galaxy-game/gateway/internal/grpcapi/freshness_replay_integration_test.go

package grpcapi

import (
	"context"
	"errors"
	"io"
	"sync"
	"testing"
	"time"

	"galaxy/gateway/internal/replay"
	"galaxy/gateway/internal/session"
	gatewayv1 "galaxy/gateway/proto/galaxy/gateway/v1"

	"github.com/stretchr/testify/assert"
	"github.com/stretchr/testify/require"
	"google.golang.org/grpc"
	"google.golang.org/grpc/codes"
	"google.golang.org/grpc/status"
)

func TestExecuteCommandRejectsStaleTimestamp(t *testing.T) {
	t.Parallel()

	tests := []struct {
		name        string
		timestampMS int64
	}{
		{
			name:        "past window",
			timestampMS: testCurrentTime.Add(-testFreshnessWindow - time.Millisecond).UnixMilli(),
		},
		{
			name:        "future window",
			timestampMS: testCurrentTime.Add(testFreshnessWindow + time.Millisecond).UnixMilli(),
		},
	}

	for _, tt := range tests {
		tt := tt

		t.Run(tt.name, func(t *testing.T) {
			t.Parallel()

			delegate := &recordingEdgeGatewayService{}
			server, runGateway := newTestGateway(t, ServerDependencies{
				Service:      delegate,
				SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
				ReplayStore:  staticReplayStore{},
			})
			defer runGateway.stop(t)

			addr := waitForListenAddr(t, server)
			conn := dialGatewayClient(t, addr)
			defer func() {
				require.NoError(t, conn.Close())
			}()

			client := gatewayv1.NewEdgeGatewayClient(conn)
			_, err := client.ExecuteCommand(context.Background(), newValidExecuteCommandRequestWithTimestamp("device-session-123", "request-123", tt.timestampMS))
			require.Error(t, err)
			assert.Equal(t, codes.FailedPrecondition, status.Code(err))
			assert.Equal(t, "request timestamp is outside the freshness window", status.Convert(err).Message())
			assert.Zero(t, delegate.executeCalls)
		})
	}
}

func TestSubscribeEventsRejectsStaleTimestamp(t *testing.T) {
	t.Parallel()

	tests := []struct {
		name        string
		timestampMS int64
	}{
		{
			name:        "past window",
			timestampMS: testCurrentTime.Add(-testFreshnessWindow - time.Millisecond).UnixMilli(),
		},
		{
			name:        "future window",
			timestampMS: testCurrentTime.Add(testFreshnessWindow + time.Millisecond).UnixMilli(),
		},
	}

	for _, tt := range tests {
		tt := tt

		t.Run(tt.name, func(t *testing.T) {
			t.Parallel()

			delegate := &recordingEdgeGatewayService{}
			server, runGateway := newTestGateway(t, ServerDependencies{
				Service:      delegate,
				SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
				ReplayStore:  staticReplayStore{},
			})
			defer runGateway.stop(t)

			addr := waitForListenAddr(t, server)
			conn := dialGatewayClient(t, addr)
			defer func() {
				require.NoError(t, conn.Close())
			}()

			client := gatewayv1.NewEdgeGatewayClient(conn)
			err := subscribeEventsError(t, context.Background(), client, newValidSubscribeEventsRequestWithTimestamp("device-session-123", "request-123", tt.timestampMS))
			require.Error(t, err)
			assert.Equal(t, codes.FailedPrecondition, status.Code(err))
			assert.Equal(t, "request timestamp is outside the freshness window", status.Convert(err).Message())
			assert.Zero(t, delegate.subscribeCalls)
		})
	}
}

func TestExecuteCommandRejectsReplay(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{}
	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: replayDuplicateBySessionAndRequest(),
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	req := newValidExecuteCommandRequest()

	_, err := client.ExecuteCommand(context.Background(), req)
	require.NoError(t, err)

	_, err = client.ExecuteCommand(context.Background(), req)
	require.Error(t, err)
	assert.Equal(t, codes.FailedPrecondition, status.Code(err))
	assert.Equal(t, "request replay detected", status.Convert(err).Message())
	assert.Equal(t, 1, delegate.executeCalls)
}

func TestSubscribeEventsRejectsReplay(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{}
	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: replayDuplicateBySessionAndRequest(),
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	req := newValidSubscribeEventsRequest()

	stream, err := client.SubscribeEvents(context.Background(), req)
	require.NoError(t, err)
	event := recvBootstrapEvent(t, stream)
	assertServerTimeBootstrapEvent(t, event, newTestResponseSignerPublicKey(), "request-123", "trace-123", testCurrentTime.UnixMilli())
	_, err = stream.Recv()
	require.ErrorIs(t, err, io.EOF)

	err = subscribeEventsError(t, context.Background(), client, req)
	require.Error(t, err)
	assert.Equal(t, codes.FailedPrecondition, status.Code(err))
	assert.Equal(t, "request replay detected", status.Convert(err).Message())
	assert.Equal(t, 1, delegate.subscribeCalls)
}

func TestExecuteCommandAllowsSameRequestIDAcrossDistinctSessions(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{
		executeCommandFunc: func(ctx context.Context, req *gatewayv1.ExecuteCommandRequest) (*gatewayv1.ExecuteCommandResponse, error) {
			return &gatewayv1.ExecuteCommandResponse{RequestId: req.GetRequestId()}, nil
		},
	}

	server, runGateway := newTestGateway(t, ServerDependencies{
		Service: delegate,
		SessionCache: staticSessionCache{
			lookupFunc: func(ctx context.Context, deviceSessionID string) (session.Record, error) {
				return newActiveSessionRecordWithSessionID(deviceSessionID), nil
			},
		},
		ReplayStore: staticReplayStore{
			reserveFunc: replayDuplicateBySessionAndRequest(),
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)

	_, err := client.ExecuteCommand(context.Background(), newValidExecuteCommandRequestWithSessionAndRequestID("device-session-123", "request-shared"))
	require.NoError(t, err)

	_, err = client.ExecuteCommand(context.Background(), newValidExecuteCommandRequestWithSessionAndRequestID("device-session-456", "request-shared"))
	require.NoError(t, err)

	assert.Equal(t, 2, delegate.executeCalls)
}

func TestSubscribeEventsAllowsSameRequestIDAcrossDistinctSessions(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{
		subscribeEventsFunc: func(req *gatewayv1.SubscribeEventsRequest, stream grpc.ServerStreamingServer[gatewayv1.GatewayEvent]) error {
			return nil
		},
	}

	server, runGateway := newTestGateway(t, ServerDependencies{
		Service: delegate,
		SessionCache: staticSessionCache{
			lookupFunc: func(ctx context.Context, deviceSessionID string) (session.Record, error) {
				return newActiveSessionRecordWithSessionID(deviceSessionID), nil
			},
		},
		ReplayStore: staticReplayStore{
			reserveFunc: replayDuplicateBySessionAndRequest(),
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)

	stream, err := client.SubscribeEvents(context.Background(), newValidSubscribeEventsRequestWithSessionAndRequestID("device-session-123", "request-shared"))
	require.NoError(t, err)
	event := recvBootstrapEvent(t, stream)
	assertServerTimeBootstrapEvent(t, event, newTestResponseSignerPublicKey(), "request-shared", "trace-123", testCurrentTime.UnixMilli())
	_, err = stream.Recv()
	require.ErrorIs(t, err, io.EOF)

	stream, err = client.SubscribeEvents(context.Background(), newValidSubscribeEventsRequestWithSessionAndRequestID("device-session-456", "request-shared"))
	require.NoError(t, err)
	event = recvBootstrapEvent(t, stream)
	assertServerTimeBootstrapEvent(t, event, newTestResponseSignerPublicKey(), "request-shared", "trace-123", testCurrentTime.UnixMilli())
	_, err = stream.Recv()
	require.ErrorIs(t, err, io.EOF)

	assert.Equal(t, 2, delegate.subscribeCalls)
}

func TestExecuteCommandRejectsReplayStoreUnavailable(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{}
	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: func(context.Context, string, string, time.Duration) error {
				return errors.New("redis down")
			},
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	_, err := client.ExecuteCommand(context.Background(), newValidExecuteCommandRequest())
	require.Error(t, err)
	assert.Equal(t, codes.Unavailable, status.Code(err))
	assert.Equal(t, "replay store is unavailable", status.Convert(err).Message())
	assert.Zero(t, delegate.executeCalls)
}

func TestSubscribeEventsRejectsReplayStoreUnavailable(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{}
	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: func(context.Context, string, string, time.Duration) error {
				return errors.New("redis down")
			},
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	err := subscribeEventsError(t, context.Background(), client, newValidSubscribeEventsRequest())
	require.Error(t, err)
	assert.Equal(t, codes.Unavailable, status.Code(err))
	assert.Equal(t, "replay store is unavailable", status.Convert(err).Message())
	assert.Zero(t, delegate.subscribeCalls)
}

func TestExecuteCommandFreshRequestReachesDelegateAndUsesDynamicReplayTTL(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{
		executeCommandFunc: func(ctx context.Context, req *gatewayv1.ExecuteCommandRequest) (*gatewayv1.ExecuteCommandResponse, error) {
			return &gatewayv1.ExecuteCommandResponse{RequestId: req.GetRequestId()}, nil
		},
	}

	var reservedDeviceSessionID string
	var reservedRequestID string
	var reservedTTL time.Duration

	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: func(ctx context.Context, deviceSessionID string, requestID string, ttl time.Duration) error {
				reservedDeviceSessionID = deviceSessionID
				reservedRequestID = requestID
				reservedTTL = ttl
				return nil
			},
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	response, err := client.ExecuteCommand(context.Background(), newValidExecuteCommandRequest())
	require.NoError(t, err)
	assert.Equal(t, "request-123", response.GetRequestId())
	assert.Equal(t, "device-session-123", reservedDeviceSessionID)
	assert.Equal(t, "request-123", reservedRequestID)
	assert.Equal(t, testFreshnessWindow, reservedTTL)
	assert.Equal(t, 1, delegate.executeCalls)
}

func TestSubscribeEventsFreshRequestReachesDelegateAndUsesDynamicReplayTTL(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{
		subscribeEventsFunc: func(req *gatewayv1.SubscribeEventsRequest, stream grpc.ServerStreamingServer[gatewayv1.GatewayEvent]) error {
			return nil
		},
	}

	var reservedTTL time.Duration

	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: func(ctx context.Context, deviceSessionID string, requestID string, ttl time.Duration) error {
				assert.Equal(t, "device-session-123", deviceSessionID)
				assert.Equal(t, "request-123", requestID)
				reservedTTL = ttl
				return nil
			},
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	stream, err := client.SubscribeEvents(context.Background(), newValidSubscribeEventsRequest())
	require.NoError(t, err)
	event := recvBootstrapEvent(t, stream)
	assertServerTimeBootstrapEvent(t, event, newTestResponseSignerPublicKey(), "request-123", "trace-123", testCurrentTime.UnixMilli())
	_, err = stream.Recv()
	require.ErrorIs(t, err, io.EOF)
	assert.Equal(t, testFreshnessWindow, reservedTTL)
	assert.Equal(t, 1, delegate.subscribeCalls)
}

func TestExecuteCommandFutureSkewUsesExtendedReplayTTL(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{
		executeCommandFunc: func(ctx context.Context, req *gatewayv1.ExecuteCommandRequest) (*gatewayv1.ExecuteCommandResponse, error) {
			return &gatewayv1.ExecuteCommandResponse{RequestId: req.GetRequestId()}, nil
		},
	}

	var reservedTTL time.Duration

	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: func(ctx context.Context, deviceSessionID string, requestID string, ttl time.Duration) error {
				reservedTTL = ttl
				return nil
			},
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	_, err := client.ExecuteCommand(
		context.Background(),
		newValidExecuteCommandRequestWithTimestamp("device-session-123", "request-123", testCurrentTime.Add(2*time.Minute).UnixMilli()),
	)
	require.NoError(t, err)
	assert.Equal(t, 7*time.Minute, reservedTTL)
	assert.Equal(t, 1, delegate.executeCalls)
}

func TestExecuteCommandBoundaryFreshnessUsesMinimumReplayTTL(t *testing.T) {
	t.Parallel()

	delegate := &recordingEdgeGatewayService{
		executeCommandFunc: func(ctx context.Context, req *gatewayv1.ExecuteCommandRequest) (*gatewayv1.ExecuteCommandResponse, error) {
			return &gatewayv1.ExecuteCommandResponse{RequestId: req.GetRequestId()}, nil
		},
	}

	var reservedTTL time.Duration

	server, runGateway := newTestGateway(t, ServerDependencies{
		Service:      delegate,
		SessionCache: staticSessionCache{lookupFunc: func(context.Context, string) (session.Record, error) { return newActiveSessionRecord(), nil }},
		ReplayStore: staticReplayStore{
			reserveFunc: func(ctx context.Context, deviceSessionID string, requestID string, ttl time.Duration) error {
				reservedTTL = ttl
				return nil
			},
		},
	})
	defer runGateway.stop(t)

	addr := waitForListenAddr(t, server)
	conn := dialGatewayClient(t, addr)
	defer func() {
		require.NoError(t, conn.Close())
	}()

	client := gatewayv1.NewEdgeGatewayClient(conn)
	_, err := client.ExecuteCommand(
		context.Background(),
		newValidExecuteCommandRequestWithTimestamp("device-session-123", "request-123", testCurrentTime.Add(-testFreshnessWindow).UnixMilli()),
	)
	require.NoError(t, err)
	assert.Equal(t, minimumReplayReservationTTL, reservedTTL)
	assert.Equal(t, 1, delegate.executeCalls)
}

func replayDuplicateBySessionAndRequest() func(context.Context, string, string, time.Duration) error {
	var (
		mu   sync.Mutex
		seen = make(map[string]struct{})
	)

	return func(ctx context.Context, deviceSessionID string, requestID string, ttl time.Duration) error {
		mu.Lock()
		defer mu.Unlock()

		key := deviceSessionID + "\x00" + requestID
		if _, ok := seen[key]; ok {
			return replay.ErrDuplicate
		}

		seen[key] = struct{}{}
		return nil
	}
}