// This test verifies that a framework attempting to subscribe
// after its failover timeout has elapsed is disallowed.
TEST_F(HttpFaultToleranceTest, SchedulerSubscribeAfterFailoverTimeout)
{
master::Flags flags = CreateMasterFlags();
flags.authenticate_frameworks = false;
v1::FrameworkInfo frameworkInfo = v1::DEFAULT_FRAMEWORK_INFO;
frameworkInfo.set_failover_timeout(Weeks(2).secs());
Try<Owned<cluster::Master>> master = StartMaster(flags);
ASSERT_SOME(master);
Future<Nothing> deactivateFramework = FUTURE_DISPATCH(
_, &master::allocator::MesosAllocatorProcess::deactivateFramework);
v1::FrameworkID frameworkId;
ContentType contentType = ContentType::PROTOBUF;
// Launch the first (i.e., failing) scheduler and wait until it receives
// a `SUBSCRIBED` event to launch the second (i.e., failover) scheduler.
{
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
Future<Nothing> connected;
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(FutureSatisfy(&connected));
v1::scheduler::TestMesos schedulerLibrary(
master.get()->pid,
contentType,
scheduler);
AWAIT_READY(connected);
Future<Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
{
Call call;
call.set_type(Call::SUBSCRIBE);
Call::Subscribe* subscribe = call.mutable_subscribe();
subscribe->mutable_framework_info()->CopyFrom(frameworkInfo);
schedulerLibrary.send(call);
}
AWAIT_READY(subscribed);
frameworkId = subscribed->framework_id();
}
// Wait until master schedules the framework for removal.
AWAIT_READY(deactivateFramework);
// Simulate framework failover timeout.
Clock::pause();
Clock::settle();
Try<Duration> failoverTimeout =
Duration::create(frameworkInfo.failover_timeout());
ASSERT_SOME(failoverTimeout);
Future<Nothing> frameworkFailoverTimeout =
FUTURE_DISPATCH(_, &Master::frameworkFailoverTimeout);
Clock::advance(failoverTimeout.get());
Clock::resume();
// Wait until master actually marks the framework as completed.
AWAIT_READY(frameworkFailoverTimeout);
// Now launch the second (i.e., failover) scheduler using the
// framework id recorded from the first scheduler.
{
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
Future<Nothing> connected;
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(FutureSatisfy(&connected))
.WillRepeatedly(Return()); // Ignore future invocations.
v1::scheduler::TestMesos schedulerLibrary(
master.get()->pid,
contentType,
scheduler);
AWAIT_READY(connected);
// Framework should get `Error` event because the framework with this id
// is marked as completed.
Future<Nothing> error;
EXPECT_CALL(*scheduler, error(_, _))
.WillOnce(FutureSatisfy(&error));
EXPECT_CALL(*scheduler, disconnected(_))
.Times(AtMost(1));
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::SUBSCRIBE);
Call::Subscribe* subscribe = call.mutable_subscribe();
subscribe->mutable_framework_info()->CopyFrom(v1::DEFAULT_FRAMEWORK_INFO);
subscribe->mutable_framework_info()->mutable_id()->CopyFrom(frameworkId);
schedulerLibrary.send(call);
}
AWAIT_READY(error);
}
}
// This test ensures that the failed over scheduler is able to send a message
// to the executor.
TEST_F(HttpFaultToleranceTest, SchedulerFailoverFrameworkToExecutorMessage)
{
master::Flags flags = CreateMasterFlags();
flags.authenticate_frameworks = false;
Try<Owned<cluster::Master>> master = StartMaster(flags);
ASSERT_SOME(master);
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
auto executor = std::make_shared<v1::MockHTTPExecutor>();
ExecutorID executorId = DEFAULT_EXECUTOR_ID;
TestContainerizer containerizer(executorId, executor);
Owned<MasterDetector> detector = master.get()->createDetector();
Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), &containerizer);
ASSERT_SOME(slave);
Future<Nothing> connected;
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(FutureSatisfy(&connected))
.WillRepeatedly(Return()); // Ignore future invocations.
ContentType contentType = ContentType::PROTOBUF;
v1::scheduler::TestMesos schedulerLibrary(
master.get()->pid,
contentType,
scheduler);
AWAIT_READY(connected);
Future<Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
Future<Event::Offers> offers;
EXPECT_CALL(*scheduler, offers(_, _))
.WillOnce(FutureArg<1>(&offers));
{
Call call;
call.set_type(Call::SUBSCRIBE);
Call::Subscribe* subscribe = call.mutable_subscribe();
subscribe->mutable_framework_info()->CopyFrom(v1::DEFAULT_FRAMEWORK_INFO);
schedulerLibrary.send(call);
}
AWAIT_READY(subscribed);
v1::FrameworkID frameworkId(subscribed->framework_id());
AWAIT_READY(offers);
EXPECT_NE(0, offers->offers().size());
EXPECT_CALL(*executor, connected(_))
.WillOnce(v1::executor::SendSubscribe(frameworkId, evolve(executorId)));
EXPECT_CALL(*executor, subscribed(_, _));
Future<Nothing> launch;
EXPECT_CALL(*executor, launch(_, _))
.WillOnce(FutureSatisfy(&launch));
const v1::Offer& offer = offers->offers(0);
v1::TaskInfo taskInfo =
evolve(createTask(devolve(offer), "", executorId));
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::ACCEPT);
Call::Accept* accept = call.mutable_accept();
accept->add_offer_ids()->CopyFrom(offer.id());
v1::Offer::Operation* operation = accept->add_operations();
operation->set_type(v1::Offer::Operation::LAUNCH);
operation->mutable_launch()->add_task_infos()->CopyFrom(taskInfo);
schedulerLibrary.send(call);
}
AWAIT_READY(launch);
auto scheduler2 = std::make_shared<v1::MockHTTPScheduler>();
Future<Nothing> connected2;
EXPECT_CALL(*scheduler2, connected(_))
.WillOnce(FutureSatisfy(&connected2));
// Failover to another scheduler instance.
v1::scheduler::TestMesos schedulerLibrary2(
master.get()->pid,
contentType,
scheduler2);
AWAIT_READY(connected2);
// The previously connected scheduler instance should receive an
// error/disconnected event.
Future<Nothing> error;
EXPECT_CALL(*scheduler, error(_, _))
.WillOnce(FutureSatisfy(&error));
Future<Nothing> disconnected;
EXPECT_CALL(*scheduler, disconnected(_))
.WillOnce(FutureSatisfy(&disconnected));
EXPECT_CALL(*scheduler2, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
EXPECT_CALL(*scheduler2, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::SUBSCRIBE);
Call::Subscribe* subscribe = call.mutable_subscribe();
subscribe->mutable_framework_info()->CopyFrom(v1::DEFAULT_FRAMEWORK_INFO);
subscribe->mutable_framework_info()->mutable_id()->CopyFrom(frameworkId);
schedulerLibrary2.send(call);
}
AWAIT_READY(error);
AWAIT_READY(disconnected);
AWAIT_READY(subscribed);
EXPECT_EQ(frameworkId, subscribed->framework_id());
Future<v1::executor::Event::Message> message;
EXPECT_CALL(*executor, message(_, _))
.WillOnce(FutureArg<1>(&message));
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::MESSAGE);
Call::Message* message = call.mutable_message();
message->mutable_agent_id()->CopyFrom(offer.agent_id());
message->mutable_executor_id()->CopyFrom(v1::DEFAULT_EXECUTOR_ID);
message->set_data("hello world");
schedulerLibrary2.send(call);
}
AWAIT_READY(message);
ASSERT_EQ("hello world", message->data());
EXPECT_CALL(*executor, shutdown(_))
.Times(AtMost(1));
EXPECT_CALL(*executor, disconnected(_))
.Times(AtMost(1));
}
// This test checks that a scheduler exit shuts down the executor.
TEST_F(HttpFaultToleranceTest, SchedulerExit)
{
master::Flags flags = CreateMasterFlags();
flags.authenticate_frameworks = false;
Try<Owned<cluster::Master>> master = StartMaster(flags);
ASSERT_SOME(master);
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
auto executor = std::make_shared<v1::MockHTTPExecutor>();
ExecutorID executorId = DEFAULT_EXECUTOR_ID;
TestContainerizer containerizer(executorId, executor);
Owned<MasterDetector> detector = master.get()->createDetector();
Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), &containerizer);
ASSERT_SOME(slave);
Future<Nothing> connected;
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(FutureSatisfy(&connected))
.WillRepeatedly(Return()); // Ignore future invocations.
ContentType contentType = ContentType::PROTOBUF;
v1::scheduler::TestMesos schedulerLibrary(
master.get()->pid,
contentType,
scheduler);
AWAIT_READY(connected);
Future<Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
Future<Event::Offers> offers;
EXPECT_CALL(*scheduler, offers(_, _))
.WillOnce(FutureArg<1>(&offers));
{
Call call;
call.set_type(Call::SUBSCRIBE);
Call::Subscribe* subscribe = call.mutable_subscribe();
subscribe->mutable_framework_info()->CopyFrom(v1::DEFAULT_FRAMEWORK_INFO);
schedulerLibrary.send(call);
}
AWAIT_READY(subscribed);
v1::FrameworkID frameworkId(subscribed->framework_id());
AWAIT_READY(offers);
EXPECT_NE(0, offers->offers().size());
EXPECT_CALL(*executor, connected(_))
.WillOnce(v1::executor::SendSubscribe(frameworkId, evolve(executorId)));
EXPECT_CALL(*executor, subscribed(_, _));
Future<Nothing> launch;
EXPECT_CALL(*executor, launch(_, _))
.WillOnce(FutureSatisfy(&launch));
const v1::Offer& offer = offers->offers(0);
v1::TaskInfo taskInfo =
evolve(createTask(devolve(offer), "", DEFAULT_EXECUTOR_ID));
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::ACCEPT);
Call::Accept* accept = call.mutable_accept();
accept->add_offer_ids()->CopyFrom(offer.id());
v1::Offer::Operation* operation = accept->add_operations();
operation->set_type(v1::Offer::Operation::LAUNCH);
operation->mutable_launch()->add_task_infos()->CopyFrom(taskInfo);
schedulerLibrary.send(call);
}
AWAIT_READY(launch);
EXPECT_CALL(*scheduler, disconnected(_))
.Times(AtMost(1));
Future<Nothing> shutdown;
EXPECT_CALL(*executor, shutdown(_))
.WillOnce(FutureSatisfy(&shutdown));
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::TEARDOWN);
schedulerLibrary.send(call);
}
// Ensure that the executor receives a `Event::Shutdown` after the
// scheduler exit.
AWAIT_READY(shutdown);
}
// This test verifies that, after a master failover, reconciliation of an
// operation that is still pending on an agent results in `OPERATION_PENDING`.
TEST_P(OperationReconciliationTest, AgentPendingOperationAfterMasterFailover)
{
Clock::pause();
mesos::internal::master::Flags masterFlags = CreateMasterFlags();
Try<Owned<cluster::Master>> master = StartMaster(masterFlags);
ASSERT_SOME(master);
Future<UpdateSlaveMessage> updateSlaveMessage =
FUTURE_PROTOBUF(UpdateSlaveMessage(), _, _);
auto detector = std::make_shared<StandaloneMasterDetector>(master.get()->pid);
mesos::internal::slave::Flags slaveFlags = CreateSlaveFlags();
Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), slaveFlags);
ASSERT_SOME(slave);
// Advance the clock to trigger agent registration.
Clock::advance(slaveFlags.registration_backoff_factor);
// Wait for the agent to register.
AWAIT_READY(updateSlaveMessage);
// Start and register a resource provider.
ResourceProviderInfo resourceProviderInfo;
resourceProviderInfo.set_type("org.apache.mesos.rp.test");
resourceProviderInfo.set_name("test");
Resource disk = createDiskResource(
"200", "*", None(), None(), createDiskSourceRaw(None(), "profile"));
Owned<MockResourceProvider> resourceProvider(
new MockResourceProvider(
resourceProviderInfo,
Resources(disk)));
// We override the mock resource provider's default action, so the operation
// will stay in `OPERATION_PENDING`.
Future<resource_provider::Event::ApplyOperation> applyOperation;
EXPECT_CALL(*resourceProvider, applyOperation(_))
.WillOnce(FutureArg<0>(&applyOperation));
Owned<EndpointDetector> endpointDetector(
mesos::internal::tests::resource_provider::createEndpointDetector(
slave.get()->pid));
updateSlaveMessage = FUTURE_PROTOBUF(UpdateSlaveMessage(), _, _);
// NOTE: We need to resume the clock so that the resource provider can
// fully register.
Clock::resume();
ContentType contentType = GetParam();
resourceProvider->start(endpointDetector, contentType);
// Wait until the agent's resources have been updated to include the
// resource provider resources.
AWAIT_READY(updateSlaveMessage);
ASSERT_TRUE(updateSlaveMessage->has_resource_providers());
ASSERT_EQ(1, updateSlaveMessage->resource_providers().providers_size());
Clock::pause();
// Start a v1 framework.
auto scheduler = std::make_shared<MockHTTPScheduler>();
FrameworkInfo frameworkInfo = DEFAULT_FRAMEWORK_INFO;
frameworkInfo.set_roles(0, DEFAULT_TEST_ROLE);
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(scheduler::SendSubscribe(frameworkInfo));
Future<scheduler::Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
// Ignore heartbeats.
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return());
// Decline offers that do not contain wanted resources.
EXPECT_CALL(*scheduler, offers(_, _))
.WillRepeatedly(scheduler::DeclineOffers());
Future<scheduler::Event::Offers> offers;
auto isRaw = [](const Resource& r) {
return r.has_disk() &&
r.disk().has_source() &&
r.disk().source().type() == Resource::DiskInfo::Source::RAW;
};
EXPECT_CALL(*scheduler, offers(_, scheduler::OffersHaveAnyResource(
std::bind(isRaw, lambda::_1))))
.WillOnce(FutureArg<1>(&offers))
.WillRepeatedly(scheduler::DeclineOffers()); // Decline successive offers.
scheduler::TestMesos mesos(
master.get()->pid, contentType, scheduler, detector);
AWAIT_READY(subscribed);
FrameworkID frameworkId(subscribed->framework_id());
// NOTE: If the framework has not declined an unwanted offer yet when
// the master updates the agent with the RAW disk resource, the new
// allocation triggered by this update won't generate an allocatable
// offer due to no CPU and memory resources. So here we first settle
// the clock to ensure that the unwanted offer has been declined, then
// advance the clock to trigger another allocation.
Clock::settle();
Clock::advance(masterFlags.allocation_interval);
AWAIT_READY(offers);
ASSERT_FALSE(offers->offers().empty());
const Offer& offer = offers->offers(0);
const AgentID& agentId = offer.agent_id();
Option<Resource> source;
Option<ResourceProviderID> resourceProviderId;
foreach (const Resource& resource, offer.resources()) {
if (isRaw(resource)) {
source = resource;
ASSERT_TRUE(resource.has_provider_id());
resourceProviderId = resource.provider_id();
break;
}
}
ASSERT_SOME(source);
ASSERT_SOME(resourceProviderId);
OperationID operationId;
operationId.set_value("operation");
mesos.send(createCallAccept(
frameworkId,
offer,
{CREATE_DISK(
source.get(),
Resource::DiskInfo::Source::MOUNT,
None(),
operationId)}));
AWAIT_READY(applyOperation);
// Simulate master failover.
EXPECT_CALL(*scheduler, disconnected(_));
detector->appoint(None());
master->reset();
master = StartMaster();
ASSERT_SOME(master);
// Settle the clock to ensure the master finishes recovering the registry.
Clock::settle();
Future<SlaveReregisteredMessage> slaveReregistered = FUTURE_PROTOBUF(
SlaveReregisteredMessage(), master.get()->pid, slave.get()->pid);
updateSlaveMessage = FUTURE_PROTOBUF(UpdateSlaveMessage(), _, _);
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(scheduler::SendSubscribe(frameworkInfo, frameworkId));
Future<scheduler::Event::Subscribed> frameworkResubscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&frameworkResubscribed));
// Simulate a new master detected event to the agent and the scheduler.
detector->appoint(master.get()->pid);
// Advance the clock, so that the agent re-registers.
Clock::advance(slaveFlags.registration_backoff_factor);
// Resume the clock to avoid deadlocks related to agent registration.
// See MESOS-8828.
Clock::resume();
// Wait for the framework and agent to re-register.
AWAIT_READY(slaveReregistered);
AWAIT_READY(updateSlaveMessage);
AWAIT_READY(frameworkResubscribed);
Clock::pause();
// Test explicit reconciliation
{
scheduler::Call::ReconcileOperations::Operation operation;
operation.mutable_operation_id()->CopyFrom(operationId);
operation.mutable_agent_id()->CopyFrom(agentId);
const Future<scheduler::APIResult> result =
mesos.call({createCallReconcileOperations(frameworkId, {operation})});
AWAIT_READY(result);
// The master should respond with '200 OK' and with a `scheduler::Response`.
ASSERT_EQ(process::http::Status::OK, result->status_code());
ASSERT_TRUE(result->has_response());
const scheduler::Response response = result->response();
ASSERT_EQ(scheduler::Response::RECONCILE_OPERATIONS, response.type());
ASSERT_TRUE(response.has_reconcile_operations());
const scheduler::Response::ReconcileOperations& reconcile =
response.reconcile_operations();
ASSERT_EQ(1, reconcile.operation_statuses_size());
const OperationStatus& operationStatus = reconcile.operation_statuses(0);
EXPECT_EQ(operationId, operationStatus.operation_id());
EXPECT_EQ(OPERATION_PENDING, operationStatus.state());
EXPECT_FALSE(operationStatus.has_uuid());
}
// Test implicit reconciliation
{
const Future<scheduler::APIResult> result =
mesos.call({createCallReconcileOperations(frameworkId, {})});
AWAIT_READY(result);
// The master should respond with '200 OK' and with a `scheduler::Response`.
ASSERT_EQ(process::http::Status::OK, result->status_code());
ASSERT_TRUE(result->has_response());
const scheduler::Response response = result->response();
ASSERT_EQ(scheduler::Response::RECONCILE_OPERATIONS, response.type());
ASSERT_TRUE(response.has_reconcile_operations());
const scheduler::Response::ReconcileOperations& reconcile =
response.reconcile_operations();
ASSERT_EQ(1, reconcile.operation_statuses_size());
const OperationStatus& operationStatus = reconcile.operation_statuses(0);
EXPECT_EQ(operationId, operationStatus.operation_id());
EXPECT_EQ(OPERATION_PENDING, operationStatus.state());
EXPECT_FALSE(operationStatus.has_uuid());
}
}
// This test checks that a failed over scheduler gets the retried status update
// when the original instance dies without acknowledging the update.
TEST_F(HttpFaultToleranceTest, SchedulerFailoverStatusUpdate)
{
master::Flags flags = CreateMasterFlags();
flags.authenticate_frameworks = false;
Try<Owned<cluster::Master>> master = StartMaster(flags);
ASSERT_SOME(master);
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
auto executor = std::make_shared<v1::MockHTTPExecutor>();
ExecutorID executorId = DEFAULT_EXECUTOR_ID;
TestContainerizer containerizer(executorId, executor);
Owned<MasterDetector> detector = master.get()->createDetector();
Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), &containerizer);
ASSERT_SOME(slave);
Future<Nothing> connected;
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(FutureSatisfy(&connected))
.WillRepeatedly(Return()); // Ignore future invocations.
ContentType contentType = ContentType::PROTOBUF;
v1::scheduler::TestMesos schedulerLibrary(
master.get()->pid,
contentType,
scheduler);
AWAIT_READY(connected);
Future<Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
Future<Event::Offers> offers;
EXPECT_CALL(*scheduler, offers(_, _))
.WillOnce(FutureArg<1>(&offers));
{
Call call;
call.set_type(Call::SUBSCRIBE);
Call::Subscribe* subscribe = call.mutable_subscribe();
subscribe->mutable_framework_info()->CopyFrom(v1::DEFAULT_FRAMEWORK_INFO);
schedulerLibrary.send(call);
}
AWAIT_READY(subscribed);
v1::FrameworkID frameworkId(subscribed->framework_id());
AWAIT_READY(offers);
EXPECT_NE(0, offers->offers().size());
EXPECT_CALL(*executor, connected(_))
.WillOnce(v1::executor::SendSubscribe(frameworkId, evolve(executorId)));
EXPECT_CALL(*executor, subscribed(_, _));
EXPECT_CALL(*executor, launch(_, _))
.WillOnce(v1::executor::SendUpdateFromTask(
frameworkId, evolve(executorId), v1::TASK_RUNNING));
Future<Nothing> acknowledged;
EXPECT_CALL(*executor, acknowledged(_, _))
.WillOnce(FutureSatisfy(&acknowledged));
Future<Event::Update> update;
EXPECT_CALL(*scheduler, update(_, _))
.WillOnce(FutureArg<1>(&update));
const v1::Offer& offer = offers->offers(0);
v1::TaskInfo taskInfo =
evolve(createTask(devolve(offer), "", executorId));
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::ACCEPT);
Call::Accept* accept = call.mutable_accept();
accept->add_offer_ids()->CopyFrom(offer.id());
v1::Offer::Operation* operation = accept->add_operations();
operation->set_type(v1::Offer::Operation::LAUNCH);
operation->mutable_launch()->add_task_infos()->CopyFrom(taskInfo);
schedulerLibrary.send(call);
}
AWAIT_READY(acknowledged);
AWAIT_READY(update);
EXPECT_EQ(v1::TASK_RUNNING, update->status().state());
EXPECT_EQ(executorId, devolve(update->status().executor_id()));
EXPECT_TRUE(update->status().has_executor_id());
EXPECT_TRUE(update->status().has_uuid());
// Failover the scheduler without acknowledging the status update.
auto scheduler2 = std::make_shared<v1::MockHTTPScheduler>();
Future<Nothing> connected2;
EXPECT_CALL(*scheduler2, connected(_))
.WillOnce(FutureSatisfy(&connected2));
// Failover to another scheduler instance.
v1::scheduler::TestMesos schedulerLibrary2(
master.get()->pid,
contentType,
scheduler2);
AWAIT_READY(connected2);
// The previously connected scheduler instance should receive an
// error/disconnected event.
Future<Nothing> error;
EXPECT_CALL(*scheduler, error(_, _))
.WillOnce(FutureSatisfy(&error));
Future<Nothing> disconnected;
EXPECT_CALL(*scheduler, disconnected(_))
.WillOnce(FutureSatisfy(&disconnected));
EXPECT_CALL(*scheduler2, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
EXPECT_CALL(*scheduler2, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
// Scheduler2 should receive the retried status update.
Future<Nothing> update2;
EXPECT_CALL(*scheduler2, update(_, _))
.WillOnce(FutureSatisfy(&update2))
.WillRepeatedly(Return()); // Ignore subsequent updates.
{
Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(Call::SUBSCRIBE);
Call::Subscribe* subscribe = call.mutable_subscribe();
subscribe->mutable_framework_info()->CopyFrom(v1::DEFAULT_FRAMEWORK_INFO);
subscribe->mutable_framework_info()->mutable_id()->CopyFrom(frameworkId);
schedulerLibrary2.send(call);
}
AWAIT_READY(error);
AWAIT_READY(disconnected);
AWAIT_READY(subscribed);
EXPECT_EQ(frameworkId, subscribed->framework_id());
Clock::pause();
// Now advance time enough for the reliable timeout to kick in and
// another status update to be sent.
Clock::advance(slave::STATUS_UPDATE_RETRY_INTERVAL_MIN);
AWAIT_READY(update2);
EXPECT_CALL(*executor, shutdown(_))
.Times(AtMost(1));
EXPECT_CALL(*executor, disconnected(_))
.Times(AtMost(1));
}
// This test verifies that reconciliation of an unknown operation that belongs
// to an unknown agent results in `OPERATION_UNKNOWN`.
TEST_P(OperationReconciliationTest, UnknownOperationUnknownAgent)
{
Clock::pause();
Try<Owned<cluster::Master>> master = StartMaster();
ASSERT_SOME(master);
auto scheduler = std::make_shared<MockHTTPScheduler>();
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(scheduler::SendSubscribe(DEFAULT_FRAMEWORK_INFO));
Future<scheduler::Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
// Ignore heartbeats.
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return());
// Decline all offers.
EXPECT_CALL(*scheduler, offers(_, _))
.WillRepeatedly(scheduler::DeclineOffers());
scheduler::TestMesos mesos(master.get()->pid, GetParam(), scheduler);
AWAIT_READY(subscribed);
FrameworkID frameworkId(subscribed->framework_id());
AgentID agentId;
agentId.set_value("agent");
OperationID operationId;
operationId.set_value("operation");
scheduler::Call::ReconcileOperations::Operation operation;
operation.mutable_operation_id()->CopyFrom(operationId);
operation.mutable_agent_id()->CopyFrom(agentId);
const Future<scheduler::APIResult> result =
mesos.call({createCallReconcileOperations(frameworkId, {operation})});
AWAIT_READY(result);
// The master should respond with '200 OK' and with a `scheduler::Response`.
ASSERT_EQ(process::http::Status::OK, result->status_code());
ASSERT_TRUE(result->has_response());
const scheduler::Response response = result->response();
ASSERT_EQ(scheduler::Response::RECONCILE_OPERATIONS, response.type());
ASSERT_TRUE(response.has_reconcile_operations());
const scheduler::Response::ReconcileOperations& reconcile =
response.reconcile_operations();
ASSERT_EQ(1, reconcile.operation_statuses_size());
const OperationStatus& operationStatus = reconcile.operation_statuses(0);
EXPECT_EQ(operationId, operationStatus.operation_id());
EXPECT_EQ(OPERATION_UNKNOWN, operationStatus.state());
EXPECT_FALSE(operationStatus.has_uuid());
}
// This test verifies that reconciliation of an unknown operation that belongs
// to an agent marked gone results in `OPERATION_GONE_BY_OPERATOR`.
TEST_P(OperationReconciliationTest, UnknownOperationAgentMarkedGone)
{
Clock::pause();
mesos::internal::master::Flags masterFlags = CreateMasterFlags();
Try<Owned<cluster::Master>> master = StartMaster(masterFlags);
ASSERT_SOME(master);
Future<SlaveRegisteredMessage> slaveRegisteredMessage =
FUTURE_PROTOBUF(SlaveRegisteredMessage(), _, _);
Owned<MasterDetector> detector = master.get()->createDetector();
mesos::internal::slave::Flags slaveFlags = CreateSlaveFlags();
Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), slaveFlags);
ASSERT_SOME(slave);
// Advance the clock to trigger agent registration.
Clock::advance(slaveFlags.registration_backoff_factor);
// Wait for the agent to register and get the agent ID.
AWAIT_READY(slaveRegisteredMessage);
const AgentID agentId = evolve(slaveRegisteredMessage->slave_id());
ContentType contentType = GetParam();
{
master::Call call;
call.set_type(master::Call::MARK_AGENT_GONE);
call.mutable_mark_agent_gone()->mutable_agent_id()->CopyFrom(agentId);
Future<process::http::Response> response = process::http::post(
master.get()->pid,
"api/v1",
createBasicAuthHeaders(DEFAULT_CREDENTIAL),
serialize(contentType, call),
stringify(contentType));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(process::http::OK().status, response);
}
auto scheduler = std::make_shared<MockHTTPScheduler>();
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(scheduler::SendSubscribe(DEFAULT_FRAMEWORK_INFO));
Future<scheduler::Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
// Ignore heartbeats.
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return());
// Decline all offers.
EXPECT_CALL(*scheduler, offers(_, _))
.WillRepeatedly(scheduler::DeclineOffers());
scheduler::TestMesos mesos(master.get()->pid, contentType, scheduler);
AWAIT_READY(subscribed);
FrameworkID frameworkId(subscribed->framework_id());
OperationID operationId;
operationId.set_value("operation");
scheduler::Call::ReconcileOperations::Operation operation;
operation.mutable_operation_id()->CopyFrom(operationId);
operation.mutable_agent_id()->CopyFrom(agentId);
const Future<scheduler::APIResult> result =
mesos.call({createCallReconcileOperations(frameworkId, {operation})});
AWAIT_READY(result);
// The master should respond with '200 OK' and with a `scheduler::Response`.
ASSERT_EQ(process::http::Status::OK, result->status_code());
ASSERT_TRUE(result->has_response());
const scheduler::Response response = result->response();
ASSERT_EQ(scheduler::Response::RECONCILE_OPERATIONS, response.type());
ASSERT_TRUE(response.has_reconcile_operations());
const scheduler::Response::ReconcileOperations& reconcile =
response.reconcile_operations();
ASSERT_EQ(1, reconcile.operation_statuses_size());
const OperationStatus& operationStatus = reconcile.operation_statuses(0);
EXPECT_EQ(operationId, operationStatus.operation_id());
EXPECT_EQ(OPERATION_GONE_BY_OPERATOR, operationStatus.state());
EXPECT_FALSE(operationStatus.has_uuid());
}
// This test verifies that executor API and operator API calls receive an
// unsuccessful response if the request contains a properly-signed
// authentication token with invalid claims.
TEST_F(ExecutorAuthorizationTest, FailedApiCalls)
{
Try<Owned<cluster::Master>> master = StartMaster();
ASSERT_SOME(master);
// Start an agent with permissive ACLs so that a task can be launched and the
// local authorizer's implicit executor authorization will be performed.
ACLs acls;
acls.set_permissive(true);
slave::Flags flags = CreateSlaveFlags();
flags.acls = acls;
Owned<MasterDetector> detector = master.get()->createDetector();
v1::Resources resources =
v1::Resources::parse("cpus:0.1;mem:32;disk:32").get();
v1::ExecutorInfo executorInfo;
executorInfo.set_type(v1::ExecutorInfo::DEFAULT);
executorInfo.mutable_executor_id()->CopyFrom(v1::DEFAULT_EXECUTOR_ID);
executorInfo.mutable_resources()->CopyFrom(resources);
auto executor = std::make_shared<v1::MockHTTPExecutor>();
Owned<TestContainerizer> containerizer(new TestContainerizer(
devolve(executorInfo.executor_id()), executor));
Try<Owned<cluster::Slave>> slave =
this->StartSlave(detector.get(), containerizer.get(), flags);
ASSERT_SOME(slave);
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
Future<Nothing> connected;
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(FutureSatisfy(&connected));
v1::scheduler::TestMesos mesos(
master.get()->pid,
ContentType::PROTOBUF,
scheduler);
AWAIT_READY(connected);
Future<v1::scheduler::Event::Subscribed> frameworkSubscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&frameworkSubscribed));
Future<v1::scheduler::Event::Offers> offers;
EXPECT_CALL(*scheduler, offers(_, _))
.WillOnce(FutureArg<1>(&offers))
.WillRepeatedly(Return()); // Ignore subsequent offers.
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
mesos.send(v1::createCallSubscribe(v1::DEFAULT_FRAMEWORK_INFO));
AWAIT_READY(frameworkSubscribed);
v1::FrameworkID frameworkId(frameworkSubscribed->framework_id());
executorInfo.mutable_framework_id()->CopyFrom(frameworkId);
AWAIT_READY(offers);
ASSERT_FALSE(offers->offers().empty());
Future<v1::executor::Mesos*> executorLib;
EXPECT_CALL(*executor, connected(_))
.WillOnce(FutureArg<0>(&executorLib));
const v1::Offer& offer = offers->offers(0);
const v1::AgentID& agentId = offer.agent_id();
{
v1::scheduler::Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(v1::scheduler::Call::ACCEPT);
v1::scheduler::Call::Accept* accept = call.mutable_accept();
accept->add_offer_ids()->CopyFrom(offer.id());
v1::Offer::Operation* operation = accept->add_operations();
operation->set_type(v1::Offer::Operation::LAUNCH_GROUP);
v1::TaskInfo taskInfo =
v1::createTask(agentId, resources, SLEEP_COMMAND(1000));
v1::TaskGroupInfo taskGroup;
taskGroup.add_tasks()->CopyFrom(taskInfo);
v1::Offer::Operation::LaunchGroup* launchGroup =
operation->mutable_launch_group();
launchGroup->mutable_executor()->CopyFrom(executorInfo);
launchGroup->mutable_task_group()->CopyFrom(taskGroup);
mesos.send(call);
}
AWAIT_READY(executorLib);
Future<v1::executor::Event::Subscribed> executorSubscribed;
EXPECT_CALL(*executor, subscribed(_, _))
.WillOnce(FutureArg<1>(&executorSubscribed));
Future<Nothing> launchGroup;
EXPECT_CALL(*executor, launchGroup(_, _))
.WillOnce(FutureSatisfy(&launchGroup));
{
v1::executor::Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.mutable_executor_id()->CopyFrom(v1::DEFAULT_EXECUTOR_ID);
call.set_type(v1::executor::Call::SUBSCRIBE);
call.mutable_subscribe();
executorLib.get()->send(call);
}
// Wait for the executor to subscribe. Once it is in the SUBSCRIBED state,
// the UPDATE and MESSAGE executor calls can be attempted.
AWAIT_READY(executorSubscribed);
AWAIT_READY(launchGroup);
// Create a principal which contains an incorrect ContainerID.
hashmap<string, string> claims;
claims["fid"] = frameworkId.value();
claims["eid"] = v1::DEFAULT_EXECUTOR_ID.value();
claims["cid"] = id::UUID::random().toString();
Principal incorrectPrincipal(None(), claims);
// Generate an authentication token which is signed using the correct key,
// but contains an invalid set of claims.
Owned<JWTSecretGenerator> jwtSecretGenerator(
new JWTSecretGenerator(DEFAULT_JWT_SECRET_KEY));
Future<Secret> authenticationToken =
jwtSecretGenerator->generate(incorrectPrincipal);
AWAIT_READY(authenticationToken);
v1::ContainerID containerId;
containerId.set_value(id::UUID::random().toString());
containerId.mutable_parent()->CopyFrom(executorSubscribed->container_id());
http::Headers headers;
headers["Authorization"] = "Bearer " + authenticationToken->value().data();
// Since the executor library has already been initialized with a valid
// authentication token, we use an HTTP helper function to send the
// executor API and operator API calls with an invalid token.
{
v1::agent::Call call;
call.set_type(v1::agent::Call::LAUNCH_NESTED_CONTAINER);
call.mutable_launch_nested_container()->mutable_container_id()
->CopyFrom(containerId);
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
}
{
v1::agent::Call call;
call.set_type(v1::agent::Call::LAUNCH_NESTED_CONTAINER_SESSION);
call.mutable_launch_nested_container_session()->mutable_container_id()
->CopyFrom(containerId);
call.mutable_launch_nested_container_session()->mutable_command()
->set_value("sleep 120");
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
}
{
v1::agent::Call call;
call.set_type(v1::agent::Call::WAIT_NESTED_CONTAINER);
call.mutable_wait_nested_container()->mutable_container_id()
->CopyFrom(containerId);
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
}
{
v1::agent::Call call;
call.set_type(v1::agent::Call::KILL_NESTED_CONTAINER);
call.mutable_kill_nested_container()->mutable_container_id()
->CopyFrom(containerId);
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
}
{
v1::agent::Call call;
call.set_type(v1::agent::Call::REMOVE_NESTED_CONTAINER);
call.mutable_remove_nested_container()->mutable_container_id()
->CopyFrom(containerId);
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
}
{
v1::agent::Call call;
call.set_type(v1::agent::Call::ATTACH_CONTAINER_OUTPUT);
call.mutable_attach_container_output()->mutable_container_id()
->CopyFrom(containerId);
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
}
const string failureMessage =
"does not contain a 'cid' claim with the correct active ContainerID";
{
v1::TaskStatus status;
status.mutable_task_id()->set_value(id::UUID::random().toString());
status.set_state(v1::TASK_RUNNING);
status.set_uuid(id::UUID::random().toBytes());
status.set_source(v1::TaskStatus::SOURCE_EXECUTOR);
v1::executor::Call call;
call.set_type(v1::executor::Call::UPDATE);
call.mutable_framework_id()->CopyFrom(frameworkId);
call.mutable_executor_id()->CopyFrom(v1::DEFAULT_EXECUTOR_ID);
call.mutable_update()->mutable_status()->CopyFrom(status);
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1/executor",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
EXPECT_TRUE(strings::contains(response->body, failureMessage));
}
{
v1::executor::Call call;
call.set_type(v1::executor::Call::MESSAGE);
call.mutable_framework_id()->CopyFrom(frameworkId);
call.mutable_executor_id()->CopyFrom(v1::DEFAULT_EXECUTOR_ID);
call.mutable_message()->set_data("executor message");
Future<http::Response> response = http::post(
slave.get()->pid,
"api/v1/executor",
headers,
serialize(ContentType::PROTOBUF, call),
stringify(ContentType::PROTOBUF));
AWAIT_EXPECT_RESPONSE_STATUS_EQ(http::Forbidden().status, response);
EXPECT_TRUE(strings::contains(response->body, failureMessage));
}
EXPECT_CALL(*executor, shutdown(_))
.Times(AtMost(1));
}
// This test verifies that default executor subscription fails if the executor
// provides a properly-signed authentication token with invalid claims.
TEST_F(ExecutorAuthorizationTest, FailedSubscribe)
{
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);
Result<Authorizer*> authorizer = Authorizer::create(acls);
ASSERT_SOME(authorizer);
slave::Flags flags = CreateSlaveFlags();
flags.acls = acls;
Owned<MasterDetector> detector = master.get()->createDetector();
auto executor = std::make_shared<v1::MockHTTPExecutor>();
v1::Resources resources =
v1::Resources::parse("cpus:0.1;mem:32;disk:32").get();
v1::ExecutorInfo executorInfo;
executorInfo.set_type(v1::ExecutorInfo::DEFAULT);
executorInfo.mutable_executor_id()->CopyFrom(v1::DEFAULT_EXECUTOR_ID);
executorInfo.mutable_resources()->CopyFrom(resources);
Owned<TestContainerizer> containerizer(
new TestContainerizer(devolve(executorInfo.executor_id()), executor));
// This pointer is passed to the agent, which will perform the cleanup.
Owned<MockSecretGenerator> mockSecretGenerator(new MockSecretGenerator());
Try<Owned<cluster::Slave>> slave = StartSlave(
detector.get(),
containerizer.get(),
mockSecretGenerator.get(),
authorizer.get(),
flags);
ASSERT_SOME(slave);
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
Future<Nothing> connected;
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(FutureSatisfy(&connected));
v1::scheduler::TestMesos mesos(
master.get()->pid,
ContentType::PROTOBUF,
scheduler);
AWAIT_READY(connected);
Future<v1::scheduler::Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
Future<v1::scheduler::Event::Offers> offers;
EXPECT_CALL(*scheduler, offers(_, _))
.WillOnce(FutureArg<1>(&offers))
.WillRepeatedly(Return()); // Ignore subsequent offers.
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return()); // Ignore heartbeats.
mesos.send(v1::createCallSubscribe(v1::DEFAULT_FRAMEWORK_INFO));
AWAIT_READY(subscribed);
v1::FrameworkID frameworkId(subscribed->framework_id());
executorInfo.mutable_framework_id()->CopyFrom(frameworkId);
AWAIT_READY(offers);
ASSERT_FALSE(offers->offers().empty());
Future<v1::executor::Mesos*> executorLib;
EXPECT_CALL(*executor, connected(_))
.WillOnce(FutureArg<0>(&executorLib));
Owned<JWTSecretGenerator> jwtSecretGenerator(
new JWTSecretGenerator(DEFAULT_JWT_SECRET_KEY));
// Create a principal which contains an incorrect ContainerID.
hashmap<string, string> claims;
claims["fid"] = frameworkId.value();
claims["eid"] = v1::DEFAULT_EXECUTOR_ID.value();
claims["cid"] = id::UUID::random().toString();
Principal principal(None(), claims);
// Generate an authentication token which is signed using the correct key,
// but contains an invalid set of claims.
Future<Secret> authenticationToken =
jwtSecretGenerator->generate(principal);
AWAIT_READY(authenticationToken);
EXPECT_CALL(*mockSecretGenerator, generate(_))
.WillOnce(Return(authenticationToken.get()));
const v1::Offer& offer = offers->offers(0);
const v1::AgentID& agentId = offer.agent_id();
{
v1::TaskInfo taskInfo =
v1::createTask(agentId, resources, SLEEP_COMMAND(1000));
v1::TaskGroupInfo taskGroup;
taskGroup.add_tasks()->CopyFrom(taskInfo);
v1::scheduler::Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.set_type(v1::scheduler::Call::ACCEPT);
v1::scheduler::Call::Accept* accept = call.mutable_accept();
accept->add_offer_ids()->CopyFrom(offer.id());
v1::Offer::Operation* operation = accept->add_operations();
operation->set_type(v1::Offer::Operation::LAUNCH_GROUP);
v1::Offer::Operation::LaunchGroup* launchGroup =
operation->mutable_launch_group();
launchGroup->mutable_executor()->CopyFrom(executorInfo);
launchGroup->mutable_task_group()->CopyFrom(taskGroup);
mesos.send(call);
}
AWAIT_READY(executorLib);
{
v1::executor::Call call;
call.mutable_framework_id()->CopyFrom(frameworkId);
call.mutable_executor_id()->CopyFrom(v1::DEFAULT_EXECUTOR_ID);
call.set_type(v1::executor::Call::SUBSCRIBE);
call.mutable_subscribe();
executorLib.get()->send(call);
}
Future<v1::executor::Event::Error> error;
EXPECT_CALL(*executor, error(_, _))
.WillOnce(FutureArg<1>(&error));
AWAIT_READY(error);
EXPECT_EQ(
error->message(),
"Received unexpected '403 Forbidden' () for SUBSCRIBE");
}
// The master reconciles operations that are missing from a re-registering
// agent.
//
// In this case, the `ApplyOperationMessage` is dropped, so the agent should
// respond with a OPERATION_DROPPED operation status update.
//
// This test verifies that if an operation ID is set, the framework receives
// the OPERATION_DROPPED operation status update.
//
// This is a regression test for MESOS-8784.
TEST_F(
MasterSlaveReconciliationTest,
ForwardOperationDroppedAfterExplicitReconciliation)
{
Clock::pause();
mesos::internal::master::Flags masterFlags = CreateMasterFlags();
Try<Owned<cluster::Master>> master = StartMaster(masterFlags);
ASSERT_SOME(master);
Future<UpdateSlaveMessage> updateSlaveMessage =
FUTURE_PROTOBUF(UpdateSlaveMessage(), _, _);
auto detector = std::make_shared<StandaloneMasterDetector>(master.get()->pid);
mesos::internal::slave::Flags slaveFlags = CreateSlaveFlags();
Try<Owned<cluster::Slave>> slave = StartSlave(detector.get(), slaveFlags);
ASSERT_SOME(slave);
// Advance the clock to trigger agent registration.
Clock::advance(slaveFlags.registration_backoff_factor);
// Wait for the agent to register.
AWAIT_READY(updateSlaveMessage);
// Start and register a resource provider.
v1::ResourceProviderInfo resourceProviderInfo;
resourceProviderInfo.set_type("org.apache.mesos.rp.test");
resourceProviderInfo.set_name("test");
v1::Resource disk = v1::createDiskResource(
"200", "*", None(), None(), v1::createDiskSourceRaw());
Owned<v1::MockResourceProvider> resourceProvider(
new v1::MockResourceProvider(resourceProviderInfo, v1::Resources(disk)));
// Make the mock resource provider answer to reconciliation events with
// OPERATION_DROPPED operation status updates.
auto reconcileOperations =
[&resourceProvider](
const v1::resource_provider::Event::ReconcileOperations& reconcile) {
foreach (const v1::UUID& operationUuid, reconcile.operation_uuids()) {
v1::resource_provider::Call call;
call.set_type(v1::resource_provider::Call::UPDATE_OPERATION_STATUS);
call.mutable_resource_provider_id()->CopyFrom(
resourceProvider->info.id());
v1::resource_provider::Call::UpdateOperationStatus*
updateOperationStatus = call.mutable_update_operation_status();
updateOperationStatus->mutable_status()->set_state(
v1::OPERATION_DROPPED);
updateOperationStatus->mutable_operation_uuid()->CopyFrom(
operationUuid);
resourceProvider->send(call);
}
};
EXPECT_CALL(*resourceProvider, reconcileOperations(_))
.WillOnce(Invoke(reconcileOperations));
Owned<EndpointDetector> endpointDetector(
mesos::internal::tests::resource_provider::createEndpointDetector(
slave.get()->pid));
updateSlaveMessage = FUTURE_PROTOBUF(UpdateSlaveMessage(), _, _);
// NOTE: We need to resume the clock so that the resource provider can
// fully register.
Clock::resume();
ContentType contentType = ContentType::PROTOBUF;
resourceProvider->start(endpointDetector, contentType);
// Wait until the agent's resources have been updated to include the
// resource provider resources.
AWAIT_READY(updateSlaveMessage);
Clock::pause();
// Start a v1 framework.
auto scheduler = std::make_shared<v1::MockHTTPScheduler>();
v1::FrameworkInfo frameworkInfo = v1::DEFAULT_FRAMEWORK_INFO;
frameworkInfo.set_roles(0, DEFAULT_TEST_ROLE);
EXPECT_CALL(*scheduler, connected(_))
.WillOnce(v1::scheduler::SendSubscribe(frameworkInfo));
Future<v1::scheduler::Event::Subscribed> subscribed;
EXPECT_CALL(*scheduler, subscribed(_, _))
.WillOnce(FutureArg<1>(&subscribed));
// Ignore heartbeats.
EXPECT_CALL(*scheduler, heartbeat(_))
.WillRepeatedly(Return());
Future<v1::scheduler::Event::Offers> offers;
EXPECT_CALL(*scheduler, offers(_, _))
.WillOnce(FutureArg<1>(&offers))
.WillRepeatedly(v1::scheduler::DeclineOffers());
v1::scheduler::TestMesos mesos(master.get()->pid, contentType, scheduler);
AWAIT_READY(subscribed);
v1::FrameworkID frameworkId(subscribed->framework_id());
AWAIT_READY(offers);
ASSERT_FALSE(offers->offers().empty());
const v1::Offer& offer = offers->offers(0);
// We'll drop the `ApplyOperationMessage` from the master to the agent.
Future<ApplyOperationMessage> applyOperationMessage =
DROP_PROTOBUF(ApplyOperationMessage(), master.get()->pid, _);
v1::Resources resources =
v1::Resources(offer.resources()).filter([](const v1::Resource& resource) {
return resource.has_provider_id();
});
ASSERT_FALSE(resources.empty());
v1::Resource reserved = *(resources.begin());
reserved.add_reservations()->CopyFrom(
v1::createDynamicReservationInfo(
frameworkInfo.roles(0), DEFAULT_CREDENTIAL.principal()));
v1::OperationID operationId;
operationId.set_value("operation");
mesos.send(v1::createCallAccept(
frameworkId, offer, {v1::RESERVE(reserved, operationId.value())}));
AWAIT_READY(applyOperationMessage);
Future<v1::scheduler::Event::UpdateOperationStatus> operationDroppedUpdate;
EXPECT_CALL(*scheduler, updateOperationStatus(_, _))
.WillOnce(FutureArg<1>(&operationDroppedUpdate));
// Simulate a spurious master change event (e.g., due to ZooKeeper
// expiration) at the slave to force re-registration.
detector->appoint(master.get()->pid);
// Advance the clock, so that the agent re-registers.
Clock::advance(slaveFlags.registration_backoff_factor);
// Wait for the framework to receive the OPERATION_DROPPED update.
AWAIT_READY(operationDroppedUpdate);
EXPECT_EQ(operationId, operationDroppedUpdate->status().operation_id());
EXPECT_EQ(v1::OPERATION_DROPPED, operationDroppedUpdate->status().state());
}