Bind the have bucket once per pass, bound by the queue's timeout, and stop the pass at the first error (issue 1132)

internal/link/have_test.go

@@ -18,11 +18,14 @@ import (
 // haveLink is a link with a bucket behind it that keeps each key's revision as the bus does.
 type haveLink struct {
 	*quietLink
-	hmu     sync.Mutex
-	kv      map[string]haveEntry
-	sets    []string // keys written, in order
-	events  []string // "report" and "have <key>", in the order they happened
-	failing bool
+	hmu                  sync.Mutex
+	kv                   map[string]haveEntry
+	sets                 []string // keys written, in order
+	events               []string // "report" and "have <key>", in the order they happened
+	failing              bool
+	binds                int
+	setTries             int
+	bindFails, bindStuck bool
 }
 
 type haveEntry struct {
@@ -44,7 +47,22 @@ func (l *haveLink) Report(ctx context.Context, node string, body []byte) error {
 
 func (l *haveLink) Health(context.Context, string, []byte) error { return nil }
 
-func (l *haveLink) HaveGet(_ context.Context, key string) ([]byte, uint64, error) {
+func (l *haveLink) HaveBind(ctx context.Context) (HaveBucket, error) {
+	l.hmu.Lock()
+	l.binds++
+	stuck, refuse := l.bindStuck, l.bindFails
+	l.hmu.Unlock()
+	if stuck {
+		<-ctx.Done()
+		return nil, ctx.Err()
+	}
+	if refuse {
+		return nil, errors.New("no grant for the have bucket")
+	}
+	return l, nil
+}
+
+func (l *haveLink) Get(_ context.Context, key string) ([]byte, uint64, error) {
 	l.hmu.Lock()
 	defer l.hmu.Unlock()
 	e, ok := l.kv[key]
@@ -54,9 +72,10 @@ func (l *haveLink) HaveGet(_ context.Context, key string) ([]byte, uint64, error
 	return e.value, e.rev, nil
 }
 
-func (l *haveLink) HaveSet(_ context.Context, key string, value []byte, revision uint64) error {
+func (l *haveLink) Set(_ context.Context, key string, value []byte, revision uint64) error {
 	l.hmu.Lock()
 	defer l.hmu.Unlock()
+	l.setTries++
 	if l.failing {
 		return errors.New("the bucket refused")
 	}
@@ -289,3 +308,34 @@ func TestASilentReconcileStillWritesTheHaves(t *testing.T) {
 		t.Errorf("an unchanged reconcile wrote %d keys", l.setCount()-before)
 	}
 }
+
+// 1132/AC7: a bind that fails, or never answers, writes nothing and returns within the queue's timeout; a write that
+// fails ends the pass, and is said once.
+func TestAFailedBindOrWriteEndsThePassAndReturnsInTime(t *testing.T) {
+	for name, set := range map[string]func(*haveLink){
+		"a bind refused":            func(l *haveLink) { l.bindFails = true },
+		"a bind that never answers": func(l *haveLink) { l.bindStuck = true },
+	} {
+		l := newHaveLink()
+		set(l)
+		q, said := aHaveQueue(l)
+		q.Timeout = 100 * time.Millisecond
+		start := time.Now()
+		if !q.tell(context.Background(), appliedModules(), applied) {
+			t.Fatalf("%s: the report did not go out", name)
+		}
+		if took := time.Since(start); took > time.Second {
+			t.Errorf("%s: the pass took %s", name, took)
+		}
+		if l.binds != 1 || l.setTries != 0 || len(*said) != 1 {
+			t.Errorf("%s: binds %d, writes tried %d, said %v; want one bind, no write, one line", name, l.binds, l.setTries, *said)
+		}
+	}
+	l := newHaveLink()
+	l.failing = true
+	q, said := aHaveQueue(l)
+	q.tell(context.Background(), appliedModules(), applied)
+	if l.binds != 1 || l.setTries != 1 || len(*said) != 1 {
+		t.Errorf("a failed write: binds %d, writes tried %d, said %v; want one of each and one line", l.binds, l.setTries, *said)
+	}
+}