meta/meeting-notes/2024-11-20-architecture-meetup/2024-11-20-architecture-meetup.md

​Architecture meetup 2024-11-20

13:30 - 15:00

Agenda Times are approx. Less is better ;) 13:30 - 13:35 Goals of the meeting (~5min) 13:35 - 14:15 Diagram discussion (~40min) 14:15 - 14:35 Status of the project (~20min) 14:35 - 14:55 Roles & responsibilities (~25min) 14:55 - 15:00 FOSSDEM (~5min)

Attendees: Bjorn, Ronny, Richard, Kevin, Nicolas, Gheorghe, Valentin

Goals: * Clarify uncertainties and "freeze" the architecture

Notes: * Robert and Koen are missing today, both are critical * Management layer: * IdentityManagement should be clarified as Authorization, Authentication, Accounting (AAA) * The Nix-Panel to used by the operator (our "customer") with high-level parameters * operator's email * DNS zone * booked storage and compute * etc * TBD in how much this will be greenfield, see architecture discussion from a few weeks ago * About the central database (multiples) * Q: what is the datamodel? TBD in more detail.
* State of the system per operator * AAA stuff * Operator bespoke info * Provider bespoke info * upstream DNS config * Netbox: provider side (physical network layout & hardware) * Q: What's the role of NixOps4 here? * Valentin: NixOps4 merely provides a mechanism. The policy is implemented by "resource providers" which are domain-specific and plugged into NixOps4 to CRUD the various data sources * TO DO: There are no "use cases" yet to describe how the services works; e.g setting up a service like Pixelfed etc. Basically a case to describe how the components work together. * (some discussion on the various representations of the system: component dependency graph, data flow graph for how deployments come together, user stories for the various actors) * Valentin proposes to focus on the component dependencies for now, as the current diagram already mostly represents those * can sketch user stories on the side * "Nix-configuration" and Proxmos are merely resource providers for NixOps4 * TODO: Glossary for definitions (make sure we all speak the same language). * Gheorghe proposes to annotate each box with the component type (e.g. "virtualisation provider") and [at least one, if there are multiple planned] concrete implementation (e.g. "proxmox") * Ronny: there may be services that happen not to run under NixOS but some other Linux distro * We will need another configuration system for those, e.g. Ansible * This would be another resource provider for NixOps4 * We shouldNix declare it out of scope, since NixOS is the more natural thing to do, or hack it together with shell scripts if absolutely required * We currently don't have services that aren't - and defintely none that can't be - nixified * Also, centrally managed systems, such as provider-side DNS management, can be handled by the provider classically, e.g. on Debian * Our particular target deSEC would be expensive to package for NixOS, but we need it exactly once per provider and it won't be redeployed * NixOS services: * The only difference between "Services" and "FediServices" is that "FediServices" have federation * More nuanced: Fediverse services are intended to be used by the general public, while the others are more interesting for academic institutions * This is a distinction per work package * Technically, all of them are NixOS modules with configurations that are somewhat specific to our architecture * Unspecified requirement: Backups for all of these * From previous discussions: storage is one of * Service data: Block storage (ZFS), and snapshots are pulled on the provider side * Essentially anything that can't be stored as blob storage * User data: Blob storage (e.g. Garage), could optionally be replicated to compatible services * Q: is this already available in Garage or would we need to build it, and is it relevant to our mission? * Storage needs to be rewired for each service so the data actually lands where it should * Need to decide whether it's worth the time investment on a case-by-case basis * "Core services": * secrets management is more of a concern for deployments (as a NixOps4 resource provider) * Secrets are a resource & need a resource provider plugin. * Which resourceproviders does NixOps need to talk & their context. * The purpose of this block is to signify that some services are mandatory for an operator-side deployment, but they exist already and only need to be interfaced with via NixOps4 * For example, there is a DNS management system and an email server, and a deployment merely needs to register with them * TODO: label this in the architecture diagram * TODO: move it into the "Management" block, maybe rename this to "existing" or "pre-defined" services * TODO: clarify mapping between archtectural components and use case actors, refine naming *

Due to other meetings we had to stop here. We still have to discuss these topics from the agenda: 14:15 - 14:35 Status of the project (~20min) 14:35 - 14:55 Roles & responsibilities (~25min) 14:55 - 15:00 FOSSDEM (~5min)