The store keeps what the records name (hq ADR 0189)

The mesh names what may go from its own build records — a digest it did not record making is never named, which is what keeps the sweep away from the images genesis pushed. An artifact stays because a definition the mesh holds names it, or because it belongs to one of the five most recent successful builds of its module. internal/artifacts asks the store to let go of one; internal/inventory decides and remembers (migration 0055); the sweep runs after a build the mesh recorded, which is when both the bytes and the keep set moved. Never fatal to a build. And the manifest side of while-stopped, refused from the definition alone: no schedule, run-once, a container the module does not declare, itself.
2026-10-02 21:49:03 +02:00
parent e09a14ba4c
commit 0412653920
9 changed files with 857 additions and 0 deletions
@@ -1511,6 +1511,13 @@ func ParseManifest(raw []byte) (Manifest, error) {
 				}
 			}
 		}
+		// **A scheduled step may hold this module's own containers still while it runs**
+		// (novox/hq ADR 0189). What the host judges is the declaration it receives — whether each
+		// id is a container placed on that machine; what belongs here is what only the definition
+		// shows: that the ids are this module's, that they are containers, and that the step is
+		// scheduled. A module naming a neighbour's container would be a module that can stop the
+		// mesh, and the manifest is where that is visible.
+		problems = append(problems, whileStoppedProblems(m, r, hasSchedule(r))...)
 	}
 	for name, own := range m.OwnSecrets {
 		if !placedOrAbsolute(own.Path) {
@@ -2042,3 +2049,71 @@ func (o OwnSecrets) Paths() map[string]string {
 // InstancesInterchangeable is the one value of a definition's `instances`: the module is the same
 // on every machine, so any instance may answer for the module.
 const InstancesInterchangeable = "interchangeable"
+
+// WhileStopped is the resource key naming the containers a scheduled step holds still while it
+// runs (novox/hq ADR 0189). Carried to the host unchanged, like `schedule`.
+const WhileStopped = "while-stopped"
+
+// hasSchedule is whether a resource declares a cadence, as a string.
+func hasSchedule(r map[string]any) bool {
+	s, _ := r["schedule"].(string)
+	return s != ""
+}
+
+// whileStoppedProblems judges one container's maintenance window against its own definition
+// (novox/hq ADR 0189).
+//
+// Three things the manifest is the only place to see: that the step is scheduled (a one-time
+// offline job says *before* rather than *instead of* — at apply the host already has a window,
+// because the declaration is applied in order and a run-once step gates what follows); that every
+// id it names is **this module's own** container; and that it does not name itself.
+//
+// The host checks the fourth — that the container is actually placed on that machine — because
+// that is a fact about the declaration and not about the definition.
+func whileStoppedProblems(m Manifest, r map[string]any, scheduled bool) []string {
+	raw, present := r[WhileStopped]
+	if !present {
+		return nil
+	}
+	ids, ok := raw.([]any)
+	if !ok {
+		return []string{fmt.Sprintf(
+			"%s declares %s on %v as a %T; it is a list of this module's container ids",
+			m.Module, WhileStopped, r["id"], raw)}
+	}
+	var problems []string
+	if len(ids) > 0 && !scheduled {
+		problems = append(problems, fmt.Sprintf(
+			"%s declares %s on %v, which has no schedule. A maintenance window is for a recurring "+
+				"step: at apply the mesh already has one, because a run-once step gates what is "+
+				"declared after it (novox/hq ADR 0189)", m.Module, WhileStopped, r["id"]))
+	}
+	containers := map[string]bool{}
+	for _, own := range m.Resources {
+		if fmt.Sprint(own["type"]) == "container" {
+			containers[fmt.Sprint(own["id"])] = true
+		}
+	}
+	for _, each := range ids {
+		id, ok := each.(string)
+		if !ok {
+			problems = append(problems, fmt.Sprintf(
+				"%s declares %s on %v naming a %T; each entry is a container's id",
+				m.Module, WhileStopped, r["id"], each))
+			continue
+		}
+		if id == fmt.Sprint(r["id"]) {
+			problems = append(problems, fmt.Sprintf(
+				"%s declares %s on %v naming itself", m.Module, WhileStopped, r["id"]))
+			continue
+		}
+		if !containers[id] {
+			problems = append(problems, fmt.Sprintf(
+				"%s declares %s on %v naming %q, which is not a container this module declares. "+
+					"A step may hold still its own module's containers and nobody else's — one "+
+					"that could quiesce a neighbour could stop the mesh",
+				m.Module, WhileStopped, r["id"], id))
+		}
+	}
+	return problems
+}
@@ -0,0 +1,89 @@
+package catalogue
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+// A scheduled step may hold its module's own containers still while it runs (novox/hq ADR 0189).
+//
+// The host judges what it receives — whether each id is a container on that machine. What the
+// definition is the only place to see is judged here, near whoever wrote it.
+
+func aStoreManifest(step map[string]any) []byte {
+	m := map[string]any{
+		"module": "distribution", "version": "1",
+		"resources": []any{
+			map[string]any{"id": "store", "type": "container", "name": "mesh-registry",
+				"image": "registry@sha256:" + strings.Repeat("a", 64)},
+			step,
+		},
+	}
+	raw, _ := json.Marshal(m)
+	return raw
+}
+
+func TestAMaintenanceWindowOnItsOwnModulesContainerIsAccepted(t *testing.T) {
+	raw := aStoreManifest(map[string]any{
+		"id": "collect", "type": "container", "name": "mesh-registry-collect",
+		"image":    "registry@sha256:" + strings.Repeat("a", 64),
+		"schedule": "30 3 * * *", "while-stopped": []any{"store"},
+	})
+	if _, err := ParseManifest(raw); err != nil {
+		t.Fatalf("a step holding its own module's container still was refused: %v", err)
+	}
+}
+
+func TestAMaintenanceWindowIsRefusedWhereTheDefinitionShowsItCannotMean(t *testing.T) {
+	for _, c := range []struct {
+		name string
+		step map[string]any
+		says string
+	}{
+		{
+			"on a step with no schedule",
+			map[string]any{"id": "collect", "type": "container", "name": "c",
+				"image":         "registry@sha256:" + strings.Repeat("a", 64),
+				"while-stopped": []any{"store"}},
+			"gates what is declared after it",
+		},
+		{
+			"on a run-once step, which already has order",
+			map[string]any{"id": "collect", "type": "container", "name": "c",
+				"image":    "registry@sha256:" + strings.Repeat("a", 64),
+				"run-once": true, "while-stopped": []any{"store"}},
+			"A maintenance window is for a recurring step",
+		},
+		{
+			"naming a container this module does not declare",
+			map[string]any{"id": "collect", "type": "container", "name": "c",
+				"image":    "registry@sha256:" + strings.Repeat("a", 64),
+				"schedule": "30 3 * * *", "while-stopped": []any{"the-broker"}},
+			"could quiesce a neighbour could stop the mesh",
+		},
+		{
+			"naming itself",
+			map[string]any{"id": "collect", "type": "container", "name": "c",
+				"image":    "registry@sha256:" + strings.Repeat("a", 64),
+				"schedule": "30 3 * * *", "while-stopped": []any{"collect"}},
+			"naming itself",
+		},
+		{
+			"written as something that is not a list",
+			map[string]any{"id": "collect", "type": "container", "name": "c",
+				"image":    "registry@sha256:" + strings.Repeat("a", 64),
+				"schedule": "30 3 * * *", "while-stopped": "store"},
+			"a list of this module's container ids",
+		},
+	} {
+		_, err := ParseManifest(aStoreManifest(c.step))
+		if err == nil {
+			t.Errorf("%s was accepted", c.name)
+			continue
+		}
+		if !strings.Contains(err.Error(), c.says) {
+			t.Errorf("%s: the refusal does not say %q:\n%v", c.name, c.says, err)
+		}
+	}
+}