Skip to main content

Architecture decisions

Each numbered record explains one choice: the situation, the decision, the alternatives, and the costs accepted. Decisions are history, not deployment instructions. An Accepted decision can still describe an experiment that a later decision ended, so follow the chain to the newest relevant record.

The original 40 decisions keep their dates and statuses; their public editions were condensed in October 2026. Later decisions are public from their first edition. The decision policy explains statuses and supersession, and the capabilities map shows what runs today.

Supersession chains​

A solid arrow points from a decision to the one that superseded it. The dashed arrow marks a decision that stayed Accepted but whose platform a later decision retired.

ChainTopicEnds at
0006 → 0016 → 0019 → 0043 → 0044SeaweedFS interface, placement, and stock runtime0044
0020 → 0021 → 0022Cilium memory and clean-state0022
0025 → 0029Network policy and Pod Security0029
0008 → 0030 → 0032Logging backend and retention0032
0041 → 0042Log-delivery canaries0042
0031 → … → 0040Lakehouse, from demo to retirement0040

All decisions​

IDDecisionStatusDate
0001Cluster reset and structured rebuildAccepted2026-04-10
0002Defer Rook-CephDeferred2026-04-10
0003Defer monitoring stack pending object storageDeferred2026-04-11
0004Storage stack frameworkAccepted2026-04-11
0005Adopt Longhorn for replicated block storageAccepted2026-04-11
0006Adopt SeaweedFS for object storageSuperseded by 00162026-04-11
0007Adopt kube-prometheus-stack for metricsAccepted2026-04-12
0008OpenTelemetry logs and traces roadmapSuperseded by 00302026-04-12
000910 Gbit SFP+ full-mesh storage networkingAccepted2026-04-15
0010Persist Tailscale node identityAccepted2026-04-15
0011Adopt an in-repository planning systemAccepted2026-04-16
0012Tailscale for internal remote workload accessAccepted2026-04-16
0013Restructure the cluster-health dashboardAccepted2026-04-17
0014Clean failed pods with a Flux-managed CronJobAccepted2026-04-17
0015Adopt Harbor as the in-cluster image registryAccepted2026-04-18
0016Adopt SeaweedFS using embedded filer S3Superseded by 00192026-04-18
0017Adopt Multus for the Longhorn storage networkAccepted2026-04-18
0018Install CNI reference plugins for MultusAccepted2026-04-19
0019SeaweedFS with the canonical spec.s3 interfaceSuperseded by 00432026-04-19
0020Size Cilium memory for cgroup working setSuperseded by 00212026-04-21
0021Raise Cilium memory and clean state on restartSuperseded by 00222026-04-23
0022Keep the Cilium memory limit; reject permanent clean-stateAccepted2026-04-23
0023Use one shared Cloudflare Tunnel for public ingressAccepted2026-04-29
0024Use Flux image automation for owned applicationsAccepted2026-05-01
0025Keep chart NetworkPolicies enabled and repair probe gapsSuperseded by 00292026-05-04
0026Use self-hosted ntfy for alert notificationsAccepted2026-05-04
0027Require explicit Flux dependencies for custom resourcesAccepted2026-05-07
0028Run ClickStack as a time-boxed observability experimentAccepted2026-06-03
0029Restore Pod Security and isolate public workloadsAccepted2026-07-09
0030Adopt Loki for Kubernetes logsSuperseded by 00322026-08-02
0031Adopt a SeaweedFS Iceberg log demoSuperseded by 00332026-08-06
0032Honor Loki's stream retention minimumAccepted2026-08-11
0033Adopt Airflow and Spark Operator for the lakehouseSuperseded by 00362026-08-11
0034Adopt the Barman Cloud plugin for CNPG backupsSuperseded by 00352026-08-12
0035Defer Airflow metadata backupsSuperseded by 00362026-08-12
0036Continue the lakehouse without metadata recoverySuperseded by 00372026-08-13
0037Remove the extended Airflow observation gateAccepted2026-08-14
0038Extend the Airflow and Spark learning review onceSuperseded by 00392026-08-14
0039Retain Airflow and Spark as a learning platformSuperseded by 00402026-08-14
0040Retire the lakehouse and TemporalAccepted2026-10-12
0041Adopt Loki stdout canariesSuperseded by 00422026-10-19
0042Bound numbered log confirmation deadlinesAccepted2026-10-19
0043Keep two SeaweedFS copies for node maintenanceSuperseded by 00442026-10-19
0044Use stock shared S3 with planned downtimeAccepted2026-10-20
0045Deploy only commits that pass the deploy gateAccepted2026-10-23