OutOfMemoryError when run EnforceRateLimitingPlanTest

[vttranlina]

Description of the bug

Exception: java.lang.OutOfMemoryError thrown from the UncaughtExceptionHandler in thread "lettuce-eventExecutorLoop***

Reproduction Steps

Run repeat until failure EnforceRateLimitingPlanTest (with my PC, it will throw when tests passed repeat at 2xx)

Additional information

• Memory Heap dump file: https://tdrive.linagora.com/shared/111111114AxHLNjGvEGJpA/drive/2rCqjwQPJJivJWemudbsoj/t/26373379ca85b4a6ffdb71a425ad6dd3243c2bba

• Test run log test-run.log

• Another log:

  10:49:24.018 [ERROR] i.n.u.HashedWheelTimer - You are creating too many HashedWheelTimer instances. HashedWheelTimer is a shared resource that must be reused across the JVM, so that only a few instances are created.
  java.lang.OutOfMemoryError: Java heap space

• Redis log: (0 clients connected)

  2023-07-17T03:57:37.173041142Z 1:M 17 Jul 2023 03:57:37.172 . 0 clients connected (0 replicas), 862312 bytes in use
  2023-07-17T03:57:42.185684229Z 1:M 17 Jul 2023 03:57:42.185 . 0 clients connected (0 replicas), 862312 bytes in use

• The Default IDE setting when run Junit: xms=512Mb xmx=1024Mb

[vttranlina]

MyIDE

[vttranlina]

I guess it relates to io.lettuce.core.resource.ClientResources, which has not been closed correctly So I tried to modify org.apache.james.rate.limiter.redis.RedisRateLimiterFactory is bellow but I can't get better

class RedisRateLimiterFactory @Inject()(redisConfiguration: RedisRateLimiterConfiguration) extends RateLimiterFactory {

  var resources: ClientResources = _
  val rateLimitjFactory: AbstractRequestRateLimiterFactory[RedisSlidingWindowRequestRateLimiter] =
    if (redisConfiguration.isCluster) {
      val resourceBuilder = ClientResources.builder()
        .threadFactoryProvider(poolName => NamedThreadFactory.withName(s"redis-driver-$poolName"))
      redisConfiguration.ioThreads.foreach(value => resourceBuilder.ioThreadPoolSize(value))
      redisConfiguration.workerThreads.foreach(value =>resourceBuilder.computationThreadPoolSize(value))
      new RedisClusterRateLimiterFactory(RedisClusterClient.create(resourceBuilder.build(),
        redisConfiguration.redisURI.value.asJava))
    } else {
      val resourceBuilder = ClientResources.builder()
        .threadFactoryProvider(poolName => NamedThreadFactory.withName(s"redis-driver-$poolName"))
      redisConfiguration.ioThreads.foreach(value => resourceBuilder.ioThreadPoolSize(value))
      redisConfiguration.workerThreads.foreach(value => resourceBuilder.computationThreadPoolSize(value))
      resources = resourceBuilder.build();

      new RedisSingleInstanceRateLimitjFactory(RedisClient.create(resources, redisConfiguration.redisURI.value.last))
    }

  def preDestroy(): Unit = {
    resources.shutdown()
  }

[quantranhong1999]

May help to get some insights from the heap dump: https://heaphero.io/heap-report-wc.jsp?p=RThCZ0lOakJ3K2dCd3FiSEc4VEhscytQdnNlRUVoSExkWjdwdlY1T01IUzlVR1JsaGU4dFpzUGtZc1BFdE5vSnQ0SXNuNkNaMzhOdWM2N2hzdVBTZFE9PQ==

5,094 instances of "io.netty.buffer.PoolThreadCache", loaded by "jdk.internal.loader.ClassLoaders$AppClassLoader @ 0xc0b26618" occupy 705,688,952 (66.28%) bytes

My guess: We initiate multiple times the lettuce driver in tests, and in somehow the driver is not freed after each test.

@BeforeEach
redisRateLimiterFactory = new RedisRateLimiterFactory(RedisRateLimiterConfiguration.from(redis.redisURI().toString, isCluster = false))

Not a big deal in tests IMO. But in production RedisRateLimiterFactory should be in SINGLETON scope IMO.

[vttranlina]

local issue