fix(group): bound push pressure and synchronize load agents

loadtest/im-pressure/coordinator_test.go

@@ -15,6 +15,7 @@ func TestCoordinatorResetsBarrierWhenConnectionCountDrops(t *testing.T) {
 		start: StartInfo{RunID: "run-reset", Scenario: "group"}, startDelay: time.Minute,
 		commitDelay: time.Second, expectedAgents: 2, ready: make(map[string]AgentReady),
 		armed: make(map[string]bool), ledgers: make(map[string][]AcceptedMessage),
+		started:   make(map[string]time.Time),
 		summaries: make(map[string]AgentSummary), complete: make(chan struct{}),
 	}
 	postReady := func(ready AgentReady) StartInfo {
@@ -52,6 +53,7 @@ func TestCoordinatorClientWaitsForBothAgentsAndMergesLedger(t *testing.T) {
 		commitDelay: 50 * time.Millisecond, expectedAgents: 2,
 		ready: make(map[string]AgentReady), ledgers: make(map[string][]AcceptedMessage),
 		armed: make(map[string]bool), summaries: make(map[string]AgentSummary), complete: make(chan struct{}),
+		started: make(map[string]time.Time),
 	}
 	server := httptest.NewServer(testCoordinatorMux(state))
 	defer server.Close()
@@ -116,6 +118,7 @@ func testCoordinatorMux(state *coordinatorState) *http.ServeMux {
 	mux := http.NewServeMux()
 	mux.HandleFunc("/v1/ready", state.handleReady)
 	mux.HandleFunc("/v1/start", state.handleStart)
+	mux.HandleFunc("/v1/started", state.handleStarted)
 	mux.HandleFunc("/v1/ledger", state.handleLedger)
 	mux.HandleFunc("/v1/observed", state.handleObserved)
 	mux.HandleFunc("/v1/history", state.handleHistory)
@@ -123,3 +126,50 @@ func testCoordinatorMux(state *coordinatorState) *http.ServeMux {
 	mux.HandleFunc("/v1/status", state.handleStatus)
 	return mux
 }
+
+func TestCoordinatorRejectsLateAgentForBothMachines(t *testing.T) {
+	startAt := time.Now().Add(-maxAgentStartSkew - time.Second).UTC()
+	state := &coordinatorState{
+		start:          StartInfo{RunID: "run-late", Scenario: "group", StartAt: startAt, Committed: true},
+		expectedAgents: 2, started: make(map[string]time.Time),
+	}
+	server := httptest.NewServer(testCoordinatorMux(state))
+	defer server.Close()
+	client := newCoordinatorClient(server.URL)
+	if err := client.started(context.Background(), "b", startAt); err == nil {
+		t.Fatal("late agent must be rejected")
+	}
+	if err := client.started(context.Background(), "a", startAt); err == nil {
+		t.Fatal("the other agent must observe the same rejected run")
+	}
+}
+
+func TestCoordinatorReleasesAgentsOnlyAfterBothReachStart(t *testing.T) {
+	startAt := time.Now().UTC()
+	state := &coordinatorState{
+		start:          StartInfo{RunID: "run-synced", Scenario: "group", StartAt: startAt, Committed: true},
+		expectedAgents: 2, started: make(map[string]time.Time),
+	}
+	server := httptest.NewServer(testCoordinatorMux(state))
+	defer server.Close()
+	client := newCoordinatorClient(server.URL)
+	aDone := make(chan error, 1)
+	go func() { aDone <- client.started(context.Background(), "a", startAt) }()
+	time.Sleep(50 * time.Millisecond)
+	select {
+	case err := <-aDone:
+		t.Fatalf("machine A passed before B arrived: %v", err)
+	default:
+	}
+	if err := client.started(context.Background(), "b", startAt); err != nil {
+		t.Fatalf("machine B start: %v", err)
+	}
+	select {
+	case err := <-aDone:
+		if err != nil {
+			t.Fatalf("machine A start: %v", err)
+		}
+	case <-time.After(time.Second):
+		t.Fatal("machine A was not released")
+	}
+}