A working private network, and four reasons it did not work

Three machines across two sites, two of them behind no reachable address, all
nine paths open. The mesh computes the graph, delivers it as a declaration, and
the nodes bring it up.

Every fault below looked like success from inside the mesh: the graph was
right, the files were right, the services were up, every node reported it had
applied. None was reachable by reasoning.

A running interface does not re-read its configuration. A node joins, every
existing node's peer list changes, the file is replaced -- and the service is
already running, so nothing reloads it. Fixed as declared state rather than a
command: the service must reflect the file. A command to restart would be an
action, and the link may not carry one. The host refused exactly that, which is
how this shape was arrived at.

A hub sharing a site with a spoke appeared twice in that spoke's peer list --
once as a direct peer, once as the route of last resort. WireGuard takes one
entry per key and refuses the file. The ordinary shape of a small mesh, and in
none of the tests written before it ran.

Two nodes at one site that neither can be dialled were peered directly. Nobody
opens the path, and the direct route is more specific than the hub's, so it
wins and blackholes -- this design's own warning arriving in its
implementation. They now route through the hub unless one end can be dialled.

And Docker sets the FORWARD policy to DROP, so a hub with ip_forward enabled
carried nothing between its spokes. The substrate at tier 1 silently breaks the
network at tier 2, and nothing in either tier's state says so. The hub inserts
its own rule above those chains and removes it on the way down.

Two weak tests found by injection along the way: one asserted the keepalive
rule only against the hub, whose peer entries happen not to set that field at
all, so it tested an absence; the other checked the firewall rules by looking
for FORWARD anywhere, which the PostDown line satisfies on its own.

internal/inventory/addresses.go

@@ -0,0 +1,76 @@
+package inventory
+
+import (
+	"context"
+	"fmt"
+	"net/netip"
+)
+
+// The mesh assigns overlay addresses. A node computes nothing about the network it is joining —
+// it generates a keypair, publishes the public half, and receives the rest
+// (novox/hq 08-connectivity).
+
+// AssignAddress gives a node its place on the private network, if it has none.
+//
+// Idempotent: a node that already has an address keeps it. Reassigning would change every other
+// node's peer list to chase it, and an address that moves is the thing declaring the hub was
+// meant to stop.
+//
+// Allocated in order from the range, taking the lowest free one. Not random: a person reading a
+// peer list should be able to guess which node an address belongs to, and reuse of a released
+// address is a smaller problem than a list nobody can hold in their head.
+func (i *Inventory) AssignAddress(ctx context.Context, node, cidr string) (string, error) {
+	prefix, err := netip.ParsePrefix(cidr)
+	if err != nil {
+		return "", fmt.Errorf("%q is not a network the mesh can allocate from: %w", cidr, err)
+	}
+
+	var existing *string
+	if err := i.store.Pool().QueryRow(ctx,
+		`select host(overlay_address) from node where id = $1`, node).Scan(&existing); err != nil {
+		return "", err
+	}
+	if existing != nil && *existing != "" {
+		return *existing, nil
+	}
+
+	taken := map[string]bool{}
+	rows, err := i.store.Pool().Query(ctx,
+		`select host(overlay_address) from node where overlay_address is not null`)
+	if err != nil {
+		return "", err
+	}
+	for rows.Next() {
+		var a string
+		if err := rows.Scan(&a); err != nil {
+			rows.Close()
+			return "", err
+		}
+		taken[a] = true
+	}
+	rows.Close()
+	if err := rows.Err(); err != nil {
+		return "", err
+	}
+
+	// The first address in a range is conventionally the network itself and is skipped, so
+	// allocation starts one past it.
+	candidate := prefix.Masked().Addr().Next()
+	for prefix.Contains(candidate) {
+		if !taken[candidate.String()] {
+			if _, err := i.store.Pool().Exec(ctx,
+				`update node set overlay_address = $2::inet where id = $1`,
+				node, candidate.String()); err != nil {
+				return "", err
+			}
+			return candidate.String(), nil
+		}
+		candidate = candidate.Next()
+	}
+	// Said plainly rather than returning an empty address that fails later on a machine. A mesh
+	// that has outgrown its range needs a person, and renumbering is not something to attempt
+	// halfway through assigning one node.
+	return "", fmt.Errorf(
+		"every address in %s is taken, so %s cannot be given one. The mesh has outgrown its "+
+			"range and renumbering it is a deliberate act", cidr, node)
+}