// This is a simple end to end test that makes sure a master using log
// storage with ZooKeeper can successfully launch a task.
TEST_F(RegistrarZooKeeperTest, TaskRunning)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  MockExecutor exec(DEFAULT_EXECUTOR_ID);
  TestContainerizer containerizer(&exec);

  Owned<MasterDetector> detector = master.get()->createDetector();
  Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), &containerizer);
  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);
  EXPECT_NE(0u, offers->size());

  TaskInfo task = createTask(offers.get()[0], "dummy", DEFAULT_EXECUTOR_ID);

  EXPECT_CALL(exec, registered(_, _, _, _));

  EXPECT_CALL(exec, launchTask(_, _))
    .WillOnce(SendStatusUpdateFromTask(TASK_RUNNING));

  Future<Nothing> resourcesUpdated;
  EXPECT_CALL(containerizer,
              update(_, Resources(offers.get()[0].resources())))
    .WillOnce(DoAll(FutureSatisfy(&resourcesUpdated),
                    Return(Nothing())));

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

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

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

  AWAIT_READY(resourcesUpdated);

  EXPECT_CALL(exec, shutdown(_))
    .Times(AtMost(1));

  driver.stop();
  driver.join();
}
// This test ensures that the command executor does not send
// TASK_KILLING to frameworks that do not support the capability.
TEST_P_TEMP_DISABLED_ON_WINDOWS(CommandExecutorTest, NoTaskKillingCapability)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

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

  slave::Flags flags = CreateSlaveFlags();
  flags.http_command_executor = GetParam();

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

  // Start the framework without the task killing capability.
  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);
  EXPECT_EQ(1u, offers->size());

  // Launch a task with the command executor.
  TaskInfo task = createTask(
      offers->front().slave_id(),
      offers->front().resources(),
      "sleep 1000");

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

  driver.launchTasks(offers->front().id(), {task});

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

  // There should only be a TASK_KILLED update.
  Future<TaskStatus> statusKilled;
  EXPECT_CALL(sched, statusUpdate(_, _))
    .WillOnce(FutureArg<1>(&statusKilled));

  driver.killTask(task.task_id());

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

  driver.stop();
  driver.join();
}
Esempio n. 3
0
TEST_F_TEMP_DISABLED_ON_WINDOWS(
    ResourceOffersTest,
    ResourceOfferWithMultipleSlaves)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  Owned<MasterDetector> detector = master.get()->createDetector();
  vector<Owned<cluster::Slave>> slaves;

  // Start 10 slaves.
  for (int i = 0; i < 10; i++) {
    slave::Flags flags = CreateSlaveFlags();
    flags.launcher = "posix";

    flags.resources = Option<std::string>("cpus:2;mem:1024");

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

  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()); // All 10 slaves might not be in first offer.

  driver.start();

  AWAIT_READY(offers);
  ASSERT_FALSE(offers->empty());
  EXPECT_GE(10u, offers->size());

  Resources resources(offers.get()[0].resources());
  EXPECT_EQ(2, resources.get<Value::Scalar>("cpus")->value());
  EXPECT_EQ(1024, resources.get<Value::Scalar>("mem")->value());

  driver.stop();
  driver.join();
}
Esempio n. 4
0
TEST_F(ResourceOffersTest, Request)
{
  TestAllocator<master::allocator::HierarchicalDRFAllocator> allocator;

  EXPECT_CALL(allocator, initialize(_, _, _, _, _, _));

  Try<Owned<cluster::Master>> master = StartMaster(&allocator);
  ASSERT_SOME(master);

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

  EXPECT_CALL(allocator, addFramework(_, _, _, _, _));

  Future<Nothing> registered;
  EXPECT_CALL(sched, registered(&driver, _, _))
    .WillOnce(FutureSatisfy(&registered));

  driver.start();

  AWAIT_READY(registered);

  vector<Request> sent;
  Request request;
  request.mutable_slave_id()->set_value("test");
  sent.push_back(request);

  Future<vector<Request>> received;
  EXPECT_CALL(allocator, requestResources(_, _))
    .WillOnce(FutureArg<1>(&received));

  driver.requestResources(sent);

  AWAIT_READY(received);
  EXPECT_EQ(sent.size(), received->size());
  EXPECT_FALSE(received->empty());
  EXPECT_EQ(request.slave_id(), received.get()[0].slave_id());

  driver.stop();
  driver.join();
}
Esempio n. 5
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();
}
// 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. 7
0
TEST_P(CpuIsolatorTest, ROOT_UserCpuUsage)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = GetParam();

  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());

  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());

  // Max out a single core in userspace. This will run for at most one
  // second.
  TaskInfo task = createTask(
      offers.get()[0],
      "while true ; do true ; done & sleep 60");

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

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

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

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

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

  // Wait up to 1 second for the child process to induce 1/8 of a
  // second of user cpu time.
  ResourceStatistics statistics;
  Duration waited = Duration::zero();
  do {
    Future<ResourceStatistics> usage = containerizer->usage(containerId);
    AWAIT_READY(usage);

    statistics = usage.get();

    // If we meet our usage expectations, we're done!
    if (statistics.cpus_user_time_secs() >= 0.125) {
      break;
    }

    os::sleep(Milliseconds(200));
    waited += Milliseconds(200);
  } while (waited < Seconds(1));

  EXPECT_LE(0.125, statistics.cpus_user_time_secs());

  driver.stop();
  driver.join();
}
Esempio n. 8
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();
}
// 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. 10
0
// This test ensures that driver based schedulers using explicit
// acknowledgements can acknowledge status updates sent from
// HTTP based executors.
TEST_F_TEMP_DISABLED_ON_WINDOWS(
    HTTPCommandExecutorTest,
    ExplicitAcknowledgements)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

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

  slave::Flags flags = CreateSlaveFlags();
  flags.http_command_executor = true;

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

  MockScheduler sched;
  MesosSchedulerDriver driver(
      &sched,
      DEFAULT_FRAMEWORK_INFO,
      master.get()->pid,
      false,
      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);
  EXPECT_EQ(1u, offers->size());

  // Launch a task with the command executor.
  TaskInfo task = createTask(
      offers->front().slave_id(),
      offers->front().resources(),
      "sleep 1000");

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

  // Ensure no status update acknowledgements are sent from the driver
  // to the master until the explicit acknowledgement is sent.
  EXPECT_NO_FUTURE_CALLS(
      mesos::scheduler::Call(),
      mesos::scheduler::Call::ACKNOWLEDGE,
      _ ,
      master.get()->pid);

  driver.launchTasks(offers->front().id(), {task});

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

  // Now send the acknowledgement.
  Future<mesos::scheduler::Call> acknowledgement = FUTURE_CALL(
      mesos::scheduler::Call(),
      mesos::scheduler::Call::ACKNOWLEDGE,
      _,
      master.get()->pid);

  driver.acknowledgeStatusUpdate(statusRunning.get());

  AWAIT_READY(acknowledgement);

  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_NE(0u, offers->size());

  // 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();
}
Esempio n. 12
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. 13
0
// This test ensures that a task will transition straight from `TASK_KILLING` to
// `TASK_KILLED`, even if the health check begins to fail during the kill policy
// grace period.
//
// TODO(gkleiman): this test takes about 7 seconds to run, consider using mock
// tasks and health checkers to speed it up.
TEST_P(CommandExecutorTest, NoTransitionFromKillingToRunning)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

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

  slave::Flags flags = CreateSlaveFlags();
  flags.http_command_executor = GetParam();

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

  // Start the framework with the task killing capability.
  FrameworkInfo::Capability capability;
  capability.set_type(FrameworkInfo::Capability::TASK_KILLING_STATE);

  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.add_capabilities()->CopyFrom(capability);

  MockScheduler sched;
  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);
  EXPECT_EQ(1u, offers->size());

  const string command = strings::format(
      "%s %s --sleep_duration=15",
      getTestHelperPath("test-helper"),
      KillPolicyTestHelper::NAME).get();

  TaskInfo task = createTask(offers->front(), command);

  // Create a health check that succeeds until a temporary file is removed.
  Try<string> temporaryPath = os::mktemp(path::join(os::getcwd(), "XXXXXX"));
  ASSERT_SOME(temporaryPath);
  const string tmpPath = temporaryPath.get();

  HealthCheck healthCheck;
  healthCheck.set_type(HealthCheck::COMMAND);
  healthCheck.mutable_command()->set_value("ls " + tmpPath + " >/dev/null");
  healthCheck.set_delay_seconds(0);
  healthCheck.set_grace_period_seconds(0);
  healthCheck.set_interval_seconds(0);

  task.mutable_health_check()->CopyFrom(healthCheck);

  // Set the kill policy grace period to 5 seconds.
  KillPolicy killPolicy;
  killPolicy.mutable_grace_period()->set_nanoseconds(Seconds(5).ns());

  task.mutable_kill_policy()->CopyFrom(killPolicy);

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

  Future<TaskStatus> statusRunning;
  Future<TaskStatus> statusHealthy;
  Future<TaskStatus> statusKilling;
  Future<TaskStatus> statusKilled;

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

  driver.launchTasks(offers->front().id(), tasks);

  AWAIT_READY(statusRunning);
  EXPECT_EQ(TASK_RUNNING, statusRunning.get().state());

  AWAIT_READY(statusHealthy);
  EXPECT_EQ(TASK_RUNNING, statusHealthy.get().state());
  EXPECT_TRUE(statusHealthy.get().has_healthy());
  EXPECT_TRUE(statusHealthy.get().healthy());

  driver.killTask(task.task_id());

  AWAIT_READY(statusKilling);
  EXPECT_EQ(TASK_KILLING, statusKilling->state());
  EXPECT_FALSE(statusKilling.get().has_healthy());

  // Remove the temporary file, so that the health check fails.
  os::rm(tmpPath);

  AWAIT_READY(statusKilled);
  EXPECT_EQ(TASK_KILLED, statusKilled->state());
  EXPECT_FALSE(statusKilled.get().has_healthy());

  driver.stop();
  driver.join();
}
Esempio n. 14
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();
}
// 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();
}
Esempio n. 16
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();
}
// This test verifies that persistent volumes are unmounted properly
// after a checkpointed framework disappears and the slave restarts.
//
// TODO(jieyu): Even though the command task specifies a new
// filesystem root, the executor (command executor) itself does not
// change filesystem root (uses the host filesystem). We need to add a
// test to test the scenario that the executor itself changes rootfs.
TEST_F(LinuxFilesystemIsolatorMesosTest,
       ROOT_RecoverOrphanedPersistentVolume)
{
  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:1024;disk(role1):1024";
  flags.isolation = "filesystem/linux,docker/runtime";
  flags.docker_registry = registry;
  flags.docker_store_dir = path::join(sandbox.get(), "store");
  flags.image_providers = "docker";

  Fetcher fetcher(flags);

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

  ASSERT_SOME(create);

  Owned<Containerizer> containerizer(create.get());

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

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

  ASSERT_SOME(slave);

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

  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());

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

  string dir1 = path::join(sandbox.get(), "dir1");
  ASSERT_SOME(os::mkdir(dir1));

  Resource persistentVolume = createPersistentVolume(
      Megabytes(64),
      "role1",
      "id1",
      "path1",
      None(),
      None(),
      frameworkInfo.principal());

  // Create a task that does nothing for a long time.
  TaskInfo task = createTask(
      offer.slave_id(),
      Resources::parse("cpus:1;mem:512").get() + persistentVolume,
      "sleep 1000");

  task.mutable_container()->CopyFrom(createContainerInfo(
      "test_image",
      {createVolumeHostPath("/tmp", dir1, Volume::RW)}));

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

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

  // Create the persistent volumes and launch task via `acceptOffers`.
  driver.acceptOffers(
      {offer.id()},
      {CREATE(persistentVolume), LAUNCH({task})});

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

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

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

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

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

  ContainerID containerId = *containers->begin();

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

  // Wipe the slave meta directory so that the slave will treat the
  // above running task as an orphan.
  ASSERT_SOME(os::rmdir(slave::paths::getMetaRootDir(flags.work_dir)));

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

  // Recreate the containerizer using the same helper as above.
  containerizer.reset();

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

  containerizer.reset(create.get());

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

  // Wait until slave recovery is complete.
  AWAIT_READY(_recover);

  // Wait until the orphan containers are cleaned up.
  AWAIT_READY(containerizer->wait(containerId));

  Try<fs::MountInfoTable> table = fs::MountInfoTable::read();
  ASSERT_SOME(table);

  // All mount targets should be under this directory.
  string directory = slave::paths::getSandboxRootDir(flags.work_dir);

  // Verify that the orphaned container's persistent volume and
  // the rootfs are unmounted.
  foreach (const fs::MountInfoTable::Entry& entry, table->entries) {
    EXPECT_FALSE(strings::contains(entry.target, directory))
      << "Target was not unmounted: " << entry.target;
  }

  driver.stop();
  driver.join();
}
// This test ensures that when explicit acknowledgements are enabled,
// acknowledgements for master-generated updates are dropped by the
// driver. We test this by creating an invalid task that uses no
// resources.
TEST_F(MesosSchedulerDriverTest, ExplicitAcknowledgementsMasterGeneratedUpdate)
{
  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 sched;
  MesosSchedulerDriver driver(
      &sched,
      DEFAULT_FRAMEWORK_INFO,
      master.get()->pid,
      false,
      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.

  // Ensure no status update acknowledgements are sent to the master.
  EXPECT_NO_FUTURE_CALLS(
      mesos::scheduler::Call(),
      mesos::scheduler::Call::ACKNOWLEDGE,
      _ ,
      master.get()->pid);

  driver.start();

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

  // Launch a task using no resources.
  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);

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

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

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

  AWAIT_READY(status);
  ASSERT_EQ(TASK_ERROR, status->state());
  ASSERT_EQ(TaskStatus::SOURCE_MASTER, status->source());
  ASSERT_EQ(TaskStatus::REASON_TASK_INVALID, status->reason());

  // Now send the acknowledgement.
  driver.acknowledgeStatusUpdate(status.get());

  // Settle the clock to ensure driver processes the acknowledgement,
  // which should get dropped due to having come from the master.
  Clock::pause();
  Clock::settle();

  driver.stop();
  driver.join();
}
// This test verifies that when reregistering, the slave sends the
// executor ID of a non-command executor task, but not the one of a
// command executor task. We then check that the master's API has
// task IDs absent only for the command executor case.
//
// This was motivated by MESOS-8135.
TEST_F(MasterSlaveReconciliationTest, SlaveReregisterTaskExecutorIds)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();

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

  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));

  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);
  EXPECT_NE(0u, offers->size());

  const Offer& offer = offers->front();
  const SlaveID& slaveId = offer.slave_id();

  Resources resources = Resources::parse(defaultTaskResourcesString).get();

  TaskInfo commandExecutorTask =
    createTask(slaveId, resources, SLEEP_COMMAND(1000));

  TaskInfo defaultExecutorTask =
    createTask(slaveId, resources, SLEEP_COMMAND(1000));

  ExecutorInfo defaultExecutorInfo;
  defaultExecutorInfo.set_type(ExecutorInfo::DEFAULT);
  defaultExecutorInfo.mutable_executor_id()->CopyFrom(DEFAULT_EXECUTOR_ID);
  defaultExecutorInfo.mutable_framework_id()->CopyFrom(frameworkId.get());
  defaultExecutorInfo.mutable_resources()->CopyFrom(resources);

  // We expect two TASK_STARTING and two TASK_RUNNING updates.
  vector<Future<TaskStatus>> taskStatuses(4);

  {
    // This variable doesn't have to be used explicitly.
    testing::InSequence inSequence;

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

    EXPECT_CALL(sched, statusUpdate(&driver, _))
      .WillRepeatedly(Return()); // Ignore subsequent updates.
  }
Esempio n. 20
0
TEST_P(MemoryIsolatorTest, ROOT_MemUsage)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.isolation = GetParam();

  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());

  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());

  TaskInfo task = createTask(offers.get()[0], "sleep 120");

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

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

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

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

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

  Future<ResourceStatistics> usage = containerizer->usage(containerId);
  AWAIT_READY(usage);

  // TODO(jieyu): Consider using a program that predictably increases
  // RSS so that we can set more meaningful expectation here.
  EXPECT_LT(0u, usage->mem_rss_bytes());

  driver.stop();
  driver.join();
}
Esempio n. 21
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. 22
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";

  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];

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

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

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

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

  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();

  // 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();
}
Esempio n. 23
0
// This test ensures that the command executor sends TASK_KILLING
// to frameworks that support the capability.
TEST_F(CommandExecutorTest, TaskKillingCapability)
{
  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);

  // Start the framework with the task killing capability.
  FrameworkInfo::Capability capability;
  capability.set_type(FrameworkInfo::Capability::TASK_KILLING_STATE);

  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.add_capabilities()->CopyFrom(capability);

  MockScheduler sched;
  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);
  EXPECT_EQ(1u, offers->size());

  // Launch a task with the command executor.
  TaskInfo task = createTask(
      offers->front().slave_id(),
      offers->front().resources(),
      "sleep 1000");

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

  driver.launchTasks(offers->front().id(), {task});

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

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

  driver.killTask(task.task_id());

  AWAIT_READY(statusKilling);
  EXPECT_EQ(TASK_KILLING, statusKilling->state());

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

  driver.stop();
  driver.join();
}
Esempio n. 24
0
TEST(FutureTest, ArrowOperator)
{
  Future<string> s = string("hello");
  EXPECT_EQ(5u, s->size());
}
Esempio n. 25
0
// This test ensures that the HTTP command executor can self terminate
// after it gets the ACK for the terminal status update from agent.
TEST_F_TEMP_DISABLED_ON_WINDOWS(HTTPCommandExecutorTest, TerminateWithACK)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

  slave::Flags flags = CreateSlaveFlags();
  flags.http_command_executor = true;

  Fetcher fetcher;

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

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

  StandaloneMasterDetector detector(master.get()->pid);

  MockSlave slave(flags, &detector, containerizer.get());
  spawn(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);
  EXPECT_EQ(1u, offers->size());

  // Launch a short lived task.
  TaskInfo task = createTask(
      offers->front().slave_id(),
      offers->front().resources(),
      "sleep 1");

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

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

  Future<Future<Option<ContainerTermination>>> termination;
  EXPECT_CALL(slave, executorTerminated(_, _, _))
    .WillOnce(FutureArg<2>(&termination));

  driver.launchTasks(offers->front().id(), {task});

  // Scheduler should first receive TASK_RUNNING followed by TASK_FINISHED.
  AWAIT_READY(statusRunning);
  EXPECT_EQ(TASK_RUNNING, statusRunning->state());

  AWAIT_READY(statusFinished);
  EXPECT_EQ(TASK_FINISHED, statusFinished->state());

  // The executor should self terminate with 0 as exit status once
  // it gets the ACK for the terminal status update from agent.
  AWAIT_READY(termination);
  ASSERT_TRUE(termination.get().isReady());
  EXPECT_EQ(0, termination.get().get().get().status());

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

  terminate(slave);
  wait(slave);
}
Esempio n. 26
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. 27
0
TEST_F(DiskQuotaTest, ResourceStatistics)
{
  Try<Owned<cluster::Master>> master = StartMaster();
  ASSERT_SOME(master);

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

  flags.resources = strings::format("disk(%s):10", DEFAULT_TEST_ROLE).get();

  // NOTE: We can't pause the clock because we need the reaper to reap
  // the 'du' subprocess.
  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);

  FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
  frameworkInfo.set_role(DEFAULT_TEST_ROLE);

  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());

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

  Resource volume = createPersistentVolume(
      Megabytes(4),
      DEFAULT_TEST_ROLE,
      "id1",
      "path1",
      None(),
      None(),
      DEFAULT_CREDENTIAL.principal());

  Resources taskResources = Resources::parse("cpus:1;mem:128").get();

  taskResources += createDiskResource(
      "3",
      DEFAULT_TEST_ROLE,
      None(),
      None());

  taskResources += volume;

  // Create a task that uses 2MB disk.
  TaskInfo task = createTask(
      offer.slave_id(),
      taskResources,
      "dd if=/dev/zero of=file bs=1048576 count=2 && "
      "dd if=/dev/zero of=path1/file bs=1048576 count=2 && "
      "sleep 1000");

  Future<TaskStatus> status1;
  Future<TaskStatus> status2;
  EXPECT_CALL(sched, statusUpdate(&driver, _))
    .WillOnce(FutureArg<1>(&status1))
    .WillOnce(FutureArg<1>(&status2))
    .WillRepeatedly(Return());       // Ignore subsequent updates.

  driver.acceptOffers(
      {offer.id()},
      {CREATE(volume),
       LAUNCH({task})});

  AWAIT_READY(status1);
  EXPECT_EQ(task.task_id(), status1->task_id());
  EXPECT_EQ(TASK_RUNNING, status1->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.
  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());
    }

    ASSERT_EQ(2u, usage->disk_statistics().size());

    bool done = true;
    foreach (const DiskStatistics& statistics, usage->disk_statistics()) {
      ASSERT_TRUE(statistics.has_limit_bytes());
      EXPECT_EQ(
          statistics.has_persistence() ? Megabytes(4) : Megabytes(3),
          statistics.limit_bytes());

      if (!statistics.has_used_bytes()) {
        done = false;
      } else {
        EXPECT_GT(
            statistics.has_persistence() ? Megabytes(4) : Megabytes(3),
            statistics.used_bytes());
      }
    }

    if (done) {
      break;
    }

    ASSERT_LT(elapsed, Seconds(5));

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

  driver.killTask(task.task_id());

  AWAIT_READY(status2);
  EXPECT_EQ(task.task_id(), status2->task_id());
  EXPECT_EQ(TASK_KILLED, status2->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();
}