparison of programming languages for machine learning and AI tasks | |
| • Deployment artifacts and strategies, including use of containers and SCP (Secure Copy Protocol) | |
| • Importance of knowing which hash is being pushed to production environments | |
| • Backup strategy and disaster recovery | |
| • Monitoring and alerting for production environments | |
| • Blue/green deployment on single machines with minimal setup | |
| • Orchestrating releases with CI/CD pipelines and artifact management | |
| • Hot code reloading and upgrading running versions of the application | |
| • Balancing monolithic architecture with operational concerns | |
| • Discussion on running a monolithic architecture with external systems interacting with it | |
| • Importance of simplicity in deployment and operations for smaller teams and organizations | |
| • Concerns about choosing Kubernetes or other widely-used tools as they may not provide a competitive advantage | |
| • The value of taking chances and making decisions that go against common practices, such as Apple's approach to shipping half-finished features | |
| • Personal experiences with tooling and deployment methods, including using Fly.io for cloud deployments | |
| • Considering bare metal or dedicated servers over cloud-native options for certain projects | |
| • Exploring the idea of building a system without persistent data storage for an art project, using Erlang hot code updates | |
| • Hot code updates and their challenges | |
| • Trade-offs between automation and manual configuration | |
| • Importance of clear documentation for system setup and operation | |
| • Balancing complexity with maintainability and understandability | |
| • Staying within one's comfort zone and familiar ecosystem | |
| • Challenges of deploying systems in different programming languages | |
| • Discussion of Dagger and its potential benefits for building SDKs in various languages | |
| • Elixir as a preferred language for CI/CD tooling and its limitations when used with YAML | |
| • Kubernetes and its complexities, including the need to reconcile declarative systems with operational requirements | |
| • Preference for using Linux due to comfort and familiarity, but acknowledging that other systems (e.g. BSDs) may have advantages in certain areas | |
| • Discussion of systemd as a complex system that can be difficult to use effectively | |
| • Mention of exploring new tools and technologies, including the use of non-Linux operating systems (e.g. FreeBSD) | |
| • Introduction of DCH Dave Cottlehuber, an expert in operational systems who works with FreeBSD and has experience with CouchDB. | |
| • The danger of going off the beaten path with technology and tools | |
| • The example of NixOS, PureScript, Haskell, OCaml, and other niche programming languages/systems | |
| • The importance of not over-introducing new technology at once | |
| • The need to balance challenging oneself with the potential risks of incompatibility |