long currentTime = roleQueryHelper.getCurrentTime();
        assertTrue(currentTime > 0);

        long currentTimeMillis = System.currentTimeMillis();
        // Allow for small time difference
        assertTrue(Math.abs(currentTime - currentTimeMillis) < 1000);
    }

    @Test
    public void test_constants() {
        assertEquals("userRoles", RoleQueryHelper.USER_ROLES);
    }

# Interval (seconds) for polling heartbeats and events.
coordinator.poll.interval=60
# Time-to-live (ms) for instance heartbeat documents.
coordinator.heartbeat.ttl=180000
# Time-to-live (ms) for operation lock documents.
coordinator.operation.ttl=7200000
# Maximum number of retries for acquiring an operation lock.
coordinator.operation.retry=3
# Time-to-live (ms) for event notification documents.
coordinator.event.ttl=600000

    assertThat(c1e0.connectionIndex).isEqualTo(1) // Fresh connection.
  }

  /**
   * Check we would use an existing connection to a later DNS result instead of connecting to the
   * first DNS result for the first time.
   */
  @Test
  fun prefersExistingCompatible() {
    server.enqueue(MockResponse())
    server.enqueue(MockResponse())
    val connectCount = AtomicInteger()
    val listener: EventListener =

     * and then to XOR them together.  The strideTable enables us to hash an int followed by 12
     * zero bytes (3 ints), while the byteTable is for advancing one byte at a time.
     * This algorithm is due to the paper "Everything we know about CRC but [are] afraid to forget"
     * by Kadatch and Jenkins, 2010.
     */

    Crc32cHasher() {
      super(16);
    }

            deferred.resolve(new SuggestResponse("", 0, Collections.emptyList(), 0, null));
        });
        resolver.start();

        // Register another callback and trigger resolve at about the same time
        Thread.sleep(50);
        resolveLatch.countDown();
        deferred.promise().then(response -> {
            callbackCount.incrementAndGet();
            callbackLatch.countDown();
        });

    private long accessTime = Long.MAX_VALUE;

    @Override
    public long getAccessTime() {
      return accessTime;
    }

    @Override
    public void setAccessTime(long time) {
      this.accessTime = time;
    }

    private ReferenceEntry<K, V> nextAccess = nullEntry();

    @Override
    public ReferenceEntry<K, V> getNextInAccessQueue() {
      return nextAccess;
    }

    private long accessTime = Long.MAX_VALUE;

    @Override
    public long getAccessTime() {
      return accessTime;
    }

    @Override
    public void setAccessTime(long time) {
      this.accessTime = time;
    }

    private ReferenceEntry<K, V> nextAccess = nullEntry();

    @Override
    public ReferenceEntry<K, V> getNextInAccessQueue() {
      return nextAccess;
    }

        // Test with zero
        serverData.serverTime = 0L;
        assertEquals(0L, serverData.serverTime);

        // Test with positive value
        serverData.serverTime = 132514080000000000L; // Windows file time
        assertEquals(132514080000000000L, serverData.serverTime);

        // Test with negative value
        serverData.serverTime = -1L;
        assertEquals(-1L, serverData.serverTime);

    ...
```

この場合、Python は `stanleyjobsox` と `stanleyjobson` の両方で `stanleyjobso` 全体を比較してから、文字列が同じでないと気づきます。したがって、「ユーザー名またはパスワードが正しくありません」と応答するまでに余分に数マイクロ秒かかります。

#### 応答時間が攻撃者を助ける { #the-time-to-answer-helps-the-attackers }

ここで、サーバーが「ユーザー名またはパスワードが正しくありません」というレスポンスを返すまでに、わずかに長い時間がかかったことに気づけば、攻撃者は何かしら正解に近づいた、すなわち先頭のいくつかの文字が正しかったことを知ることができます。

すると、`johndoe` よりも `stanleyjobsox` に近いものを狙って再試行できます。

            Graph graph, Graph.Vertex before, Graph.Vertex after, Collection<Lifecycle.Phase> phases) {
        // We add ordering between internal phases.
        // This would be wrong at execution time, but we are here computing a list and not a graph,
        // so in order to obtain the expected order, we add these links between phases.
        Lifecycle.Phase prev = null;
        for (Lifecycle.Phase child : phases) {