// 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));
}
// 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());
}