Let two controllers overlap safely while one hands over to the other (hq issue 213)

The controller's machine moves it from the container to a process by
starting the process first and removing the container once the process
is up (mesh-host's `replaces`). For that moment two controllers share the
store and the bus. Checked what each does:

- the seat's verbs: a queue group per seat, each call answered once. Safe.
- the controller's consumers on CONTROL and EVENTS: push consumers with
  no delivery group, so the second bind is refused with "consumer is
  already bound" and serve exited. The process would restart for ever,
  the host would never see it up, and the container would never go. The
  second controller now stands by and binds when the first lets go
  (tested on a real bus; fails without the change).
- plans: read, changed and saved whole by the 30s timer, by build
  outcomes, by a merge and by `plans stop`. Two timers would each ask a
  tier the other had just asked. Working the plans now takes a
  session-level advisory lock on the inventory: the timer skips while
  another holds it, the other paths wait for it. Build asks happen only
  inside plan work and are covered by the same lock.

internal/link/receive_nats.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"log"
 	"strings"
 	"time"
 
@@ -78,10 +79,15 @@ func (n *natsInbound) Receive(ctx context.Context, act func(context.Context, Con
 	// than creating one here: the consumer is an object with a configuration — ack policy, ack
 	// wait, redelivery — and a client that creates its own would be a second opinion about it.
 	control := make(chan *nats.Msg, Prefetch)
-	said, err := js.ChanSubscribe("", control, nats.Bind("CONTROL", broker.ControllerName))
+	said, err := standingBy(ctx, log.Default(), "CONTROL", func() (*nats.Subscription, error) {
+		return js.ChanSubscribe("", control, nats.Bind("CONTROL", broker.ControllerName))
+	})
 	if err != nil {
 		return fmt.Errorf("subscribing to what nodes say: %w", err)
 	}
+	if said == nil {
+		return nil // stopped while standing by
+	}
 	defer func() { _ = said.Unsubscribe() }()
 
 	// Heartbeats, on core NATS and off any stream (design 25 §3). Their own subscription because
@@ -97,10 +103,15 @@ func (n *natsInbound) Receive(ctx context.Context, act func(context.Context, Con
 	var events chan *nats.Msg
 	if len(n.follows) > 0 {
 		events = make(chan *nats.Msg, Prefetch)
-		followed, err := js.ChanSubscribe("", events, nats.Bind("EVENTS", broker.ControllerName))
+		followed, err := standingBy(ctx, log.Default(), "EVENTS", func() (*nats.Subscription, error) {
+			return js.ChanSubscribe("", events, nats.Bind("EVENTS", broker.ControllerName))
+		})
 		if err != nil {
 			return fmt.Errorf("subscribing to what the catalogue says: %w", err)
 		}
+		if followed == nil {
+			return nil
+		}
 		defer func() { _ = followed.Unsubscribe() }()
 	}
 
@@ -324,3 +335,44 @@ func (m *natsControl) forget() {
 type replyAddressed struct {
 	ReplyTo string `json:"reply_to,omitempty"`
 }
+
+// StandbyPoll is how often a controller standing by looks again for its consumers. A variable so a
+// test need not wait.
+var StandbyPoll = 2 * time.Second
+
+// standingBy binds one of the controller's consumers, waiting while another controller holds it.
+//
+// **Two controllers, one consumer** (novox/hq issue 213). The controller's consumers are push
+// consumers with no delivery group, so the server lets one subscription bind each — on purpose:
+// two would each act on every message (issue 146). When a machine hands its controller over from
+// the container to the process, the host starts the process first and removes the container only
+// once the process is up; the process then finds the consumers bound. Exiting on that would never
+// be up, so the container would never go. It stands by instead — the seat's verbs are already
+// served from a queue group, and the plans wait on their lock — and binds as soon as the other lets
+// go. Nil and no error is ctx ending while it waited.
+func standingBy(ctx context.Context, logger interface{ Printf(string, ...any) }, stream string,
+	bind func() (*nats.Subscription, error)) (*nats.Subscription, error) {
+	said := false
+	for {
+		sub, err := bind()
+		if err == nil {
+			if said {
+				logger.Printf("took the controller's consumer on %s: the controller that held it let go", stream)
+			}
+			return sub, nil
+		}
+		if !strings.Contains(err.Error(), "already bound") {
+			return nil, err
+		}
+		if !said {
+			logger.Printf("another controller holds the controller's consumer on %s; standing by "+
+				"until it lets go", stream)
+			said = true
+		}
+		select {
+		case <-ctx.Done():
+			return nil, nil
+		case <-time.After(StandbyPoll):
+		}
+	}
+}

internal/link/standby_nats_test.go

@@ -0,0 +1,69 @@
+package link
+
+import (
+	"context"
+	"encoding/json"
+	"os"
+	"testing"
+	"time"
+
+	"github.com/novox/mesh-controller/internal/broker"
+)
+
+// novox/hq issue 213: while a machine hands its controller over, the new controller (the process)
+// is started while the old one (the container) still holds the controller's consumers. It must not
+// exit — the host would read that as a replacement that did not come up and never remove the
+// container — and must not act on what the old one is handed. It stands by, and takes the consumers
+// when the old one lets go.
+func TestNatsASecondControllerStandsByAndTakesOverWhenTheFirstLetsGo(t *testing.T) {
+	js := aBus(t)
+	was := StandbyPoll
+	StandbyPoll = 50 * time.Millisecond
+	defer func() { StandbyPoll = was }()
+
+	old := &counted{}
+	_, stopOld := servingOn(t, js, old)
+	eventually(t, "the first controller binding its consumer", func() bool {
+		info, err := js.Context().ConsumerInfo("CONTROL", broker.ControllerName)
+		return err == nil && info.PushBound
+	})
+
+	// The new one, on a connection of its own as the process would have.
+	second, err := broker.Dial(os.Getenv("MESH_TEST_NATS"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(second.Close)
+	fresh := &counted{}
+	s := &Server{inbound: Nats(second), bus: OverNATS{Conn: second.Conn(), JS: second.Context()},
+		listener: fresh, log: quiet()}
+	ctx, stopNew := context.WithCancel(context.Background())
+	defer stopNew()
+	ended := make(chan error, 1)
+	go func() { ended <- s.Serve(ctx) }()
+
+	select {
+	case err := <-ended:
+		t.Fatalf("the second controller stopped instead of standing by: %v", err)
+	case <-time.After(500 * time.Millisecond):
+	}
+
+	report := func(declared string) {
+		body, _ := json.Marshal(Report{Node: "anchor", Declared: declared, Applied: []string{"store"}})
+		if _, err := js.Context().Publish(ReportSubject("anchor"), body); err != nil {
+			t.Fatal(err)
+		}
+	}
+	report("d1")
+	eventually(t, "the holding controller hearing the report", func() bool { return old.count() == 1 })
+	if fresh.count() != 0 {
+		t.Fatal("the controller standing by acted on a report the holder was handed")
+	}
+
+	stopOld()
+	report("d2")
+	eventually(t, "the second controller taking over once the first let go", func() bool { return fresh.count() == 1 })
+	if old.count() != 1 {
+		t.Errorf("the first controller heard %d reports", old.count())
+	}
+}