Merge pull request 'Judge a duplicate hand-out before marking a build in hand, and never hand every build back over a stale mark (issue 1768)' (#429) from fix/1768-in-hand-marker into main

internal/link/admit_test.go

@@ -0,0 +1,28 @@
+package link
+
+import (
+	"context"
+	"testing"
+
+	"github.com/nats-io/nats.go"
+)
+
+// 1768/AC1: a request dropped as a duplicate of one another builder runs marks nothing in hand: the claim is not made,
+// so no mark is left behind to fail every later claim.
+func TestADuplicateDropIsJudgedBeforeTheClaimSoLeavesNoMark(t *testing.T) {
+	claims := 0
+	m := &natsMachine{
+		opts:      MachineOptions{Claim: func() bool { claims++; return true }},
+		duplicate: func(context.Context, BuildRequest) string { return "ace is running it" },
+	}
+	if m.admit(context.Background(), &nats.Msg{}, BuildRequest{ID: "build-1"}) {
+		t.Fatal("a duplicate was admitted")
+	}
+	if claims != 0 {
+		t.Fatalf("a dropped duplicate was claimed %d times, which leaves the in-hand mark behind", claims)
+	}
+	m.duplicate = func(context.Context, BuildRequest) string { return "" }
+	if !m.admit(context.Background(), &nats.Msg{}, BuildRequest{ID: "build-2"}) || claims != 1 {
+		t.Fatalf("a request that is no duplicate was not claimed once: claims=%d", claims)
+	}
+}

internal/link/builds_nats.go

@@ -162,6 +162,8 @@ type natsMachine struct {
 	// build waits, then the checks of pull requests marked ready, then every other check (novox/hq ADR 0406 §3).
 	subs []*takeQueue
 	opts MachineOptions
+	// duplicate stands in for runningElsewhere in tests; nil is the real one.
+	duplicate func(context.Context, BuildRequest) string
 }
 
 // takeQueue is one queue a machine pulls from.
@@ -356,18 +358,7 @@ func (m *natsMachine) Take(ctx context.Context, do func(context.Context, Build))
 				continue
 			}
 		}
-		// **Drained in the moment it was fetched** (novox/hq ADR 0335): handed back at once, as a pause
-		// landing then is.
-		if m.opts.Claim != nil && !m.opts.Claim() {
-			_ = msg.Nak()
-			continue
-		}
-		// **A duplicate of a check another builder is running is dropped** (issue 1649): an ask redelivered to this machine
-		// because its first holder did not say, in time, that it had it. The running key names that builder and this same
-		// ask; running it a second time would spend a second place on one check.
-		if why := m.runningElsewhere(ctx, request); why != "" {
-			fmt.Fprintf(os.Stderr, "%s is handed out again while another builder runs it and is dropped: %s\n", request.ID, why)
-			_ = msg.Ack()
+		if !m.admit(ctx, msg, request) {
 			continue
 		}
 		// A build outlives the acknowledgement window many times over; said while it runs,
@@ -384,6 +375,33 @@ func (m *natsMachine) Take(ctx context.Context, do func(context.Context, Build))
 	}
 }
 
+// admit decides whether a fetched request is run here, and answers the message when it is not.
+//
+// **The duplicate is judged before the request is claimed** (issue 1768): a claim marks the build in hand, and a request
+// dropped after it left the mark behind, so every later claim failed and every ask was handed back at once until all
+// had spent their deliveries. A request dropped here has marked nothing.
+//
+// **Drained in the moment it was fetched** (novox/hq ADR 0335): handed back at once, as a pause landing then is.
+func (m *natsMachine) admit(ctx context.Context, msg *nats.Msg, request BuildRequest) bool {
+	// **A duplicate of a check another builder is running is dropped** (issue 1649): an ask redelivered to this machine
+	// because its first holder did not say, in time, that it had it. The running key names that builder and this same
+	// ask; running it a second time would spend a second place on one check.
+	elsewhere := m.runningElsewhere
+	if m.duplicate != nil {
+		elsewhere = m.duplicate
+	}
+	if why := elsewhere(ctx, request); why != "" {
+		fmt.Fprintf(os.Stderr, "%s is handed out again while another builder runs it and is dropped: %s\n", request.ID, why)
+		_ = msg.Ack()
+		return false
+	}
+	if m.opts.Claim != nil && !m.opts.Claim() {
+		_ = msg.Nak()
+		return false
+	}
+	return true
+}
+
 // runningElsewhere says why an ask need not run here, or nothing: the bucket of the checks running says another builder
 // runs this very ask (its id, its head and its trunk commit). A key that cannot be read runs the check.
 func (m *natsMachine) runningElsewhere(ctx context.Context, r BuildRequest) string {