Merge pull request 'The network carries the registry trust (ADR 0082, issues 042/048/062)' (#29) from feat/the-network-carries-registry-trust into main

cmd/mesh-controller/network.go

@@ -152,6 +152,24 @@ func network(ctx context.Context, inv *inventory.Inventory, on map[string]bool,
 		return overlay.Empty(), nil
 	}
 	g, err := overlay.From(nodes, overlayCIDR(), "")
+	if g != nil {
+		// The artifact store, as this network reaches it. Found rather than configured: the
+		// provider is whichever module offers it, on whichever machine holds that module — and if
+		// nothing does yet (genesis raises the registry before the catalogue knows it), there is
+		// no trust to write and nothing is written (novox/hq ADR 0082).
+		//
+		// Refused rather than composed without it when the question could not be answered: a
+		// declaration missing the trust because a lookup failed is a machine that cannot pull,
+		// delivered by a push that reported success — and nothing recomposes it until the next
+		// push (the shape of novox/hq issues 042/048, reappearing as a race).
+		at, port, found, storeErr := artifactStoreOnNetwork(ctx, inv, on)
+		if storeErr != nil {
+			return nil, fmt.Errorf("finding the artifact store this network reaches: %w", storeErr)
+		}
+		if found {
+			g.TrustRegistry(overlay.InternalName(at) + ":" + port)
+		}
+	}
 	if err != nil && len(refused) > 0 {
 		// The network is missing something, and some machines could not be resolved at all. Those
 		// are almost always the same fact: a node that does not resolve contributes nothing, so
@@ -383,3 +401,44 @@ func namesInTheMesh(ctx context.Context, inv *inventory.Inventory) (map[string]s
 	}
 	return out, nil
 }
+
+// artifactStoreOnNetwork is the machine and port the mesh's artifact store answers on, when a
+// module providing it is assigned to a machine that is on the private network.
+//
+// A lookup failure is an error, never "not found": collapsing the two composed a declaration
+// without the trust whenever the inventory hiccuped, delivered by a push that reported success —
+// and nothing recomposed the machine until the next push. "No store" must mean the mesh has none,
+// not that the question went unanswered.
+func artifactStoreOnNetwork(ctx context.Context, inv *inventory.Inventory,
+	on map[string]bool) (node, port string, found bool, err error) {
+
+	shelf, err := inv.Catalogue(ctx)
+	if err != nil {
+		return "", "", false, fmt.Errorf("reading the catalogue: %w", err)
+	}
+	providers := map[string]string{} // module -> served port
+	for name, m := range shelf {
+		served, offers := m.Serves[catalogue.ArtifactStoreProvision]
+		if !offers {
+			continue
+		}
+		if p, ok := served["port"]; ok {
+			providers[name] = fmt.Sprintf("%v", p)
+		}
+	}
+	if len(providers) == 0 {
+		return "", "", false, nil
+	}
+	for machine := range on {
+		assigned, err := inv.Assigned(ctx, machine)
+		if err != nil {
+			return "", "", false, fmt.Errorf("reading what %s is assigned: %w", machine, err)
+		}
+		for _, a := range assigned {
+			if p, ok := providers[a]; ok {
+				return machine, p, true, nil
+			}
+		}
+	}
+	return "", "", false, nil
+}

internal/catalogue/declaration.go

@@ -484,16 +484,7 @@ func (r Resolution) Declaration(with Rendering) ([]map[string]any, error) {
 			// the daemon's. Written `<module>.<id>`, and a dot is what marks it as already
 			// answered: prefixing it again would point at nothing, silently, and the daemon would
 			// serve the old names for ever while everything reported success.
-			if reflects, ok := resource["restart-on"].([]any); ok {
+			if renamed := reflectsRenamed(m.Module, resource["restart-on"]); renamed != nil {
-				var renamed []any
-				for _, id := range reflects {
-					named := fmt.Sprint(id)
-					if strings.Contains(named, ".") {
-						renamed = append(renamed, named)
-						continue
-					}
-					renamed = append(renamed, m.Module+"."+named)
-				}
 				copied["restart-on"] = renamed
 			}
 			out = append(out, copied)
@@ -566,6 +557,36 @@ type Contribution struct {
 // Named after both. Named after the machine alone, two modules on one node wrote to one path: the
 // second overwrote the first, and the provisioner — reading a directory — saw one consumer where
 // there were two.
+// reflectsRenamed is a resource's restart-on list under the module's prefix, or nil when it has
+// none. It reads both the shape JSON parsing produces ([]any) and the shape code composing
+// resources natively produces ([]string): a reference that was skipped because its list arrived
+// in the other shape would point at nothing — silently, with the service never restarting and
+// every check passing, which is how a runtime kept serving without the registry trust its
+// daemon file already carried.
+func reflectsRenamed(module string, reflects any) []any {
+	var names []string
+	switch v := reflects.(type) {
+	case []any:
+		for _, id := range v {
+			names = append(names, fmt.Sprint(id))
+		}
+	case []string:
+		names = v
+	}
+	if names == nil {
+		return nil
+	}
+	var renamed []any
+	for _, named := range names {
+		if strings.Contains(named, ".") {
+			renamed = append(renamed, named)
+			continue
+		}
+		renamed = append(renamed, module+"."+named)
+	}
+	return renamed
+}
+
 func grantPath(directory, consumer, module string) string {
 	return strings.TrimRight(directory, "/") + "/" + consumer + "." + module + ".secret"
 }

internal/catalogue/reflects_renamed_test.go

@@ -0,0 +1,32 @@
+package catalogue
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestReflectsRenamedReadsBothShapes(t *testing.T) {
+	// The list arrives as []any when the resource was parsed from JSON, and as []string when it
+	// was composed in code — the overlay's registry trust is the second kind. A shape that was
+	// skipped would leave the reference unprefixed, pointing at nothing, and the service would
+	// never restart while every check passed (novox/hq issues 042/048, the run-8 diagnosis).
+	parsed := reflectsRenamed("mesh-wireguard", []any{"registry-trust"})
+	if !reflect.DeepEqual(parsed, []any{"mesh-wireguard.registry-trust"}) {
+		t.Fatalf("parsed shape: %v", parsed)
+	}
+	composed := reflectsRenamed("mesh-wireguard", []string{"registry-trust"})
+	if !reflect.DeepEqual(composed, []any{"mesh-wireguard.registry-trust"}) {
+		t.Fatalf("composed shape: %v", composed)
+	}
+
+	// A name already under a module keeps it: that is how a service reflects a file another
+	// module put on the machine.
+	kept := reflectsRenamed("resolver", []string{"mesh-wireguard.fact-node-names", "own-config"})
+	if !reflect.DeepEqual(kept, []any{"mesh-wireguard.fact-node-names", "resolver.own-config"}) {
+		t.Fatalf("dotted name was not kept: %v", kept)
+	}
+
+	if got := reflectsRenamed("any", nil); got != nil {
+		t.Fatalf("no list should rename to nothing, got %v", got)
+	}
+}

internal/overlay/generator.go

@@ -83,8 +83,17 @@ type Generator struct {
 	// keyPath is where each node keeps the private half it generated. Named rather than carried:
 	// the mesh has never seen it and never will.
 	keyPath string
+	// registry is the mesh's artifact store as the network reaches it (host:port), or empty when
+	// the mesh has none. Being on the network is what grants a machine the right to pull from it
+	// (novox/hq ADR 0082), so the module that puts a machine on the network is what writes the
+	// runtime's trust — the same reasoning that has it write /etc/hosts.
+	registry string
 }
 
+// TrustRegistry names the artifact store this network's machines pull from in the clear —
+// the overlay is the transport security (ADR 0082).
+func (g *Generator) TrustRegistry(hostPort string) { g.registry = hostPort }
+
 // From builds a generator over the machines that are part of the network.
 //
 // The nodes given are the ones assigned the module — not every node the mesh knows. A machine
@@ -123,7 +132,29 @@ func (g *Generator) Resources(node string) ([]map[string]any, bool, error) {
 	if err := json.Unmarshal(raw, &parsed); err != nil {
 		return nil, false, err
 	}
-	return parsed.Resources, true, nil
+	resources := parsed.Resources
+	if g.registry != "" {
+		trust, err := json.Marshal(map[string]any{"insecure-registries": []string{g.registry}})
+		if err != nil {
+			return nil, false, err
+		}
+		resources = append(resources,
+			map[string]any{
+				// Merged, not owned: the runtime's daemon file is the machine's, and this states
+				// one fact into it. The registry speaks plain HTTP because every path to it is
+				// already inside the overlay's encryption (ADR 0082) — this line is the runtime
+				// being told what the mesh already means.
+				"id": "registry-trust", "type": "file", "path": "/etc/docker/daemon.json",
+				"content": string(trust) + "\n", "mode": "0644", "merge": "json",
+			},
+			map[string]any{
+				// The runtime reloads nothing for this setting, so it is restarted when the fact
+				// changes — once, at joining, before the machine runs anything that would mind.
+				"id": "registry-trust-reload", "type": "service", "unit": "docker.service",
+				"state": "running", "restart-on": []string{"registry-trust"},
+			})
+	}
+	return resources, true, nil
 }
 
 // Nodes are the machines this generator was built over, so a caller can say who is on the network.

internal/overlay/generator_test.go

@@ -0,0 +1,56 @@
+package overlay
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestTheNetworkCarriesRegistryTrust(t *testing.T) {
+	// novox/hq ADR 0082: being on the network is what grants a machine the right to pull from the
+	// mesh's artifact store in the clear, so the network module writes the runtime's trust — and
+	// writes nothing when the mesh has no store to trust.
+	nodes := []Node{
+		{Name: "anchor", Site: "lab", Hub: true, Endpoint: "192.0.2.10:51820", Key: "k1", Address: "10.42.0.1"},
+		{Name: "node2", Site: "lab", Key: "k2", Address: "10.42.0.2"},
+	}
+	g, err := From(nodes, "10.42.0.0/16", "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	plain, _, err := g.Resources("node2")
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, r := range plain {
+		if r["id"] == "registry-trust" {
+			t.Fatal("trust was written with no artifact store to trust")
+		}
+	}
+
+	g.TrustRegistry("anchor.internal:5000")
+	trusted, part, err := g.Resources("node2")
+	if err != nil || !part {
+		t.Fatalf("resources: %v part=%v", err, part)
+	}
+	var file, service map[string]any
+	for _, r := range trusted {
+		switch r["id"] {
+		case "registry-trust":
+			file = r
+		case "registry-trust-reload":
+			service = r
+		}
+	}
+	if file == nil || service == nil {
+		t.Fatalf("the trust file or its reload is missing: %v", trusted)
+	}
+	if file["path"] != "/etc/docker/daemon.json" || file["merge"] != "json" {
+		t.Fatalf("the trust is not a merged daemon.json: %v", file)
+	}
+	if content, _ := file["content"].(string); !strings.Contains(content, `"anchor.internal:5000"`) {
+		t.Fatalf("the trust does not name the store: %v", file["content"])
+	}
+	if service["unit"] != "docker.service" {
+		t.Fatalf("the reload does not restart the runtime: %v", service)
+	}
+}