mounts: clear a failed mount of its own whose source was away, and the failure of a unit it takes over, so neither is said as the machine's own failed unit (hq ADR 0263 rules 4 and 6)

modules/mounts/cmd/mounts/failed_test.go

@@ -0,0 +1,101 @@
+package main
+
+import (
+	"context"
+	"strings"
+	"testing"
+	"time"
+)
+
+// A mount the module took over, or one whose source was away when it was touched, is never left as a
+// failed unit: the node-engine would say it as a failed unit of the machine's own, with a notification each
+// time (hq ADR 0263 rules 4 and 6). A mount that failed while its source answers stays failed: a fault.
+
+func failedLook(t *testing.T, f *fakeSystem) map[string]ShareStatus {
+	t.Helper()
+	st := f.applier(time.Date(2026, 10, 9, 14, 0, 0, 0, time.UTC)).Look(context.Background())
+	out := map[string]ShareStatus{}
+	for _, s := range st.Shares {
+		out[s.Name] = s
+	}
+	return out
+}
+
+func TestTheFailureOfTheUnitTheModuleTakesOverIsCleared(t *testing.T) {
+	f := newSystem(t)
+	f.write(ConfigFile, "shares=library=/srv/library\nserver=homeserver\nat=192.0.2.10\nnode=workstation\n")
+	f.write(ExportsFile, exportsAnswer(t, library))
+	f.reachable["192.0.2.10:2049"] = true
+	// fstab's generator made srv-library.mount for the same mount point, and it timed out at boot.
+	f.units["srv-library.mount"] = map[string]string{"ActiveState": "failed", "Result": "timeout"}
+
+	states := failedLook(t, f)
+
+	if !f.called("systemctl reset-failed srv-library.mount") {
+		t.Fatalf("the generator's failure is cleared once the module's unit is written: %q", f.calls)
+	}
+	if s := states["library"]; s.State != StateArmed {
+		t.Errorf("the share is judged on its own mounts, not on the generator's: %+v", s)
+	}
+}
+
+func TestAMountThatFailedBecauseItsSourceWasAwayIsCleared(t *testing.T) {
+	f := newSystem(t)
+	f.write(ConfigFile, "shares=library=/srv/library\nsources=games=smb://192.0.2.76/Share@/mnt/games\n"+
+		"server=homeserver\nat=192.0.2.10\nnode=workstation\naccount=operator\n"+
+		"secret=/var/lib/mesh/mounts/smb-credentials.secret\nstate=/var/lib/mesh/mounts\n")
+	f.write(ExportsFile, exportsAnswer(t, library))
+	f.write("/var/lib/mesh/mounts/smb-credentials.secret", "[games]\nusername=root\npassword="+testPassword+"\n")
+	a := f.applier(time.Date(2026, 10, 9, 14, 0, 0, 0, time.UTC))
+	a.Look(context.Background()) // writes and arms both pairs; neither source answers
+	f.calls = nil
+
+	// Someone opened both mount points: the gaming device is off and the server is away, so both mounts failed.
+	f.units["mnt-games.mount"] = map[string]string{"ActiveState": "failed", "Result": "timeout"}
+	f.units["srv-library.mount"] = map[string]string{"ActiveState": "failed", "Result": "timeout"}
+	cleared := a.clearFailed(context.Background())
+
+	for _, unit := range []string{"mnt-games.mount", "srv-library.mount"} {
+		if !f.called("systemctl reset-failed " + unit) {
+			t.Errorf("%s failed because its source was away and is cleared: %q", unit, f.calls)
+		}
+	}
+	if len(cleared) != 2 || !strings.Contains(strings.Join(cleared, "; "), "games was not there to mount") {
+		t.Errorf("each clearing is said: %q", cleared)
+	}
+	states := failedLook(t, f)
+	if states["games"].State != StateAbsent || states["library"].State != StateUnreachable {
+		t.Errorf("the module still says each as absent or unreachable: %+v", states)
+	}
+}
+
+func TestAMountThatFailedWhileItsSourceAnswersStaysFailed(t *testing.T) {
+	f := newSystem(t)
+	f.write(ConfigFile, "shares=library=/srv/library\nserver=homeserver\nat=192.0.2.10\nnode=workstation\n")
+	f.write(ExportsFile, exportsAnswer(t, library))
+	f.reachable["192.0.2.10:2049"] = true
+	a := f.applier(time.Date(2026, 10, 9, 14, 0, 0, 0, time.UTC))
+	a.Look(context.Background())
+	f.calls = nil
+
+	f.units["srv-library.mount"] = map[string]string{"ActiveState": "failed", "Result": "exit-code"}
+	if cleared := a.clearFailed(context.Background()); len(cleared) != 0 || f.called("systemctl reset-failed") {
+		t.Fatalf("a mount refused by a server that answers is a fault and is left failed: %q %q", cleared, f.calls)
+	}
+	if s := failedLook(t, f)["library"]; s.State != StateFailed {
+		t.Errorf("it is said failed: %+v", s)
+	}
+}
+
+func TestOnlyTheActingProcessClears(t *testing.T) {
+	f := newSystem(t)
+	f.write(ConfigFile, "shares=library=/srv/library\nserver=homeserver\nat=192.0.2.10\nnode=workstation\n")
+	f.write(ExportsFile, exportsAnswer(t, library))
+	f.units["srv-library.mount"] = map[string]string{"ActiveState": "failed", "Result": "timeout"}
+	a := f.applier(time.Now())
+	a.Act = false
+	a.Look(context.Background())
+	if f.called("systemctl reset-failed") {
+		t.Errorf("the judging process changes nothing: %q", f.calls)
+	}
+}