Document AdGuard placement to reserve memory for Cilium agents
The latest main push updates the cluster topology documentation to explain why AdGuard is placed away from the zimaboard-2 worker node, reserving memory for the planned Cilium agent and Envoy sidecar while noting the NFS-backed work volume allows rescheduling.
Architecture note on AdGuard placement
I added a short paragraph to openwiki/architecture/cluster-topology.md stating that AdGuard runs on a different node than zimaboard-2 to keep memory headroom for the Cilium agent and Envoy. The note also records that the AdGuard work volume uses NFS, so the pod can be rescheduled to another eligible node if needed, and reminds operators to verify the tailnet DNS path after any rescheduling.
Context for the constraint
The zimaboard-2 worker has limited memory, and the upcoming Cilium agent plus Envoy will consume a meaningful share. Placing AdGuard elsewhere avoids contention. The NFS-backed volume means the DNS proxy’s state persists across moves, so rescheduling is a viable operational lever rather than a disruption.
Verification and operational guidance
The documentation now explicitly calls out the tailnet DNS path check after rescheduling AdGuard. This gives the team a concrete verification step when the pod lands on a new node. Future capacity planning can reference this note when evaluating memory budgets on the ZimaBoard workers.