Phase 2: Move non-critical services to k3s #2

New issue

Open

opened 2026-05-04 19:36:42 +00:00 by agent-cos · 1 comment

agent-cos commented

2026-05-04 19:36:42 +00:00

Goal

Migrate lightweight services (MeOS Dashboard, MCP servers) to k3s. Free RAM on n97-1.

Tasks

Deploy MeOS Dashboard manifest: kubectl apply -f apps/meos-dashboard/deployment.yaml
Deploy OpenBrain MCP on rpi4-1 (ARM): kubectl apply -f apps/openbrain/deployment.yaml
Deploy MCP servers on rpi4-1: kubectl apply -f apps/mcp-servers/deployment.yaml
Taint ARM nodes: kubectl taint nodes rpi4-1 arm64=true:NoSchedule (same for rpi4-2)
Verify ARM deployments have correct tolerations
Update n97-1 nginx upstreams to point at k8s ClusterIP services
Disable bare-metal MCP services on n97-1
Verify: OpenBrain, forgejo-mcp, lean-experiment still functional

Acceptance

MeOS Dashboard accessible via k8s service
MCP servers running on ARM worker
n97-1 RAM freed by ~200-400MB from moved services
All services still reachable from os.scoble.me

Depends On

#1 (k3s control plane must be up)
rpi4-1 and rpi4-2 must be bootstrapped as workers first (see #3)

## Goal Migrate lightweight services (MeOS Dashboard, MCP servers) to k3s. Free RAM on n97-1. ## Tasks - [ ] Deploy MeOS Dashboard manifest: `kubectl apply -f apps/meos-dashboard/deployment.yaml` - [ ] Deploy OpenBrain MCP on rpi4-1 (ARM): `kubectl apply -f apps/openbrain/deployment.yaml` - [ ] Deploy MCP servers on rpi4-1: `kubectl apply -f apps/mcp-servers/deployment.yaml` - [ ] Taint ARM nodes: `kubectl taint nodes rpi4-1 arm64=true:NoSchedule` (same for rpi4-2) - [ ] Verify ARM deployments have correct tolerations - [ ] Update n97-1 nginx upstreams to point at k8s ClusterIP services - [ ] Disable bare-metal MCP services on n97-1 - [ ] Verify: OpenBrain, forgejo-mcp, lean-experiment still functional ## Acceptance - MeOS Dashboard accessible via k8s service - MCP servers running on ARM worker - n97-1 RAM freed by ~200-400MB from moved services - All services still reachable from os.scoble.me ## Depends On - #1 (k3s control plane must be up) - rpi4-1 and rpi4-2 must be bootstrapped as workers first (see #3)

sscoble commented

2026-05-25 02:18:13 +00:00

Owner

Triage: Still pending — blocked on workers (#3)

Phase 2 depends on:

#1 ✅ (control plane is running — closed)
#3 ❌ (workers not bootstrapped — none of n97-2, n97-3, rpi4-1, rpi4-2 are joined)

Cannot proceed until at least one worker node (preferably rpi4-1 or rpi4-2) is joined to the cluster and tainted for ARM workloads.

## Triage: Still pending — blocked on workers (#3) Phase 2 depends on: - #1 ✅ (control plane is running — closed) - #3 ❌ (workers not bootstrapped — none of n97-2, n97-3, rpi4-1, rpi4-2 are joined) Cannot proceed until at least one worker node (preferably rpi4-1 or rpi4-2) is joined to the cluster and tainted for ARM workloads.