Fix singleton session demand identity

cmd/gc/adoption_barrier.go

@@ -133,8 +133,13 @@ func runAdoptionBarrier(
 		// base template name (e.g., "city-worker-3" -> "worker").
 		cfgAgent, isConfigAgent := agentBySession[sessionName]
 		isPoolInstance := false
+		staleSingletonSuffix := false
 		if !isConfigAgent {
-			if base := resolvePoolBase(sessionName, store, cityName, st, agentByQN); base != nil {
+			if base := resolveCanonicalSingletonSuffixBase(sessionName, store, cityName, st, agentByQN); base != nil {
+				cfgAgent = base
+				isConfigAgent = true
+				staleSingletonSuffix = true
+			} else if base := resolvePoolBase(sessionName, store, cityName, st, agentByQN); base != nil {
 				cfgAgent = base
 				isConfigAgent = true
 				isPoolInstance = true
@@ -189,15 +194,19 @@ func runAdoptionBarrier(
 		// instance expansion, to avoid false positives on direct session
 		// names that end in numbers.
 		slot := parsePoolSlot(sessionName)
-		if slot > 0 && isConfigAgent && cfgAgent.SupportsInstanceExpansion() {
+		switch {
+		case slot > 0 && staleSingletonSuffix:
+			fmt.Fprintf(stderr, "adoption barrier: adopting stale singleton suffix session %s as canonical agent %s without pool_slot metadata\n", //nolint:errcheck
+				sessionName, cfgAgent.QualifiedName())
+		case slot > 0 && isConfigAgent && cfgAgent.SupportsInstanceExpansion():
 			detail.PoolSlot = slot
 			meta["pool_slot"] = strconv.Itoa(slot)
 			if maxSess := cfgAgent.EffectiveMaxActiveSessions(); maxSess != nil && *maxSess >= 0 && slot > *maxSess {
 				detail.OutOfBounds = true
 				fmt.Fprintf(stderr, "adoption barrier: %s pool slot %d exceeds max %d (adopt-then-drain)\n", //nolint:errcheck
 					sessionName, slot, *maxSess)
 			}
-		} else if slot > 0 && !isConfigAgent {
+		case slot > 0 && !isConfigAgent:
 			// Defensive log (ga-fiw): a session ending in "-N" did not match
 			// any configured agent — either by exact session name or by pool
 			// base resolution. This is the orphan shape that produced the
@@ -305,6 +314,25 @@ func resolvePoolBase(sessionName string, store beads.Store, cityName, sessionTem
 	return nil
 }
 
+func resolveCanonicalSingletonSuffixBase(sessionName string, store beads.Store, cityName, sessionTemplate string, agentByQN map[string]*config.Agent) *config.Agent {
+	slot := parsePoolSlot(sessionName)
+	if slot == 0 {
+		return nil
+	}
+	suffix := fmt.Sprintf("-%d", slot)
+	baseSessName := sessionName[:len(sessionName)-len(suffix)]
+	for _, a := range agentByQN {
+		if !a.UsesCanonicalSingletonPoolIdentity() {
+			continue
+		}
+		sn := lookupSessionNameOrLegacy(store, cityName, a.QualifiedName(), sessionTemplate)
+		if sn == baseSessName {
+			return a
+		}
+	}
+	return nil
+}
+
 // parsePoolSlot extracts the numeric pool slot from a session name suffix.
 // Returns 0 if no slot suffix is found.
 func parsePoolSlot(sessionName string) int {

cmd/gc/adoption_barrier_test.go

@@ -730,19 +730,15 @@ func TestAdoptionBarrier_SingletonWithNumericSuffix(t *testing.T) {
 	}
 }
 
-// TestAdoptionBarrier_OrphanDashNSessionLogsWarning verifies the ga-fiw
-// defensive log: when a running session ends in "-N" but no configured pool
-// agent claims it (because the matching base agent has max_active_sessions=1
-// and SupportsInstanceExpansion()=false), the barrier still adopts the
-// session but emits a stderr warning so the leak is traceable.
-func TestAdoptionBarrier_OrphanDashNSessionLogsWarning(t *testing.T) {
+func TestAdoptionBarrier_StaleDashNSingletonAdoptsCanonicalIdentity(t *testing.T) {
 	store := beads.NewMemStore()
 	// "refinery-1" looks like a pool instance but the base "refinery" agent
-	// has max_active_sessions=1, so resolvePoolBase rejects the suffix.
+	// has max_active_sessions=1, so it should be treated as stale singleton
+	// state rather than a live pool slot.
 	sp := &fakeAdoptionProvider{running: []string{"refinery-1"}}
 	cfg := &config.City{
 		Agents: []config.Agent{
-			{Name: "refinery", MaxActiveSessions: intPtr(1)},
+			{Name: "refinery", MaxActiveSessions: intPtr(1), ScaleCheck: "printf 1"},
 		},
 	}
 	var stderr bytes.Buffer
@@ -751,14 +747,22 @@ func TestAdoptionBarrier_OrphanDashNSessionLogsWarning(t *testing.T) {
 	if result.Adopted != 1 {
 		t.Errorf("Adopted = %d, want 1", result.Adopted)
 	}
-	if !bytes.Contains(stderr.Bytes(), []byte("refinery-1 ends in -1")) {
-		t.Errorf("stderr missing orphan -N warning; got: %s", stderr.String())
+	if !bytes.Contains(stderr.Bytes(), []byte("adopting stale singleton suffix session refinery-1")) {
+		t.Errorf("stderr missing stale singleton adoption warning; got: %s", stderr.String())
 	}
-	// Verify no pool_slot metadata (we explicitly decline to stamp it).
 	beadList, _ := store.ListByLabel(sessionBeadLabel, 0)
 	for _, b := range beadList {
+		if b.Metadata["agent_name"] != "refinery" {
+			t.Errorf("stale singleton agent_name = %q, want canonical refinery", b.Metadata["agent_name"])
+		}
+		if !containsString(b.Labels, "agent:refinery") {
+			t.Errorf("stale singleton labels = %v, want canonical agent label", b.Labels)
+		}
+		if containsString(b.Labels, "agent:refinery-1") {
+			t.Errorf("stale singleton labels = %v, must not include phantom pool identity", b.Labels)
+		}
 		if b.Metadata["pool_slot"] != "" {
-			t.Errorf("orphan -N session should not have pool_slot metadata, got %q", b.Metadata["pool_slot"])
+			t.Errorf("stale singleton session should not have pool_slot metadata, got %q", b.Metadata["pool_slot"])
 		}
 	}
 }