foreach (const Future<TaskStatus>& taskStatus, taskStatuses) {
    AWAIT_READY(taskStatus);

    Option<TaskState> taskState = taskStates.get(taskStatus->task_id());
    ASSERT_SOME(taskState);

    switch (taskState.get()) {
      case TASK_STAGING: {
        ASSERT_EQ(TASK_STARTING, taskStatus->state())
          << taskStatus->DebugString();

        taskStates[taskStatus->task_id()] = TASK_STARTING;
        break;
      }
      case TASK_STARTING: {
        ASSERT_EQ(TASK_RUNNING, taskStatus->state())
          << taskStatus->DebugString();

        taskStates[taskStatus->task_id()] = TASK_RUNNING;
        break;
      }
      default: {
        FAIL() << "Unexpected task update: " << taskStatus->DebugString();
        break;
      }
    }
  }
Esempio n. 2
0
// This test verifies that disk quota isolator recovers properly after
// the slave restarts.
TEST_F(DiskQuotaTest, SlaveRecovery)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "posix/cpu,posix/mem,disk/du";
  flags.container_disk_watch_interval = Milliseconds(1);

  Fetcher fetcher(flags);

  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, true, &fetcher);

  ASSERT_SOME(_containerizer);

  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), containerizer.get(), flags);

  ASSERT_SOME(slave);

  MockScheduler sched;

  // Enable checkpointing for the framework.
  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_checkpoint(true);

  MesosSchedulerDriver driver(
      &sched, frameworkInfo, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return());      // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  const Offer& offer = offers.get()[0];

  // Create a task that uses 2MB disk.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:3").get(),
      "dd if=/dev/zero of=file bs=1048576 count=2 && sleep 1000");

  Future<TaskStatus> status;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&status))
    .WillRepeatedly(Return());       // Ignore subsequent updates.

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(status);
  EXPECT_EQ(task.task_id(), status->task_id());
  EXPECT_EQ(TASK_RUNNING, status->state());

  Future<hashset<ContainerID>> containers = containerizer->containers();

  AWAIT_READY(containers);
  ASSERT_EQ(1u, containers->size());

  const ContainerID& containerId = *(containers->begin());

  // Stop the slave.
  slave.get()->terminate();

  Future<ReregisterExecutorMessage> reregisterExecutorMessage =
    FUTURE_PROTOBUF(ReregisterExecutorMessage(), _, _);

  Future<Nothing> _recover = FUTURE_DISPATCH(_, &Slave::_recover);

  _containerizer = MesosContainerizer::create(flags, true, &fetcher);
  ASSERT_SOME(_containerizer);

  containerizer.reset(_containerizer.get());

  detector = master.get()->createDetector();

  slave = StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  Clock::pause();

  AWAIT_READY(_recover);

  // Wait for slave to schedule reregister timeout.
  Clock::settle();

  // Ensure the executor re-registers before completing recovery.
  AWAIT_READY(reregisterExecutorMessage);

  // Ensure the slave considers itself recovered.
  Clock::advance(flags.executor_reregistration_timeout);

  // NOTE: We resume the clock because we need the reaper to reap the
  // 'du' subprocess.
  Clock::resume();

  // Wait until disk usage can be retrieved.
  Duration elapsed = Duration::zero();
  while (true) {
    Future<ResourceStatistics> usage = containerizer->usage(containerId);
    AWAIT_READY(usage);

    ASSERT_TRUE(usage->has_disk_limit_bytes());
    EXPECT_EQ(Megabytes(3), Bytes(usage->disk_limit_bytes()));

    if (usage->has_disk_used_bytes()) {
      EXPECT_LE(usage->disk_used_bytes(), usage->disk_limit_bytes());

      // NOTE: This is to capture the regression in MESOS-2452. The data
      // stored in the executor meta directory should be less than 64K.
      if (usage->disk_used_bytes() > Kilobytes(64).bytes()) {
        break;
      }
    }

    ASSERT_LT(elapsed, Seconds(15));

    os::sleep(Milliseconds(1));
    elapsed += Milliseconds(1);
  }

  driver.stop();
  driver.join();
}
Esempio n. 3
0
// This test verifies that the container will not be killed if
// disk_enforce_quota flag is false (even if the disk usage exceeds
// its quota).
TEST_F(DiskQuotaTest, NoQuotaEnforcement)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "posix/cpu,posix/mem,disk/du";

  // NOTE: We can't pause the clock because we need the reaper to reap
  // the 'du' subprocess.
  flags.container_disk_watch_interval = Milliseconds(1);
  flags.enforce_container_disk_quota = false;

  Fetcher fetcher(flags);

  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, true, &fetcher);

  ASSERT_SOME(_containerizer);

  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), containerizer.get(), flags);

  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return());      // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  const Offer& offer = offers.get()[0];

  // Create a task that uses 2MB disk.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:1").get(),
      "dd if=/dev/zero of=file bs=1048576 count=2 && sleep 1000");

  Future<TaskStatus> status;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&status))
    .WillRepeatedly(Return());       // Ignore subsequent updates.

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(status);
  EXPECT_EQ(task.task_id(), status->task_id());
  EXPECT_EQ(TASK_RUNNING, status->state());

  Future<hashset<ContainerID>> containers = containerizer->containers();

  AWAIT_READY(containers);
  ASSERT_EQ(1u, containers->size());

  const ContainerID& containerId = *(containers->begin());

  // Wait until disk usage can be retrieved and the usage actually
  // exceeds the limit. If the container is killed due to quota
  // enforcement (which shouldn't happen), the 'usage' call will
  // return a failed future, leading to a failed test.
  Duration elapsed = Duration::zero();
  while (true) {
    Future<ResourceStatistics> usage = containerizer->usage(containerId);
    AWAIT_READY(usage);

    ASSERT_TRUE(usage->has_disk_limit_bytes());
    EXPECT_EQ(Megabytes(1), Bytes(usage->disk_limit_bytes()));

    if (usage->has_disk_used_bytes() &&
        usage->disk_used_bytes() > usage->disk_limit_bytes()) {
      break;
    }

    ASSERT_LT(elapsed, Seconds(5));

    os::sleep(Milliseconds(1));
    elapsed += Milliseconds(1);
  }

  driver.stop();
  driver.join();
}
Esempio n. 4
0
// Test that memory pressure listening is restarted after recovery.
TEST_F(MemoryPressureMesosTest, CGROUPS_ROOT_SlaveRecovery)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();

  // We only care about memory cgroup for this test.
  flags.isolation = "cgroups/mem";

  Fetcher fetcher(flags);

  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, true, &fetcher);

  ASSERT_SOME(_containerizer);
  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  // Enable checkpointing for the framework.
  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_checkpoint(true);

  MesosSchedulerDriver driver(
      &sched, frameworkInfo, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return());      // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  Offer offer = offers.get()[0];

  // Run a task that triggers memory pressure event. We request 1G
  // disk because we are going to write a 512 MB file repeatedly.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:256;disk:1024").get(),
      "while true; do dd count=512 bs=1M if=/dev/zero of=./temp; done");

  Future<TaskStatus> starting;
  Future<TaskStatus> running;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&starting))
    .WillOnce(FutureArg<1>(&running))
    .WillRepeatedly(Return()); // Ignore subsequent updates.

  Future<Nothing> runningAck =
    FUTURE_DISPATCH(_, &Slave::_statusUpdateAcknowledgement);

  Future<Nothing> startingAck =
    FUTURE_DISPATCH(_, &Slave::_statusUpdateAcknowledgement);

  driver.launchTasks(offers.get()[0].id(), {task});

  AWAIT_READY(starting);
  EXPECT_EQ(task.task_id(), starting->task_id());
  EXPECT_EQ(TASK_STARTING, starting->state());

  AWAIT_READY(startingAck);

  AWAIT_READY(running);
  EXPECT_EQ(task.task_id(), running->task_id());
  EXPECT_EQ(TASK_RUNNING, running->state());

  // Wait for the ACK to be checkpointed.
  AWAIT_READY_FOR(runningAck, Seconds(120));

  // We restart the slave to let it recover.
  slave.get()->terminate();

  // Set up so we can wait until the new slave updates the container's
  // resources (this occurs after the executor has re-registered).
  Future<Nothing> update =
    FUTURE_DISPATCH(_, &MesosContainerizerProcess::update);

  // Use the same flags.
  _containerizer = MesosContainerizer::create(flags, true, &fetcher);
  ASSERT_SOME(_containerizer);
  containerizer.reset(_containerizer.get());

  Future<SlaveReregisteredMessage> reregistered =
      FUTURE_PROTOBUF(SlaveReregisteredMessage(), master.get()->pid, _);

  slave = StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  AWAIT_READY(reregistered);

  // Wait until the containerizer is updated.
  AWAIT_READY(update);

  Future<hashset<ContainerID>> containers = containerizer->containers();
  AWAIT_READY(containers);
  ASSERT_EQ(1u, containers->size());

  ContainerID containerId = *(containers->begin());

  // Wait a while for some memory pressure events to occur.
  Duration waited = Duration::zero();
  do {
    Future<ResourceStatistics> usage = containerizer->usage(containerId);
    AWAIT_READY(usage);

    if (usage->mem_low_pressure_counter() > 0) {
      // We will check the correctness of the memory pressure counters
      // later, because the memory-hammering task is still active
      // and potentially incrementing these counters.
      break;
    }

    os::sleep(Milliseconds(100));
    waited += Milliseconds(100);
  } while (waited < Seconds(5));

  EXPECT_LE(waited, Seconds(5));

  // Pause the clock to ensure that the reaper doesn't reap the exited
  // command executor and inform the containerizer/slave.
  Clock::pause();
  Clock::settle();

  Future<TaskStatus> killed;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&killed));

  // Stop the memory-hammering task.
  driver.killTask(task.task_id());

  AWAIT_READY_FOR(killed, Seconds(120));
  EXPECT_EQ(task.task_id(), killed->task_id());
  EXPECT_EQ(TASK_KILLED, killed->state());

  // Now check the correctness of the memory pressure counters.
  Future<ResourceStatistics> usage = containerizer->usage(containerId);
  AWAIT_READY(usage);

  EXPECT_GE(usage->mem_low_pressure_counter(),
            usage->mem_medium_pressure_counter());
  EXPECT_GE(usage->mem_medium_pressure_counter(),
            usage->mem_critical_pressure_counter());

  Clock::resume();

  driver.stop();
  driver.join();
}
// This test checks the behavior of passed invalid limits.
TEST_F(PosixRLimitsIsolatorTest, InvalidLimits)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "posix/rlimits";

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched,
      DEFAULT_FRAMEWORK_INFO,
      master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;

  EXPECT_CALL(sched, resourceOffers(_, _))
      .WillOnce(FutureArg<1>(&offers))
      .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_NE(0u, offers->size());

  TaskInfo task = createTask(
      offers.get()[0].slave_id(),
      offers.get()[0].resources(),
      "true");

  ContainerInfo* container = task.mutable_container();
  container->set_type(ContainerInfo::MESOS);

  // Set impossible limit soft > hard.
  RLimitInfo rlimitInfo;
  RLimitInfo::RLimit* rlimit = rlimitInfo.add_rlimits();
  rlimit->set_type(RLimitInfo::RLimit::RLMT_CPU);
  rlimit->set_soft(100);
  rlimit->set_hard(1);

  container->mutable_rlimit_info()->CopyFrom(rlimitInfo);

  Future<TaskStatus> taskStatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
      .WillOnce(FutureArg<1>(&taskStatus));

  driver.launchTasks(offers.get()[0].id(), {task});

  AWAIT_READY(taskStatus);
  EXPECT_EQ(task.task_id(), taskStatus->task_id());
  EXPECT_EQ(TASK_FAILED, taskStatus->state());
  EXPECT_EQ(TaskStatus::REASON_EXECUTOR_TERMINATED, taskStatus->reason());

  driver.stop();
  driver.join();
}
// Tests that the logrotate container logger only closes FDs when it
// is supposed to and does not interfere with other FDs on the agent.
TEST_F(ContainerLoggerTest, LOGROTATE_ModuleFDOwnership)
{
  // Create a master, agent, and framework.
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  Future<SlaveRegisteredMessage> slaveRegisteredMessage =
    FUTURE_PROTOBUF(SlaveRegisteredMessage(), _, _);

  // We'll need access to these flags later.
  slave::Flags flags = CreateSlaveFlags();

  // Use the non-default container logger that rotates logs.
  flags.container_logger = LOGROTATE_CONTAINER_LOGGER_NAME;

  Fetcher fetcher(flags);

  // We use an actual containerizer + executor since we want something to run.
  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, false, &fetcher);

  ASSERT_SOME(_containerizer);
  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  AWAIT_READY(slaveRegisteredMessage);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  Future<FrameworkID> frameworkId;
  EXPECT_CALL(sched, registered(&driver, _, _))
    .WillOnce(FutureArg<1>(&frameworkId));

  // Wait for an offer, and start a task.
  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();
  AWAIT_READY(frameworkId);

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  // Start a task that will keep running until the end of the test.
  TaskInfo task = createTask(offers.get()[0], "sleep 100");

  Future<TaskStatus> statusStarting;
  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusKilled;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusStarting))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusKilled))
    .WillRepeatedly(Return());       // Ignore subsequent updates.

  driver.launchTasks(offers.get()[0].id(), {task});

  AWAIT_READY(statusStarting);
  EXPECT_EQ(TASK_STARTING, statusStarting->state());

  AWAIT_READY(statusRunning);
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  // Open multiple files, so that we're fairly certain we've opened
  // the same FDs (integers) opened by the container logger.
  vector<int> fds;
  for (int i = 0; i < 50; i++) {
    Try<int> fd = os::open(os::DEV_NULL, O_RDONLY);
    ASSERT_SOME(fd);

    fds.push_back(fd.get());
  }

  // Kill the task, which also kills the executor.
  driver.killTask(statusRunning->task_id());

  AWAIT_READY(statusKilled);
  EXPECT_EQ(TASK_KILLED, statusKilled->state());

  Future<Nothing> executorTerminated =
    FUTURE_DISPATCH(_, &Slave::executorTerminated);

  AWAIT_READY(executorTerminated);

  // Close all the FDs we opened.  Every `close` should succeed.
  foreach (int fd, fds) {
    ASSERT_SOME(os::close(fd));
  }
Esempio n. 7
0
// This test launches a container which has an image and joins host
// network, and then verifies that the container can access Internet.
TEST_F(CniIsolatorTest, ROOT_INTERNET_CURL_LaunchContainerInHostNetwork)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "docker/runtime,filesystem/linux";
  flags.image_providers = "docker";
  flags.docker_store_dir = path::join(sandbox.get(), "store");

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(&driver, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_EQ(1u, offers->size());

  const Offer& offer = offers.get()[0];

  // NOTE: We use a non-shell command here because 'sh' might not be
  // in the PATH. 'alpine' does not specify env PATH in the image.
  CommandInfo command;
  command.set_shell(false);
  command.set_value("/bin/ping");
  command.add_arguments("/bin/ping");
  command.add_arguments("-c1");
  command.add_arguments("google.com");

  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128").get(),
      command);

  Image image;
  image.set_type(Image::DOCKER);
  image.mutable_docker()->set_name("alpine");

  ContainerInfo* container = task.mutable_container();
  container->set_type(ContainerInfo::MESOS);
  container->mutable_mesos()->mutable_image()->CopyFrom(image);

  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusFinished;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusFinished));

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY_FOR(statusRunning, Seconds(60));
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFinished);
  EXPECT_EQ(task.task_id(), statusFinished->task_id());
  EXPECT_EQ(TASK_FINISHED, statusFinished->state());

  driver.stop();
  driver.join();
}
Esempio n. 8
0
TEST_F(ResourceOffersTest, ResourcesGetReofferedAfterTaskInfoError)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  Owned<MasterDetector> detector = master.get()->createDetector();
  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get());
  ASSERT_SOME(slave);

  MockScheduler sched1;
  MesosSchedulerDriver driver1(
      &sched1, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched1, registered(&driver1, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched1, resourceOffers(&driver1, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver1.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  TaskInfo task;
  task.set_name("");
  task.mutable_task_id()->set_value("1");
  task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
  task.mutable_executor()->MergeFrom(DEFAULT_EXECUTOR_INFO);

  Resource* cpus = task.add_resources();
  cpus->set_name("cpus");
  cpus->set_type(Value::SCALAR);
  cpus->mutable_scalar()->set_value(-1);

  Resource* mem = task.add_resources();
  mem->set_name("mem");
  mem->set_type(Value::SCALAR);
  mem->mutable_scalar()->set_value(static_cast<double>(Gigabytes(1).bytes()));

  vector<TaskInfo> tasks;
  tasks.push_back(task);

  Future<TaskStatus> status;
  EXPECT_CALL(sched1, statusUpdate(&driver1, _))
    .WillOnce(FutureArg<1>(&status));

  driver1.launchTasks(offers.get()[0].id(), tasks);

  AWAIT_READY(status);
  EXPECT_EQ(task.task_id(), status->task_id());
  EXPECT_EQ(TASK_ERROR, status->state());
  EXPECT_EQ(TaskStatus::REASON_TASK_INVALID, status->reason());
  EXPECT_TRUE(status->has_message());
  EXPECT_TRUE(strings::contains(status->message(), "Invalid scalar resource"))
    << status->message();

  MockScheduler sched2;
  MesosSchedulerDriver driver2(
      &sched2, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched2, registered(&driver2, _, _));

  EXPECT_CALL(sched2, resourceOffers(&driver2, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver2.start();

  AWAIT_READY(offers);

  driver1.stop();
  driver1.join();

  driver2.stop();
  driver2.join();
}
Esempio n. 9
0
// Verify that we can get accurate resource statistics from the XFS
// disk isolator.
TEST_F(ROOT_XFS_QuotaTest, ResourceStatistics)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();

  Fetcher fetcher(flags);
  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, true, &fetcher);

  ASSERT_SOME(_containerizer);
  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return());      // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  Offer offer = offers.get()[0];

  // Create a task that uses 4 of 3MB disk but doesn't fail. We will verify
  // that the allocated disk is filled.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:3").get(),
      "dd if=/dev/zero of=file bs=1048576 count=4 || sleep 1000");

  Future<TaskStatus> startingStatus;
  Future<TaskStatus> runningStatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&startingStatus))
    .WillOnce(FutureArg<1>(&runningStatus))
    .WillRepeatedly(Return()); // Ignore subsequent updates.

  driver.launchTasks(offers.get()[0].id(), {task});

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), startingStatus->task_id());
  EXPECT_EQ(TASK_STARTING, startingStatus->state());

  AWAIT_READY(runningStatus);
  EXPECT_EQ(task.task_id(), runningStatus->task_id());
  EXPECT_EQ(TASK_RUNNING, runningStatus->state());

  Future<hashset<ContainerID>> containers = containerizer.get()->containers();
  AWAIT_READY(containers);
  ASSERT_EQ(1u, containers->size());

  ContainerID containerId = *(containers->begin());
  Timeout timeout = Timeout::in(Seconds(5));

  while (true) {
    Future<ResourceStatistics> usage = containerizer.get()->usage(containerId);
    AWAIT_READY(usage);

    ASSERT_TRUE(usage->has_disk_limit_bytes());
    EXPECT_EQ(Megabytes(3), Bytes(usage->disk_limit_bytes()));

    if (usage->has_disk_used_bytes()) {
      // Usage must always be <= the limit.
      EXPECT_LE(usage->disk_used_bytes(), usage->disk_limit_bytes());

      // Usage might not be equal to the limit, but it must hit
      // and not exceed the limit.
      if (usage->disk_used_bytes() >= usage->disk_limit_bytes()) {
        EXPECT_EQ(
            usage->disk_used_bytes(), usage->disk_limit_bytes());
        EXPECT_EQ(Megabytes(3), Bytes(usage->disk_used_bytes()));
        break;
      }
    }

    ASSERT_FALSE(timeout.expired());
    os::sleep(Milliseconds(100));
  }

  driver.stop();
  driver.join();
}
Esempio n. 10
0
// This is the same logic as DiskUsageExceedsQuota except we turn off disk quota
// enforcement, so exceeding the quota should be allowed.
TEST_F(ROOT_XFS_QuotaTest, DiskUsageExceedsQuotaNoEnforce)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  Owned<MasterDetector> detector = master.get()->createDetector();

  slave::Flags flags = CreateSlaveFlags();
  flags.enforce_container_disk_quota = false;

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(&driver, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  const Offer& offer = offers.get()[0];

  // Create a task which requests 1MB disk, but actually uses more
  // than 2MB disk.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:1").get(),
      "dd if=/dev/zero of=file bs=1048576 count=2");

  Future<TaskStatus> startingStatus;
  Future<TaskStatus> runningStatus;
  Future<TaskStatus> finishedStatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&startingStatus))
    .WillOnce(FutureArg<1>(&runningStatus))
    .WillOnce(FutureArg<1>(&finishedStatus));

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), startingStatus->task_id());
  EXPECT_EQ(TASK_STARTING, startingStatus->state());

  AWAIT_READY(runningStatus);
  EXPECT_EQ(task.task_id(), runningStatus->task_id());
  EXPECT_EQ(TASK_RUNNING, runningStatus->state());

  // We expect the task to succeed even though it exceeded
  // the disk quota.
  AWAIT_READY(finishedStatus);
  EXPECT_EQ(task.task_id(), finishedStatus->task_id());
  EXPECT_EQ(TASK_FINISHED, finishedStatus->state());

  driver.stop();
  driver.join();
}
Esempio n. 11
0
// Verify that a task that tries to consume more space than it has requested
// is only allowed to consume exactly the assigned resources. We tell dd
// to write 2MB but only give it 1MB of resources and (roughly) verify that
// it exits with a failure (that should be a write error).
TEST_F(ROOT_XFS_QuotaTest, DiskUsageExceedsQuota)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  Owned<MasterDetector> detector = master.get()->createDetector();
  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), CreateSlaveFlags());
  ASSERT_SOME(slave);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(&driver, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  const Offer& offer = offers.get()[0];

  // Create a task which requests 1MB disk, but actually uses more
  // than 2MB disk.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:1").get(),
      "dd if=/dev/zero of=file bs=1048576 count=2");

  Future<TaskStatus> startingStatus;
  Future<TaskStatus> runningStatus;
  Future<TaskStatus> failedStatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&startingStatus))
    .WillOnce(FutureArg<1>(&runningStatus))
    .WillOnce(FutureArg<1>(&failedStatus));

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), startingStatus->task_id());
  EXPECT_EQ(TASK_STARTING, startingStatus->state());

  AWAIT_READY(runningStatus);
  EXPECT_EQ(task.task_id(), runningStatus->task_id());
  EXPECT_EQ(TASK_RUNNING, runningStatus->state());

  AWAIT_READY(failedStatus);
  EXPECT_EQ(task.task_id(), failedStatus->task_id());
  EXPECT_EQ(TASK_FAILED, failedStatus->state());

  // Unlike the 'disk/du' isolator, the reason for task failure
  // should be that dd got an IO error.
  EXPECT_EQ(TaskStatus::SOURCE_EXECUTOR, failedStatus->source());
  EXPECT_EQ("Command exited with status 1", failedStatus->message());

  driver.stop();
  driver.join();
}
// Tests that the task fails when it attempts to write to a persistent volume
// mounted as read-only. Note that although we use a shared persistent volume,
// the behavior is the same for non-shared persistent volumes.
TEST_F(LinuxFilesystemIsolatorMesosTest,
       ROOT_WriteAccessSharedPersistentVolumeReadOnlyMode)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  string registry = path::join(sandbox.get(), "registry");
  AWAIT_READY(DockerArchive::create(registry, "test_image"));

  slave::Flags flags = CreateSlaveFlags();
  flags.resources = "cpus:2;mem:128;disk(role1):128";
  flags.isolation = "filesystem/linux,docker/runtime";
  flags.docker_registry = registry;
  flags.docker_store_dir = path::join(sandbox.get(), "store");
  flags.image_providers = "docker";

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;
  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_roles(0, "role1");
  frameworkInfo.add_capabilities()->set_type(
      FrameworkInfo::Capability::SHARED_RESOURCES);

  MesosSchedulerDriver driver(
      &sched,
      frameworkInfo,
      master.get()->pid,
      DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(&driver, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  // We create a shared volume which shall be used by the task to
  // write to that volume.
  Resource volume = createPersistentVolume(
      Megabytes(4),
      "role1",
      "id1",
      "volume_path",
      None(),
      None(),
      frameworkInfo.principal(),
      true); // Shared volume.

  // The task uses the shared volume as read-only.
  Resource roVolume = volume;
  roVolume.mutable_disk()->mutable_volume()->set_mode(Volume::RO);

  Resources taskResources =
    Resources::parse("cpus:1;mem:64;disk(role1):1").get() + roVolume;

  TaskInfo task = createTask(
      offers.get()[0].slave_id(),
      taskResources,
      "echo hello > volume_path/file");

  // The task fails to write to the volume since the task's resources
  // intends to use the volume as read-only.
  Future<TaskStatus> statusStarting;
  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusFailed;

  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusStarting))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusFailed));

  driver.acceptOffers(
      {offers.get()[0].id()},
      {CREATE(volume),
       LAUNCH({task})});

  AWAIT_READY(statusStarting);
  EXPECT_EQ(task.task_id(), statusStarting->task_id());
  EXPECT_EQ(TASK_STARTING, statusStarting->state());

  AWAIT_READY(statusRunning);
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFailed);
  EXPECT_EQ(task.task_id(), statusFailed->task_id());
  EXPECT_EQ(TASK_FAILED, statusFailed->state());

  driver.stop();
  driver.join();
}
// This test verifies that the volume usage accounting for sandboxes
// with bind-mounted volumes (while linux filesystem isolator is used)
// works correctly by creating a file within the volume the size of
// which exceeds the sandbox quota.
TEST_F(LinuxFilesystemIsolatorMesosTest,
       ROOT_VolumeUsageExceedsSandboxQuota)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  string registry = path::join(sandbox.get(), "registry");
  AWAIT_READY(DockerArchive::create(registry, "test_image"));

  slave::Flags flags = CreateSlaveFlags();
  flags.resources = "cpus:2;mem:128;disk(role1):128";
  flags.isolation = "disk/du,filesystem/linux,docker/runtime";
  flags.docker_registry = registry;
  flags.docker_store_dir = path::join(sandbox.get(), "store");
  flags.image_providers = "docker";

  // NOTE: We can't pause the clock because we need the reaper to reap
  // the 'du' subprocess.
  flags.container_disk_watch_interval = Milliseconds(1);
  flags.enforce_container_disk_quota = true;

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;
  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_roles(0, "role1");

  MesosSchedulerDriver driver(
      &sched,
      frameworkInfo,
      master.get()->pid,
      DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(&driver, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  // We request a sandbox (1MB) that is smaller than the persistent
  // volume (4MB) and attempt to create a file in that volume that is
  // twice the size of the sanbox (2MB).
  Resources volume = createPersistentVolume(
      Megabytes(4),
      "role1",
      "id1",
      "volume_path",
      None(),
      None(),
      frameworkInfo.principal());

  Resources taskResources =
      Resources::parse("cpus:1;mem:64;disk(role1):1").get() + volume;

  // We sleep to give quota enforcement (du) a chance to kick in.
  TaskInfo task = createTask(
      offers.get()[0].slave_id(),
      taskResources,
      "dd if=/dev/zero of=volume_path/file bs=1048576 count=2 && sleep 1");

  Future<TaskStatus> statusStarting;
  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusFinished;

  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusStarting))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusFinished));

  driver.acceptOffers(
      {offers.get()[0].id()},
      {CREATE(volume),
      LAUNCH({task})});

  AWAIT_READY(statusStarting);
  EXPECT_EQ(task.task_id(), statusStarting->task_id());
  EXPECT_EQ(TASK_STARTING, statusStarting->state());

  AWAIT_READY(statusRunning);
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFinished);
  EXPECT_EQ(task.task_id(), statusFinished->task_id());
  EXPECT_EQ(TASK_FINISHED, statusFinished->state());

  driver.stop();
  driver.join();
}
Esempio n. 14
0
// This test verifies that a task group is launched on the agent if the executor
// provides a valid authentication token specifying its own ContainerID.
TEST_F(ExecutorAuthorizationTest, RunTaskGroup)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  // Start an agent with permissive ACLs so that a task can be launched.
  ACLs acls;
  acls.set_permissive(true);

  slave::Flags flags = CreateSlaveFlags();
  flags.acls = acls;

  Owned<MasterDetector> detector = master.get()->createDetector();
  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, frameworkInfo, master.get()->pid, DEFAULT_CREDENTIAL);

  Future<FrameworkID> frameworkId;
  EXPECT_CALL(sched, registered(&driver, _, _))
    .WillOnce(FutureArg<1>(&frameworkId));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(frameworkId);

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  Offer offer = offers.get()[0];

  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:0.5;mem:32").get(),
      "sleep 1000");

  Future<TaskStatus> status;

  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&status));

  Resources executorResources =
    allocatedResources(Resources::parse("cpus:0.1;mem:32;disk:32").get(), "*");

  ExecutorInfo executor;
  executor.mutable_executor_id()->set_value("default");
  executor.set_type(ExecutorInfo::DEFAULT);
  executor.mutable_framework_id()->CopyFrom(frameworkId.get());
  executor.mutable_resources()->CopyFrom(executorResources);

  TaskGroupInfo taskGroup;
  taskGroup.add_tasks()->CopyFrom(task);

  driver.acceptOffers({offer.id()}, {LAUNCH_GROUP(executor, taskGroup)});

  AWAIT_READY(status);

  ASSERT_EQ(task.task_id(), status->task_id());
  EXPECT_EQ(TASK_STARTING, status->state());

  driver.stop();
  driver.join();
}
Esempio n. 15
0
// This test verifies that docker image default entrypoint is executed
// correctly using registry puller. This corresponds to the case in runtime
// isolator logic table: sh=0, value=0, argv=1, entrypoint=1, cmd=0.
TEST_F(DockerRuntimeIsolatorTest,
       ROOT_CURL_INTERNET_DockerDefaultEntryptRegistryPuller)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "docker/runtime,filesystem/linux";
  flags.image_providers = "docker";
  flags.docker_store_dir = path::join(os::getcwd(), "store");

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(&driver, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_EQ(1u, offers->size());

  const Offer& offer = offers.get()[0];

  TaskInfo task;
  task.set_name("test-task");
  task.mutable_task_id()->set_value(UUID::random().toString());
  task.mutable_slave_id()->CopyFrom(offer.slave_id());
  task.mutable_resources()->CopyFrom(Resources::parse("cpus:1;mem:128").get());
  task.mutable_command()->set_shell(false);
  task.mutable_command()->add_arguments("hello world");

  Image image;
  image.set_type(Image::DOCKER);

  // 'mesosphere/inky' image is used in docker containerizer test, which
  // contains entrypoint as 'echo' and cmd as null.
  image.mutable_docker()->set_name("mesosphere/inky");

  ContainerInfo* container = task.mutable_container();
  container->set_type(ContainerInfo::MESOS);
  container->mutable_mesos()->mutable_image()->CopyFrom(image);

  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusFinished;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusFinished));

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY_FOR(statusRunning, Seconds(60));
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFinished);
  EXPECT_EQ(task.task_id(), statusFinished->task_id());
  EXPECT_EQ(TASK_FINISHED, statusFinished->state());

  driver.stop();
  driver.join();
}
Esempio n. 16
0
// This is the same logic as ResourceStatistics, except the task should
// be allowed to exceed the disk quota, and usage statistics should report
// that the quota was exceeded.
TEST_F(ROOT_XFS_QuotaTest, ResourceStatisticsNoEnforce)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.enforce_container_disk_quota = false;

  Fetcher fetcher(flags);
  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, true, &fetcher);

  ASSERT_SOME(_containerizer);
  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);
  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  Offer offer = offers.get()[0];

  // Create a task that uses 4MB of 3MB disk and fails if it can't
  // write the full amount.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:3").get(),
      "dd if=/dev/zero of=file bs=1048576 count=4 && sleep 1000");

  Future<TaskStatus> startingStatus;
  Future<TaskStatus> runningStatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&startingStatus))
    .WillOnce(FutureArg<1>(&runningStatus))
    .WillRepeatedly(Return()); // Ignore subsequent updates.

  driver.launchTasks(offers.get()[0].id(), {task});

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), startingStatus->task_id());
  EXPECT_EQ(TASK_STARTING, startingStatus->state());

  AWAIT_READY(runningStatus);
  EXPECT_EQ(task.task_id(), runningStatus->task_id());
  EXPECT_EQ(TASK_RUNNING, runningStatus->state());

  Future<hashset<ContainerID>> containers = containerizer.get()->containers();
  AWAIT_READY(containers);
  ASSERT_EQ(1u, containers->size());

  ContainerID containerId = *(containers->begin());
  Duration diskTimeout = Seconds(5);
  Timeout timeout = Timeout::in(diskTimeout);

  while (true) {
    Future<ResourceStatistics> usage = containerizer.get()->usage(containerId);
    AWAIT_READY(usage);

    ASSERT_TRUE(usage->has_disk_limit_bytes());
    EXPECT_EQ(Megabytes(3), Bytes(usage->disk_limit_bytes()));

    if (usage->has_disk_used_bytes()) {
      if (usage->disk_used_bytes() >= Megabytes(4).bytes()) {
        break;
      }
    }

    // The stopping condition for this test is that the isolator is
    // able to report that we wrote the full amount of data without
    // being constrained by the task disk limit.
    EXPECT_LE(usage->disk_used_bytes(), Megabytes(4).bytes());

    ASSERT_FALSE(timeout.expired())
      << "Used " << Bytes(usage->disk_used_bytes())
      << " of expected " << Megabytes(4)
      << " within the " << diskTimeout << " timeout";

    os::sleep(Milliseconds(100));
  }

  driver.stop();
  driver.join();
}
Esempio n. 17
0
// This test verifies that docker image default cmd is executed correctly.
// This corresponds to the case in runtime isolator logic table: sh=0,
// value=0, argv=1, entrypoint=0, cmd=1.
TEST_F(DockerRuntimeIsolatorTest, ROOT_DockerDefaultCmdLocalPuller)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  const string directory = path::join(os::getcwd(), "archives");

  Future<Nothing> testImage =
    DockerArchive::create(directory, "alpine", "null", "[\"sh\"]");

  AWAIT_READY(testImage);

  ASSERT_TRUE(os::exists(path::join(directory, "alpine.tar")));

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "docker/runtime,filesystem/linux";
  flags.image_providers = "docker";
  flags.docker_registry = directory;

  // Make docker store directory as a temparary directory. Because the
  // manifest of the test image is changeable, the image cached on
  // previous tests should never be used.
  flags.docker_store_dir = path::join(os::getcwd(), "store");

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(&driver, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_EQ(1u, offers->size());

  const Offer& offer = offers.get()[0];

  TaskInfo task;
  task.set_name("test-task");
  task.mutable_task_id()->set_value(UUID::random().toString());
  task.mutable_slave_id()->CopyFrom(offer.slave_id());
  task.mutable_resources()->CopyFrom(Resources::parse("cpus:1;mem:128").get());
  task.mutable_command()->set_shell(false);
  task.mutable_command()->add_arguments("-c");
  task.mutable_command()->add_arguments("echo 'hello world'");

  Image image;
  image.set_type(Image::DOCKER);
  image.mutable_docker()->set_name("alpine");

  ContainerInfo* container = task.mutable_container();
  container->set_type(ContainerInfo::MESOS);
  container->mutable_mesos()->mutable_image()->CopyFrom(image);

  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusFinished;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusFinished));

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY_FOR(statusRunning, Seconds(60));
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFinished);
  EXPECT_EQ(task.task_id(), statusFinished->task_id());
  EXPECT_EQ(TASK_FINISHED, statusFinished->state());

  driver.stop();
  driver.join();
}
Esempio n. 18
0
// In this test, the framework is not checkpointed. This ensures that when we
// stop the slave, the executor is killed and we will need to recover the
// working directories without getting any checkpointed recovery state.
TEST_F(ROOT_XFS_QuotaTest, NoCheckpointRecovery)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();

  Fetcher fetcher(flags);
  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, true, &fetcher);

  ASSERT_SOME(_containerizer);

  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(
      detector.get(),
      containerizer.get(),
      flags);

  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  Offer offer = offers.get()[0];

  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:1").get(),
      "dd if=/dev/zero of=file bs=1048576 count=1; sleep 1000");

  Future<TaskStatus> runningStatus;
  Future<TaskStatus> startingStatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&startingStatus))
    .WillOnce(FutureArg<1>(&runningStatus))
    .WillOnce(Return());

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), startingStatus->task_id());
  EXPECT_EQ(TASK_STARTING, startingStatus->state());

  AWAIT_READY(runningStatus);
  EXPECT_EQ(task.task_id(), runningStatus->task_id());
  EXPECT_EQ(TASK_RUNNING, runningStatus->state());

  Future<ResourceUsage> usage1 =
    process::dispatch(slave.get()->pid, &Slave::usage);
  AWAIT_READY(usage1);

  // We should have 1 executor using resources.
  ASSERT_EQ(1, usage1->executors().size());

  Future<hashset<ContainerID>> containers = containerizer->containers();

  AWAIT_READY(containers);
  ASSERT_EQ(1u, containers->size());

  ContainerID containerId = *containers->begin();

  // Restart the slave.
  slave.get()->terminate();

  Future<SlaveReregisteredMessage> slaveReregisteredMessage =
    FUTURE_PROTOBUF(SlaveReregisteredMessage(), _, _);

  _containerizer = MesosContainerizer::create(flags, true, &fetcher);
  ASSERT_SOME(_containerizer);

  containerizer.reset(_containerizer.get());

  slave = StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  // Wait until slave recovery is complete.
  Future<Nothing> _recover = FUTURE_DISPATCH(_, &Slave::_recover);
  AWAIT_READY_FOR(_recover, Seconds(60));

  // Wait until the orphan containers are cleaned up.
  AWAIT_READY_FOR(containerizer.get()->wait(containerId), Seconds(60));
  AWAIT_READY(slaveReregisteredMessage);

  Future<ResourceUsage> usage2 =
    process::dispatch(slave.get()->pid, &Slave::usage);
  AWAIT_READY(usage2);

  // We should have no executors left because we didn't checkpoint.
  ASSERT_TRUE(usage2->executors().empty());

  Try<std::list<string>> sandboxes = os::glob(path::join(
      slave::paths::getSandboxRootDir(mountPoint.get()),
      "*",
      "frameworks",
      "*",
      "executors",
      "*",
      "runs",
      "*"));

  ASSERT_SOME(sandboxes);

  // One sandbox and one symlink.
  ASSERT_EQ(2u, sandboxes->size());

  // Scan the remaining sandboxes and make sure that no projects are assigned.
  foreach (const string& sandbox, sandboxes.get()) {
    // Skip the "latest" symlink.
    if (os::stat::islink(sandbox)) {
      continue;
    }

    EXPECT_NONE(xfs::getProjectId(sandbox));
  }

  driver.stop();
  driver.join();
}
Esempio n. 19
0
// This test launches a command task which has checkpoint enabled, and
// agent is terminated when the task is running, after agent is restarted,
// kill the task and then verify we can receive TASK_KILLED for the task.
TEST_F(CniIsolatorTest, ROOT_SlaveRecovery)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "network/cni";

  flags.network_cni_plugins_dir = cniPluginDir;
  flags.network_cni_config_dir = cniConfigDir;

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  // Enable checkpointing for the framework.
  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_checkpoint(true);

  MesosSchedulerDriver driver(
      &sched, frameworkInfo, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(&driver, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_EQ(1u, offers->size());

  const Offer& offer = offers.get()[0];

  CommandInfo command;
  command.set_value("sleep 1000");

  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128").get(),
      command);

  ContainerInfo* container = task.mutable_container();
  container->set_type(ContainerInfo::MESOS);

  // Make sure the container join the mock CNI network.
  container->add_network_infos()->set_name("__MESOS_TEST__");

  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusKilled;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusKilled));

  EXPECT_CALL(sched, offerRescinded(&driver, _))
    .Times(AtMost(1));

  Future<Nothing> ack =
    FUTURE_DISPATCH(_, &Slave::_statusUpdateAcknowledgement);

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(statusRunning);
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  // Wait for the ACK to be checkpointed.
  AWAIT_READY(ack);

  // Stop the slave after TASK_RUNNING is received.
  slave.get()->terminate();

  // Restart the slave.
  slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  // Kill the task.
  driver.killTask(task.task_id());

  AWAIT_READY(statusKilled);
  EXPECT_EQ(task.task_id(), statusKilled->task_id());
  EXPECT_EQ(TASK_KILLED, statusKilled->state());

  driver.stop();
  driver.join();
}
Esempio n. 20
0
// In this test, the framework is checkpointed so we expect the executor to
// persist across the slave restart and to have the same resource usage before
// and after.
TEST_F(ROOT_XFS_QuotaTest, CheckpointRecovery)
{
  slave::Flags flags = CreateSlaveFlags();
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  Owned<MasterDetector> detector = master.get()->createDetector();
  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), CreateSlaveFlags());
  ASSERT_SOME(slave);

  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_checkpoint(true);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, frameworkInfo, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  Offer offer = offers.get()[0];

  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:1").get(),
      "dd if=/dev/zero of=file bs=1048576 count=1; sleep 1000");

  Future<TaskStatus> startingStatus;
  Future<TaskStatus> runningStatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&startingStatus))
    .WillOnce(FutureArg<1>(&runningStatus));

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), startingStatus->task_id());
  EXPECT_EQ(TASK_STARTING, startingStatus->state());

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), runningStatus->task_id());
  EXPECT_EQ(TASK_RUNNING, runningStatus->state());

  Future<ResourceUsage> usage1 =
    process::dispatch(slave.get()->pid, &Slave::usage);
  AWAIT_READY(usage1);

  // We should have 1 executor using resources.
  ASSERT_EQ(1, usage1->executors().size());

  // Restart the slave.
  slave.get()->terminate();

  Future<SlaveReregisteredMessage> slaveReregisteredMessage =
    FUTURE_PROTOBUF(SlaveReregisteredMessage(), _, _);

  slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  // Wait for the slave to re-register.
  AWAIT_READY(slaveReregisteredMessage);

  Future<ResourceUsage> usage2 =
    process::dispatch(slave.get()->pid, &Slave::usage);
  AWAIT_READY(usage2);

  // We should have still have 1 executor using resources.
  ASSERT_EQ(1, usage1->executors().size());

  Try<std::list<string>> sandboxes = os::glob(path::join(
      slave::paths::getSandboxRootDir(mountPoint.get()),
      "*",
      "frameworks",
      "*",
      "executors",
      "*",
      "runs",
      "*"));

  ASSERT_SOME(sandboxes);

  // One sandbox and one symlink.
  ASSERT_EQ(2u, sandboxes->size());

  // Scan the remaining sandboxes. We ought to still have project IDs
  // assigned to them all.
  foreach (const string& sandbox, sandboxes.get()) {
    // Skip the "latest" symlink.
    if (os::stat::islink(sandbox)) {
      continue;
    }

    EXPECT_SOME(xfs::getProjectId(sandbox));
  }

  driver.stop();
  driver.join();
}
Esempio n. 21
0
TEST_F(MemoryPressureMesosTest, CGROUPS_ROOT_Statistics)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();

  // We only care about memory cgroup for this test.
  flags.isolation = "cgroups/mem";
  flags.agent_subsystems = None();

  Fetcher fetcher;

  Try<MesosContainerizer*> _containerizer =
    MesosContainerizer::create(flags, true, &fetcher);

  ASSERT_SOME(_containerizer);
  Owned<MesosContainerizer> containerizer(_containerizer.get());

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave =
    StartSlave(detector.get(), containerizer.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched, DEFAULT_FRAMEWORK_INFO, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return());      // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  EXPECT_NE(0u, offers.get().size());

  Offer offer = offers.get()[0];

  // Run a task that triggers memory pressure event. We request 1G
  // disk because we are going to write a 512 MB file repeatedly.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:256;disk:1024").get(),
      "while true; do dd count=512 bs=1M if=/dev/zero of=./temp; done");

  Future<TaskStatus> running;
  Future<TaskStatus> killed;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&running))
    .WillOnce(FutureArg<1>(&killed))
    .WillRepeatedly(Return());       // Ignore subsequent updates.

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(running);
  EXPECT_EQ(task.task_id(), running.get().task_id());
  EXPECT_EQ(TASK_RUNNING, running.get().state());

  Future<hashset<ContainerID>> containers = containerizer->containers();
  AWAIT_READY(containers);
  ASSERT_EQ(1u, containers.get().size());

  ContainerID containerId = *(containers.get().begin());

  // Wait a while for some memory pressure events to occur.
  Duration waited = Duration::zero();
  do {
    Future<ResourceStatistics> usage = containerizer->usage(containerId);
    AWAIT_READY(usage);

    if (usage.get().mem_low_pressure_counter() > 0) {
      // We will check the correctness of the memory pressure counters
      // later, because the memory-hammering task is still active
      // and potentially incrementing these counters.
      break;
    }

    os::sleep(Milliseconds(100));
    waited += Milliseconds(100);
  } while (waited < Seconds(5));

  EXPECT_LE(waited, Seconds(5));

  // Pause the clock to ensure that the reaper doesn't reap the exited
  // command executor and inform the containerizer/slave.
  Clock::pause();
  Clock::settle();

  // Stop the memory-hammering task.
  driver.killTask(task.task_id());

  AWAIT_READY_FOR(killed, Seconds(120));
  EXPECT_EQ(task.task_id(), killed->task_id());
  EXPECT_EQ(TASK_KILLED, killed->state());

  // Now check the correctness of the memory pressure counters.
  Future<ResourceStatistics> usage = containerizer->usage(containerId);
  AWAIT_READY(usage);

  EXPECT_GE(usage.get().mem_low_pressure_counter(),
            usage.get().mem_medium_pressure_counter());
  EXPECT_GE(usage.get().mem_medium_pressure_counter(),
            usage.get().mem_critical_pressure_counter());

  Clock::resume();

  driver.stop();
  driver.join();
}
Esempio n. 22
0
// In this test, the agent initially doesn't enable disk isolation
// but then restarts with XFS disk isolation enabled. We verify that
// the old container launched before the agent restart is
// successfully recovered.
TEST_F(ROOT_XFS_QuotaTest, RecoverOldContainers)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  Owned<MasterDetector> detector = master.get()->createDetector();

  slave::Flags flags = CreateSlaveFlags();

  // `CreateSlaveFlags()` enables `disk/xfs` so here we reset
  // `isolation` to empty.
  flags.isolation.clear();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_checkpoint(true);

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched, frameworkInfo, master.get()->pid, DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;
  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  Offer offer = offers.get()[0];

  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:128;disk:1").get(),
      "dd if=/dev/zero of=file bs=1024 count=1; sleep 1000");

  Future<TaskStatus> startingStatus;
  Future<TaskStatus> runningstatus;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&startingStatus))
    .WillOnce(FutureArg<1>(&runningstatus));

  driver.launchTasks(offer.id(), {task});

  AWAIT_READY(startingStatus);
  EXPECT_EQ(task.task_id(), startingStatus->task_id());
  EXPECT_EQ(TASK_STARTING, startingStatus->state());

  AWAIT_READY(runningstatus);
  EXPECT_EQ(task.task_id(), runningstatus->task_id());
  EXPECT_EQ(TASK_RUNNING, runningstatus->state());

  {
    Future<ResourceUsage> usage =
      process::dispatch(slave.get()->pid, &Slave::usage);
    AWAIT_READY(usage);

    // We should have 1 executor using resources but it doesn't have
    // disk limit enabled.
    ASSERT_EQ(1, usage->executors().size());
    const ResourceUsage_Executor& executor = usage->executors().Get(0);
    ASSERT_TRUE(executor.has_statistics());
    ASSERT_FALSE(executor.statistics().has_disk_limit_bytes());
  }

  // Restart the slave.
  slave.get()->terminate();

  Future<SlaveReregisteredMessage> slaveReregisteredMessage =
    FUTURE_PROTOBUF(SlaveReregisteredMessage(), _, _);

  // This time use the agent flags that include XFS disk isolation.
  slave = StartSlave(detector.get(), CreateSlaveFlags());
  ASSERT_SOME(slave);

  // Wait for the slave to re-register.
  AWAIT_READY(slaveReregisteredMessage);

  {
    Future<ResourceUsage> usage =
      process::dispatch(slave.get()->pid, &Slave::usage);
    AWAIT_READY(usage);

    // We should still have 1 executor using resources but it doesn't
    // have disk limit enabled.
    ASSERT_EQ(1, usage->executors().size());
    const ResourceUsage_Executor& executor = usage->executors().Get(0);
    ASSERT_TRUE(executor.has_statistics());
    ASSERT_FALSE(executor.statistics().has_disk_limit_bytes());
  }

  driver.stop();
  driver.join();
}
// This test confirms that if a task exceeds configured resource
// limits it is forcibly terminated.
TEST_F(PosixRLimitsIsolatorTest, TaskExceedingLimit)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "posix/rlimits";

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched,
      DEFAULT_FRAMEWORK_INFO,
      master.get()->pid,
      DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;

  EXPECT_CALL(sched, resourceOffers(_, _))
    .WillOnce(FutureArg<1>(&offers))
    .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());

  // The task attempts to use an infinite amount of CPU time.
  TaskInfo task = createTask(
      offers.get()[0].slave_id(),
      offers.get()[0].resources(),
      "while true; do true; done");

  ContainerInfo* container = task.mutable_container();
  container->set_type(ContainerInfo::MESOS);

  // Limit the process to use maximally 1 second of CPU time.
  RLimitInfo rlimitInfo;
  RLimitInfo::RLimit* cpuLimit = rlimitInfo.add_rlimits();
  cpuLimit->set_type(RLimitInfo::RLimit::RLMT_CPU);
  cpuLimit->set_soft(1);
  cpuLimit->set_hard(1);

  container->mutable_rlimit_info()->CopyFrom(rlimitInfo);

  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusFailed;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&statusRunning))
    .WillOnce(FutureArg<1>(&statusFailed));

  driver.launchTasks(offers.get()[0].id(), {task});

  AWAIT_READY(statusRunning);
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFailed);
  EXPECT_EQ(task.task_id(), statusFailed->task_id());
  EXPECT_EQ(TASK_FAILED, statusFailed->state());

  driver.stop();
  driver.join();
}
// This test confirms that setting no values for the soft and hard
// limits implies an unlimited resource.
TEST_F(PosixRLimitsIsolatorTest, UnsetLimits) {
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = "posix/rlimits";

  Owned<MasterDetector> detector = master.get()->createDetector();

  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), flags);
  ASSERT_SOME(slave);

  MockScheduler sched;

  MesosSchedulerDriver driver(
      &sched,
      DEFAULT_FRAMEWORK_INFO,
      master.get()->pid,
      DEFAULT_CREDENTIAL);

  EXPECT_CALL(sched, registered(_, _, _));

  Future<vector<Offer>> offers;

  EXPECT_CALL(sched, resourceOffers(_, _))
      .WillOnce(FutureArg<1>(&offers))
      .WillRepeatedly(Return()); // Ignore subsequent offers.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_NE(0u, offers->size());

  TaskInfo task = createTask(
      offers.get()[0].slave_id(),
      offers.get()[0].resources(),
      "exit `ulimit -c | grep -q unlimited`");

  // Force usage of C locale as we interpret a potentially translated
  // string in the task's command.
  mesos::Environment::Variable* locale =
      task.mutable_command()->mutable_environment()->add_variables();
  locale->set_name("LC_ALL");
  locale->set_value("C");

  ContainerInfo* container = task.mutable_container();
  container->set_type(ContainerInfo::MESOS);

  // Setting rlimit for core without soft or hard limit signifies
  // unlimited range.
  RLimitInfo rlimitInfo;
  RLimitInfo::RLimit* rlimit = rlimitInfo.add_rlimits();
  rlimit->set_type(RLimitInfo::RLimit::RLMT_CORE);

  container->mutable_rlimit_info()->CopyFrom(rlimitInfo);

  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusFinal;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
      .WillOnce(FutureArg<1>(&statusRunning))
      .WillOnce(FutureArg<1>(&statusFinal));

  driver.launchTasks(offers.get()[0].id(), {task});

  AWAIT_READY(statusRunning);
  EXPECT_EQ(task.task_id(), statusRunning->task_id());
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFinal);
  EXPECT_EQ(task.task_id(), statusFinal->task_id());
  EXPECT_EQ(TASK_FINISHED, statusFinal->state());

  driver.stop();
  driver.join();
}