virtual void resourceOffers(SchedulerDriver* driver,
const vector<Offer>& offers)
{
for (size_t i = 0; i < offers.size(); i++) {
const Offer& offer = offers[i];
Resources remaining = offer.resources();
static Resources TASK_RESOURCES = Resources::parse(
"cpus:" + stringify<float>(CPUS_PER_TASK) +
";mem:" + stringify<size_t>(MEM_PER_TASK)).get();
size_t maxTasks = 0;
while (remaining.flatten().contains(TASK_RESOURCES)) {
maxTasks++;
remaining -= TASK_RESOURCES;
}
// Launch tasks.
vector<TaskInfo> tasks;
for (size_t i = 0; i < maxTasks / 2 && crawlQueue.size() > 0; i++) {
string url = crawlQueue.front();
crawlQueue.pop();
string urlId = "C" + stringify<size_t>(processed[url]);
TaskInfo task;
task.set_name("Crawler " + urlId);
task.mutable_task_id()->set_value(urlId);
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_executor()->MergeFrom(crawler);
task.mutable_resources()->MergeFrom(TASK_RESOURCES);
task.set_data(url);
tasks.push_back(task);
tasksLaunched++;
cout << "Crawler " << urlId << " " << url << endl;
}
for (size_t i = maxTasks/2; i < maxTasks && renderQueue.size() > 0; i++) {
string url = renderQueue.front();
renderQueue.pop();
string urlId = "R" + stringify<size_t>(processed[url]);
TaskInfo task;
task.set_name("Renderer " + urlId);
task.mutable_task_id()->set_value(urlId);
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_executor()->MergeFrom(renderer);
task.mutable_resources()->MergeFrom(TASK_RESOURCES);
task.set_data(url);
tasks.push_back(task);
tasksLaunched++;
cout << "Renderer " << urlId << " " << url << endl;
}
driver->launchTasks(offer.id(), tasks);
}
}
// This test checks that a scheduler exit shuts down the executor.
TEST_F(FaultToleranceTest, SchedulerExit)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
MockExecutor exec(DEFAULT_EXECUTOR_ID);
Try<PID<Slave> > slave = StartSlave(&exec);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(&sched, DEFAULT_FRAMEWORK_INFO, master.get());
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.get().size());
AWAIT_READY(offers);
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
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));
EXPECT_CALL(exec, registered(_, _, _, _));
EXPECT_CALL(exec, launchTask(_, _))
.WillOnce(SendStatusUpdateFromTask(TASK_RUNNING));
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(status);
EXPECT_EQ(TASK_RUNNING, status.get().state());
EXPECT_CALL(exec, shutdown(_))
.Times(AtMost(1));
driver.stop();
driver.join();
Shutdown();
}
void resourceOffers(const vector<Offer>& offers)
{
foreach (const Offer& offer, offers) {
cout << "Received offer " << offer.id() << " with "
<< Resources(offer.resources())
<< endl;
static const Resources TASK_RESOURCES = Resources::parse(
"cpus:" + stringify(CPUS_PER_TASK) +
";mem:" + stringify(MEM_PER_TASK)).get();
Resources remaining = offer.resources();
// Launch tasks.
vector<TaskInfo> tasks;
while (tasksLaunched < totalTasks &&
remaining.flatten().contains(TASK_RESOURCES)) {
int taskId = tasksLaunched++;
cout << "Launching task " << taskId << " using offer "
<< offer.id() << endl;
TaskInfo task;
task.set_name("Task " + lexical_cast<string>(taskId));
task.mutable_task_id()->set_value(
lexical_cast<string>(taskId));
task.mutable_agent_id()->MergeFrom(offer.agent_id());
task.mutable_executor()->MergeFrom(executor);
Option<Resources> resources =
remaining.find(TASK_RESOURCES.flatten(framework.role()));
CHECK_SOME(resources);
task.mutable_resources()->CopyFrom(resources.get());
remaining -= resources.get();
tasks.push_back(task);
}
Call call;
CHECK(framework.has_id());
call.mutable_framework_id()->CopyFrom(framework.id());
call.set_type(Call::ACCEPT);
Call::Accept* accept = call.mutable_accept();
accept->add_offer_ids()->CopyFrom(offer.id());
Offer::Operation* operation = accept->add_operations();
operation->set_type(Offer::Operation::LAUNCH);
foreach (const TaskInfo& taskInfo, tasks) {
operation->mutable_launch()->add_task_infos()->CopyFrom(taskInfo);
}
void offers(const vector<Offer>& offers)
{
CHECK_EQ(SUBSCRIBED, state);
static const Try<Resources> TASK_RESOURCES = Resources::parse(resources);
if (TASK_RESOURCES.isError()) {
EXIT(EXIT_FAILURE)
<< "Failed to parse resources '" << resources << "': "
<< TASK_RESOURCES.error();
}
foreach (const Offer& offer, offers) {
Resources offered = offer.resources();
if (!launched && offered.flatten().contains(TASK_RESOURCES.get())) {
TaskInfo task;
task.set_name(name);
task.mutable_task_id()->set_value(name);
task.mutable_agent_id()->MergeFrom(offer.agent_id());
// Takes resources first from the specified role, then from '*'.
Option<Resources> resources =
offered.find(TASK_RESOURCES.get().flatten(frameworkInfo.role()));
CHECK_SOME(resources);
task.mutable_resources()->CopyFrom(resources.get());
CommandInfo* commandInfo = task.mutable_command();
if (shell) {
CHECK_SOME(command);
commandInfo->set_shell(true);
commandInfo->set_value(command.get());
} else {
// TODO(gilbert): Treat 'command' as executable value and arguments.
commandInfo->set_shell(false);
}
if (environment.isSome()) {
Environment* environment_ = commandInfo->mutable_environment();
foreachpair (
const string& name, const string& value, environment.get()) {
Environment::Variable* environmentVariable =
environment_->add_variables();
environmentVariable->set_name(name);
environmentVariable->set_value(value);
}
}
// TODO(benh): Move this into utils, make more generic, and use in
// other tests.
vector<TaskInfo> createTasks(const Offer& offer)
{
TaskInfo task;
task.set_name("test-task");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_resources()->MergeFrom(offer.resources());
task.mutable_executor()->MergeFrom(DEFAULT_EXECUTOR_INFO);
vector<TaskInfo> tasks;
tasks.push_back(task);
return tasks;
}
virtual void resourceOffers(
SchedulerDriver* driver,
const vector<Offer>& offers)
{
static const Try<Resources> TASK_RESOURCES = Resources::parse(resources);
if (TASK_RESOURCES.isError()) {
cerr << "Failed to parse resources '" << resources
<< "': " << TASK_RESOURCES.error() << endl;
driver->abort();
return;
}
foreach (const Offer& offer, offers) {
if (!launched &&
Resources(offer.resources()).contains(TASK_RESOURCES.get())) {
TaskInfo task;
task.set_name(name);
task.mutable_task_id()->set_value(name);
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_resources()->CopyFrom(TASK_RESOURCES.get());
task.mutable_command()->set_value(command);
if (uri.isSome()) {
task.mutable_command()->add_uris()->set_value(uri.get());
}
if (dockerImage.isSome()) {
ContainerInfo containerInfo;
containerInfo.set_type(ContainerInfo::DOCKER);
ContainerInfo::DockerInfo dockerInfo;
dockerInfo.set_image(dockerImage.get());
containerInfo.mutable_docker()->CopyFrom(dockerInfo);
task.mutable_container()->CopyFrom(containerInfo);
}
vector<TaskInfo> tasks;
tasks.push_back(task);
driver->launchTasks(offer.id(), tasks);
cout << "task " << name << " submitted to slave "
<< offer.slave_id() << endl;
launched = true;
} else {
driver->declineOffer(offer.id());
}
}
}
inline TaskInfo createTask(
const Offer& offer,
const std::string& command,
const std::string& name = "test-task",
const std::string& id = UUID::random().toString())
{
TaskInfo task;
task.set_name(name);
task.mutable_task_id()->set_value(id);
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_resources()->MergeFrom(offer.resources());
task.mutable_command()->set_value(command);
return task;
}
virtual void resourceOffers(SchedulerDriver* driver,
const vector<Offer>& offers)
{
foreach (const Offer& offer, offers) {
cout << "Received offer " << offer.id() << " with " << offer.resources()
<< endl;
static const Resources TASK_RESOURCES = Resources::parse(
"cpus:" + stringify(CPUS_PER_TASK) +
";mem:" + stringify(MEM_PER_TASK)).get();
Resources remaining = offer.resources();
// Launch tasks.
vector<TaskInfo> tasks;
while (tasksLaunched < totalTasks &&
remaining.flatten().contains(TASK_RESOURCES)) {
int taskId = tasksLaunched++;
cout << "Launching task " << taskId << " using offer "
<< offer.id() << endl;
TaskInfo task;
task.set_name("Task " + lexical_cast<string>(taskId));
task.mutable_task_id()->set_value(lexical_cast<string>(taskId));
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_executor()->MergeFrom(executor);
Try<Resources> flattened = TASK_RESOURCES.flatten(role);
CHECK_SOME(flattened);
Option<Resources> resources = remaining.find(flattened.get());
CHECK_SOME(resources);
task.mutable_resources()->MergeFrom(resources.get());
remaining -= resources.get();
tasks.push_back(task);
}
driver->launchTasks(offer.id(), tasks);
}
// For use with a MockScheduler, for example:
// EXPECT_CALL(sched, resourceOffers(_, _))
// .WillOnce(LaunchTasks(TASKS, CPUS, MEM));
// Launches up to TASKS no-op tasks, if possible,
// each with CPUS cpus and MEM memory.
ACTION_P4(LaunchTasks, tasks, cpus, mem, role)
{
SchedulerDriver* driver = arg0;
std::vector<Offer> offers = arg1;
int numTasks = tasks;
int launched = 0;
for (size_t i = 0; i < offers.size(); i++) {
const Offer& offer = offers[i];
const Resources TASK_RESOURCES = Resources::parse(
"cpus:" + stringify(cpus) + ";mem:" + stringify(mem)).get();
int nextTaskId = 0;
std::vector<TaskInfo> tasks;
Resources remaining = offer.resources();
while (TASK_RESOURCES <= remaining.flatten() && launched < numTasks) {
TaskInfo task;
task.set_name("TestTask");
task.mutable_task_id()->set_value(stringify(nextTaskId++));
task.mutable_slave_id()->MergeFrom(offer.slave_id());
ExecutorInfo executor;
executor.mutable_executor_id()->set_value("default");
executor.mutable_command()->set_value(":");
task.mutable_executor()->MergeFrom(executor);
Option<Resources> resources = remaining.find(TASK_RESOURCES, role);
CHECK_SOME(resources);
task.mutable_resources()->MergeFrom(resources.get());
remaining -= resources.get();
tasks.push_back(task);
launched++;
}
driver->launchTasks(offer.id(), tasks);
}
}
virtual void resourceOffers(SchedulerDriver* driver,
const vector<Offer>& offers)
{
cout << "." << flush;
for (size_t i = 0; i < offers.size(); i++) {
const Offer& offer = offers[i];
static const Resources TASK_RESOURCES = Resources::parse(
"cpus:" + stringify(CPUS_PER_TASK) +
";mem:" + stringify(MEM_PER_TASK)).get();
Resources remaining = offer.resources();
// Launch tasks.
vector<TaskInfo> tasks;
while (tasksLaunched < totalTasks &&
TASK_RESOURCES <= remaining.flatten()) {
int taskId = tasksLaunched++;
cout << "Starting task " << taskId << " on "
<< offer.hostname() << endl;
TaskInfo task;
task.set_name("Task " + lexical_cast<string>(taskId));
task.mutable_task_id()->set_value(lexical_cast<string>(taskId));
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_executor()->MergeFrom(executor);
Option<Resources> resources = remaining.find(TASK_RESOURCES, role);
CHECK_SOME(resources);
task.mutable_resources()->MergeFrom(resources.get());
remaining -= resources.get();
tasks.push_back(task);
}
driver->launchTasks(offer.id(), tasks);
}
}
inline TaskInfo createTask(
const Offer& offer,
const std::string& command,
const Option<mesos::ExecutorID>& executorId = None(),
const std::string& name = "test-task",
const std::string& id = UUID::random().toString())
{
TaskInfo task;
task.set_name(name);
task.mutable_task_id()->set_value(id);
task.mutable_slave_id()->CopyFrom(offer.slave_id());
task.mutable_resources()->CopyFrom(offer.resources());
if (executorId.isSome()) {
ExecutorInfo executor;
executor.mutable_executor_id()->CopyFrom(executorId.get());
executor.mutable_command()->set_value(command);
task.mutable_executor()->CopyFrom(executor);
} else {
task.mutable_command()->set_value(command);
}
return task;
}
vector<TaskInfo> populateTasks(
const string& cmd,
CommandInfo healthCommand,
const Offer& offer,
int gracePeriodSeconds = 0,
const Option<int>& consecutiveFailures = None(),
const Option<map<string, string> >& env = None())
{
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->CopyFrom(offer.slave_id());
task.mutable_resources()->CopyFrom(offer.resources());
CommandInfo command;
command.set_value(cmd);
Environment::Variable* variable =
command.mutable_environment()->add_variables();
// We need to set the correct directory to launch health check process
// instead of the default for tests.
variable->set_name("MESOS_LAUNCHER_DIR");
variable->set_value(path::join(tests::flags.build_dir, "src"));
task.mutable_command()->CopyFrom(command);
HealthCheck healthCheck;
if (env.isSome()) {
foreachpair (const string& name, const string value, env.get()) {
Environment::Variable* variable =
healthCommand.mutable_environment()->mutable_variables()->Add();
variable->set_name(name);
variable->set_value(value);
}
}
// This test runs a command without the command user field set. The
// command will verify the assumption that the command is run as the
// slave user (in this case, root).
TEST_F(SlaveTest, ROOT_RunTaskWithCommandInfoWithoutUser)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
// Need flags for 'executor_registration_timeout'.
slave::Flags flags = CreateSlaveFlags();
flags.isolation = "posix/cpu,posix/mem";
Try<MesosContainerizer*> containerizer =
MesosContainerizer::create(flags, false);
CHECK_SOME(containerizer);
Try<PID<Slave> > slave = StartSlave(containerizer.get());
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _))
.Times(1);
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.get().size());
// Launch a task with the command executor.
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
Result<string> user = os::user();
CHECK_SOME(user) << "Failed to get current user name"
<< (user.isError() ? ": " + user.error() : "");
// Command executor will run as user running test.
CommandInfo command;
command.set_value("test `whoami` = " + user.get());
task.mutable_command()->MergeFrom(command);
vector<TaskInfo> tasks;
tasks.push_back(task);
Future<TaskStatus> statusRunning;
Future<TaskStatus> statusFinished;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&statusRunning))
.WillOnce(FutureArg<1>(&statusFinished));
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(statusRunning);
EXPECT_EQ(TASK_RUNNING, statusRunning.get().state());
AWAIT_READY(statusFinished);
EXPECT_EQ(TASK_FINISHED, statusFinished.get().state());
driver.stop();
driver.join();
Shutdown(); // Must shutdown before 'containerizer' gets deallocated.
}
// This test verifies that when an executor terminates before
// registering with slave, it is properly cleaned up.
TEST_F(SlaveTest, RemoveUnregisteredTerminatedExecutor)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
MockExecutor exec(DEFAULT_EXECUTOR_ID);
TestContainerizer containerizer(&exec);
Try<PID<Slave> > slave = StartSlave(&containerizer);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _))
.Times(1);
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.get().size());
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
task.mutable_executor()->MergeFrom(DEFAULT_EXECUTOR_INFO);
vector<TaskInfo> tasks;
tasks.push_back(task);
// Drop the registration message from the executor to the slave.
Future<process::Message> registerExecutorMessage =
DROP_MESSAGE(Eq(RegisterExecutorMessage().GetTypeName()), _, _);
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(registerExecutorMessage);
Future<TaskStatus> status;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&status));
Future<Nothing> schedule =
FUTURE_DISPATCH(_, &GarbageCollectorProcess::schedule);
// Now kill the executor.
containerizer.destroy(offers.get()[0].framework_id(), DEFAULT_EXECUTOR_ID);
AWAIT_READY(status);
EXPECT_EQ(TASK_LOST, status.get().state());
// We use 'gc.schedule' as a signal for the executor being cleaned
// up by the slave.
AWAIT_READY(schedule);
EXPECT_CALL(exec, shutdown(_))
.Times(AtMost(1));
driver.stop();
driver.join();
Shutdown(); // Must shutdown before 'containerizer' gets deallocated.
}
// Test that we can run the mesos-executor and specify an "override"
// command to use via the --override argument.
TEST_F(SlaveTest, MesosExecutorWithOverride)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
TestContainerizer containerizer;
Try<PID<Slave> > slave = StartSlave(&containerizer);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _))
.Times(1);
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.get().size());
// Launch a task with the command executor.
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
CommandInfo command;
command.set_value("sleep 10");
task.mutable_command()->MergeFrom(command);
vector<TaskInfo> tasks;
tasks.push_back(task);
// Expect the launch and just assume it was sucessful since we'll be
// launching the executor ourselves manually below.
Future<Nothing> launch;
EXPECT_CALL(containerizer, launch(_, _, _, _, _, _, _))
.WillOnce(DoAll(FutureSatisfy(&launch),
Return(true)));
// Expect wait after launch is called but don't return anything
// until after we've finished everything below.
Future<Nothing> wait;
process::Promise<containerizer::Termination> promise;
EXPECT_CALL(containerizer, wait(_))
.WillOnce(DoAll(FutureSatisfy(&wait),
Return(promise.future())));
driver.launchTasks(offers.get()[0].id(), tasks);
// Once we get the launch the mesos-executor with --override.
AWAIT_READY(launch);
// Set up fake environment for executor.
map<string, string> environment;
environment["MESOS_SLAVE_PID"] = stringify(slave.get());
environment["MESOS_SLAVE_ID"] = stringify(offers.get()[0].slave_id());
environment["MESOS_FRAMEWORK_ID"] = stringify(offers.get()[0].framework_id());
environment["MESOS_EXECUTOR_ID"] = stringify(task.task_id());
environment["MESOS_DIRECTORY"] = "";
// Create temporary file to store validation string. If command is
// succesfully replaced, this file will end up containing the string
// 'Hello World\n'. Otherwise, the original task command i.e.
// 'sleep' will be called and the test will fail.
Try<std::string> file = os::mktemp();
ASSERT_SOME(file);
string executorCommand =
path::join(tests::flags.build_dir, "src", "mesos-executor") +
" --override -- /bin/sh -c 'echo hello world >" + file.get() + "'";
// Expect two status updates, one for once the mesos-executor says
// the task is running and one for after our overridden command
// above finishes.
Future<TaskStatus> status1, status2;
EXPECT_CALL(sched, statusUpdate(_, _))
.WillOnce(FutureArg<1>(&status1))
.WillOnce(FutureArg<1>(&status2));
Try<process::Subprocess> executor =
process::subprocess(
executorCommand,
process::Subprocess::PIPE(),
process::Subprocess::PIPE(),
process::Subprocess::PIPE(),
environment);
ASSERT_SOME(executor);
// Scheduler should receive the TASK_RUNNING update.
AWAIT_READY(status1);
ASSERT_EQ(TASK_RUNNING, status1.get().state());
AWAIT_READY(status2);
ASSERT_EQ(TASK_FINISHED, status2.get().state());
AWAIT_READY(wait);
containerizer::Termination termination;
termination.set_killed(false);
termination.set_message("Killed executor");
termination.set_status(0);
promise.set(termination);
driver.stop();
driver.join();
AWAIT_READY(executor.get().status());
// Verify file contents.
Try<std::string> validate = os::read(file.get());
ASSERT_SOME(validate);
EXPECT_EQ(validate.get(), "hello world\n");
os::rm(file.get());
Shutdown();
}
void ChapelScheduler::resourceOffers(SchedulerDriver* driver,
const vector<Offer>& offers)
{
// offers only contain resources describing a single node -> for more details read include/mesos/mesos.proto
//
cout << "***\tProcessing Offers!" << endl;
const int remainingCpusReq = cpusReq - launchedTsks.size();
if(remainingCpusReq == 0) {
for(size_t k = 0; k < offers.size(); k++) {
const Offer& offer = offers[k];
driver->declineOffer(offer.id());
}
cout << "\t\tChapelScheduler declined offer because resource requirements satisfied" << endl;
}
// cycle through all the offers and resource a task
// each offer corresponds to a single compute node
//
const static Resources TASK_RESOURCES = Resources::parse(mesosReq).get();
vector<TaskInfo> tsks;
for(size_t i = 0; i < offers.size(); i++) {
const Offer& offer = offers[i];
if(tsks.size() == remainingCpusReq) {
driver->declineOffer(offer.id());
continue; // need to cycle through the remaining offers and decline them
}
Resources remaining = offer.resources();
/* attempting to exercise multi-tenancy capabilities in mesos
* given an offer from a node, try to maximize the number of jobs
* that can be allocated to that node given the job's resource
* requirements
*
* if the desired number of nodes and jobs are met, then launch
* all the jobs on that node's offer
*
* this means some nodes will get multiple tasks assigned for
* execution
*/
vector<TaskInfo> tol;
while(remaining.flatten().contains(TASK_RESOUCES) && ((remainingCpusReq-tsks.size()) > 0)) {
const string tid = stringify<size_t>(tsks.size());
TaskInfo task;
task.set_name("Chapel Remote Program Task\t" + tid);
task.mutable_task_id()->set_value(tid);
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_command()->MergeFrom(chplCmdInfo);
task.mutable_resources()->MergeFrom(TASK_RESOURCES);
task.set_data(remoteCmd);
tol.push_back(task); // tol means "to launch"
tsks.push_back(task); // tsks tracks tasks launched for framework termination purposes
remaining-=TASK_RESOURCES;
tasksLaunched+=1;
cout << "\t\t+++\tLaunching # of Tasks!\t" << tol.size() << " of " << tasksLaunched << endl;
}
// after all the tasks for this offer have been "resourced"
// launch the tasks using this offer.id
//
driver->launchTasks(offer.id(), tol);
}
const size_t pendingTsksSize = tsks.size();
cout << endl << "\tAcquired # tasks " << pendingTsksSize << " required # of tasks " << cpusReq << " remaining required # tasks " << remainingCpusReq << endl << endl;
if(pendingTsksSize > 0) {
for(vector<TaskInfo>::iterator i = tsks.begin(); i != tsks.end(); i++) {
launchedTsks.insert(make_pair(i->task_id().value(), *i));
}
}
}
// This test ensures that a killTask() can happen between runTask()
// and _runTask() and then gets "handled properly". This means that
// the task never gets started, but also does not get lost. The end
// result is status TASK_KILLED. Essentially, killing the task is
// realized while preparing to start it. See MESOS-947.
// Temporarily disabled due to MESOS-1945.
TEST_F(SlaveTest, DISABLED_KillTaskBetweenRunTaskParts)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
MockExecutor exec(DEFAULT_EXECUTOR_ID);
TestContainerizer containerizer(&exec);
StandaloneMasterDetector detector(master.get());
MockSlave slave(CreateSlaveFlags(), &detector, &containerizer);
process::spawn(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _))
.Times(1);
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.get().size());
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
task.mutable_executor()->MergeFrom(DEFAULT_EXECUTOR_INFO);
vector<TaskInfo> tasks;
tasks.push_back(task);
EXPECT_CALL(exec, registered(_, _, _, _))
.Times(0);
EXPECT_CALL(exec, launchTask(_, _))
.Times(0);
EXPECT_CALL(exec, shutdown(_))
.Times(0);
Future<TaskStatus> status;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillRepeatedly(FutureArg<1>(&status));
EXPECT_CALL(slave, runTask(_, _, _, _, _))
.WillOnce(Invoke(&slave, &MockSlave::unmocked_runTask));
// Saved arguments from Slave::_runTask().
Future<bool> future;
FrameworkInfo frameworkInfo;
FrameworkID frameworkId;
// Skip what Slave::_runTask() normally does, save its arguments for
// later, tie reaching the critical moment when to kill the task to
// a future.
Future<Nothing> _runTask;
EXPECT_CALL(slave, _runTask(_, _, _, _, _))
.WillOnce(DoAll(FutureSatisfy(&_runTask),
SaveArg<0>(&future),
SaveArg<1>(&frameworkInfo),
SaveArg<2>(&frameworkId)));
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(_runTask);
Future<Nothing> killTask;
EXPECT_CALL(slave, killTask(_, _, _))
.WillOnce(DoAll(Invoke(&slave, &MockSlave::unmocked_killTask),
FutureSatisfy(&killTask)));
driver.killTask(task.task_id());
// Since this is the only task ever for this framework, the
// framework should get removed in Slave::_runTask().
// Thus we can observe that this happens before Shutdown().
Future<Nothing> removeFramework;
EXPECT_CALL(slave, removeFramework(_))
.WillOnce(DoAll(Invoke(&slave, &MockSlave::unmocked_removeFramework),
FutureSatisfy(&removeFramework)));
AWAIT_READY(killTask);
slave.unmocked__runTask(
future, frameworkInfo, frameworkId, master.get(), task);
AWAIT_READY(removeFramework);
AWAIT_READY(status);
EXPECT_EQ(TASK_KILLED, status.get().state());
driver.stop();
driver.join();
process::terminate(slave);
process::wait(slave);
Shutdown(); // Must shutdown before 'containerizer' gets deallocated.
}
// The purpose of this test is to ensure that when slaves are removed
// from the master, and then attempt to send exited executor messages,
// we send a ShutdownMessage to the slave. Why? Because during a
// network partition, the master will remove a partitioned slave, thus
// sending its tasks to LOST. At this point, when the partition is
// removed, the slave may attempt to send exited executor messages if
// it was unaware that the master removed it. We've already
// notified frameworks that the tasks under the executors were LOST,
// so we have to have the slave shut down.
TEST_F(PartitionTest, PartitionedSlaveExitedExecutor)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
// Allow the master to PING the slave, but drop all PONG messages
// from the slave. Note that we don't match on the master / slave
// PIDs because it's actually the SlaveObserver Process that sends
// the pings.
Future<Message> ping = FUTURE_MESSAGE(Eq("PING"), _, _);
DROP_MESSAGES(Eq("PONG"), _, _);
MockExecutor exec(DEFAULT_EXECUTOR_ID);
TestContainerizer containerizer(&exec);
Try<PID<Slave> > slave = StartSlave(&containerizer);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), 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());
driver.start();
AWAIT_READY(frameworkId);
AWAIT_READY(offers);
ASSERT_NE(0u, offers.get().size());
// Launch a task. This allows us to have the slave send an
// ExitedExecutorMessage.
TaskID taskId;
taskId.set_value("1");
TaskInfo task;
task.set_name("");
task.mutable_task_id()->MergeFrom(taskId);
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
task.mutable_executor()->MergeFrom(DEFAULT_EXECUTOR_INFO);
task.mutable_executor()->mutable_command()->set_value("sleep 60");
vector<TaskInfo> tasks;
tasks.push_back(task);
// Set up the expectations for launching the task.
EXPECT_CALL(exec, registered(_, _, _, _));
EXPECT_CALL(exec, launchTask(_, _))
.WillOnce(SendStatusUpdateFromTask(TASK_RUNNING));
// Drop all the status updates from the slave, so that we can
// ensure the ExitedExecutorMessage is what triggers the slave
// shutdown.
DROP_PROTOBUFS(StatusUpdateMessage(), _, master.get());
driver.launchTasks(offers.get()[0].id(), tasks);
// Drop the first shutdown message from the master (simulated
// partition) and allow the second shutdown message to pass when
// triggered by the ExitedExecutorMessage.
Future<ShutdownMessage> shutdownMessage =
DROP_PROTOBUF(ShutdownMessage(), _, slave.get());
Future<TaskStatus> lostStatus;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&lostStatus));
Future<Nothing> slaveLost;
EXPECT_CALL(sched, slaveLost(&driver, _))
.WillOnce(FutureSatisfy(&slaveLost));
Clock::pause();
// Now, induce a partition of the slave by having the master
// timeout the slave.
uint32_t pings = 0;
while (true) {
AWAIT_READY(ping);
pings++;
if (pings == master::MAX_SLAVE_PING_TIMEOUTS) {
break;
}
ping = FUTURE_MESSAGE(Eq("PING"), _, _);
Clock::advance(master::SLAVE_PING_TIMEOUT);
Clock::settle();
}
Clock::advance(master::SLAVE_PING_TIMEOUT);
Clock::settle();
// The master will have notified the framework of the lost task.
AWAIT_READY(lostStatus);
EXPECT_EQ(TASK_LOST, lostStatus.get().state());
// Wait for the master to attempt to shut down the slave.
AWAIT_READY(shutdownMessage);
// The master will notify the framework that the slave was lost.
AWAIT_READY(slaveLost);
shutdownMessage = FUTURE_PROTOBUF(ShutdownMessage(), _, slave.get());
// Induce an ExitedExecutorMessage from the slave.
containerizer.destroy(
frameworkId.get(), DEFAULT_EXECUTOR_INFO.executor_id());
// Upon receiving the message, the master will shutdown the slave.
AWAIT_READY(shutdownMessage);
Clock::resume();
driver.stop();
driver.join();
Shutdown();
}
// This test has been temporarily disabled due to MESOS-1257.
TEST_F(ExternalContainerizerTest, DISABLED_Launch)
{
Try<PID<Master> > master = this->StartMaster();
ASSERT_SOME(master);
Flags testFlags;
slave::Flags flags = this->CreateSlaveFlags();
flags.isolation = "external";
flags.containerizer_path =
testFlags.build_dir + "/src/examples/python/test-containerizer";
MockExternalContainerizer containerizer(flags);
Try<PID<Slave> > slave = this->StartSlave(&containerizer, flags);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), 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.get().size());
TaskInfo task;
task.set_name("isolator_test");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->CopyFrom(offers.get()[0].slave_id());
task.mutable_resources()->CopyFrom(offers.get()[0].resources());
Resources resources(offers.get()[0].resources());
Option<Bytes> mem = resources.mem();
ASSERT_SOME(mem);
Option<double> cpus = resources.cpus();
ASSERT_SOME(cpus);
const std::string& file = path::join(flags.work_dir, "ready");
// This task induces user/system load in a child process by
// running top in a child process for ten seconds.
task.mutable_command()->set_value(
#ifdef __APPLE__
// Use logging mode with 30,000 samples with no interval.
"top -l 30000 -s 0 2>&1 > /dev/null & "
#else
// Batch mode, with 30,000 samples with no interval.
"top -b -d 0 -n 30000 2>&1 > /dev/null & "
#endif
"touch " + file + "; " // Signals that the top command is running.
"sleep 60");
Future<TaskStatus> status;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&status))
.WillRepeatedly(Return()); // Ignore rest for now.
Future<ContainerID> containerId;
EXPECT_CALL(containerizer, launch(_, _, _, _, _, _, _, _))
.WillOnce(DoAll(FutureArg<0>(&containerId),
Invoke(&containerizer,
&MockExternalContainerizer::_launch)));
driver.launchTasks(offers.get()[0].id(), {task});
AWAIT_READY(containerId);
AWAIT_READY(status);
EXPECT_EQ(TASK_RUNNING, status.get().state());
// Wait for the task to begin inducing cpu time.
while (!os::exists(file));
ExecutorID executorId;
executorId.set_value(task.task_id().value());
// We'll wait up to 10 seconds for the child process to induce
// 1/8 of a second of user and system cpu time in total.
// TODO(bmahler): Also induce rss memory consumption, by re-using
// the balloon framework.
ResourceStatistics statistics;
Duration waited = Duration::zero();
do {
Future<ResourceStatistics> usage = containerizer.usage(containerId.get());
AWAIT_READY(usage);
statistics = usage.get();
// If we meet our usage expectations, we're done!
// NOTE: We are currently getting dummy-data from the test-
// containerizer python script matching these expectations.
// TODO(tillt): Consider working with real data.
if (statistics.cpus_user_time_secs() >= 0.120 &&
statistics.cpus_system_time_secs() >= 0.05 &&
statistics.mem_rss_bytes() >= 1024u) {
break;
}
os::sleep(Milliseconds(100));
waited += Milliseconds(100);
} while (waited < Seconds(10));
EXPECT_GE(statistics.cpus_user_time_secs(), 0.120);
EXPECT_GE(statistics.cpus_system_time_secs(), 0.05);
EXPECT_EQ(statistics.cpus_limit(), cpus.get());
EXPECT_GE(statistics.mem_rss_bytes(), 1024u);
EXPECT_EQ(statistics.mem_limit_bytes(), mem.get().bytes());
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&status));
driver.killTask(task.task_id());
AWAIT_READY(status);
EXPECT_EQ(TASK_KILLED, status.get().state());
driver.stop();
driver.join();
this->Shutdown();
}
// This test verifies that the slave run task label decorator can add
// and remove labels from a task during the launch sequence. A task
// with two labels ("foo":"bar" and "bar":"baz") is launched and will
// get modified by the slave hook to strip the "foo":"bar" pair and
// add a new "baz":"qux" pair.
TEST_F(HookTest, VerifySlaveRunTaskHook)
{
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);
ASSERT_EQ(1u, offers.get().size());
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->CopyFrom(offers.get()[0].slave_id());
task.mutable_resources()->CopyFrom(offers.get()[0].resources());
task.mutable_executor()->CopyFrom(DEFAULT_EXECUTOR_INFO);
// Add two labels: (1) will be removed by the hook to ensure that
// runTaskHook can remove labels (2) will be preserved to ensure
// that the framework can add labels to the task and have those be
// available by the end of the launch task sequence when hooks are
// used (to protect against hooks removing labels completely).
Labels* labels = task.mutable_labels();
labels->add_labels()->CopyFrom(createLabel("foo", "bar"));
labels->add_labels()->CopyFrom(createLabel("bar", "baz"));
EXPECT_CALL(exec, registered(_, _, _, _));
Future<TaskInfo> taskInfo;
EXPECT_CALL(exec, launchTask(_, _))
.WillOnce(DoAll(
FutureArg<1>(&taskInfo),
SendStatusUpdateFromTask(TASK_RUNNING)));
driver.launchTasks(offers.get()[0].id(), {task});
AWAIT_READY(taskInfo);
// The master hook will hang an extra label off.
const Labels& labels_ = taskInfo.get().labels();
ASSERT_EQ(3, labels_.labels_size());
// The slave run task hook will prepend a new "baz":"qux" label.
EXPECT_EQ("baz", labels_.labels(0).key());
EXPECT_EQ("qux", labels_.labels(0).value());
// Master launch task hook will still hang off test label.
EXPECT_EQ(testLabelKey, labels_.labels(1).key());
EXPECT_EQ(testLabelValue, labels_.labels(1).value());
// And lastly, we only expect the "foo":"bar" pair to be stripped by
// the module. The last pair should be the original "bar":"baz"
// pair set by the test.
EXPECT_EQ("bar", labels_.labels(2).key());
EXPECT_EQ("baz", labels_.labels(2).value());
EXPECT_CALL(exec, shutdown(_))
.Times(AtMost(1));
driver.stop();
driver.join();
}
// Test that the prepare launch docker hook execute before launch
// a docker container. Test hook create a file "foo" in the sandbox
// directory. When the docker container launched, the sandbox directory
// is mounted to the docker container. We validate the hook by verifying
// the "foo" file exists in the docker container or not.
TEST_F(HookTest, ROOT_DOCKER_VerifySlavePreLaunchDockerHook)
{
Try<Owned<cluster::Master>> master = StartMaster();
ASSERT_SOME(master);
MockDocker* mockDocker =
new MockDocker(tests::flags.docker, tests::flags.docker_socket);
Shared<Docker> docker(mockDocker);
slave::Flags flags = CreateSlaveFlags();
Fetcher fetcher;
Try<ContainerLogger*> logger =
ContainerLogger::create(flags.container_logger);
ASSERT_SOME(logger);
MockDockerContainerizer containerizer(
flags,
&fetcher,
Owned<ContainerLogger>(logger.get()),
docker);
Owned<MasterDetector> detector = master.get()->createDetector();
Try<Owned<cluster::Slave>> slave =
StartSlave(detector.get(), &containerizer, 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);
ASSERT_NE(0u, offers.get().size());
const Offer& offer = offers.get()[0];
SlaveID slaveId = offer.slave_id();
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->CopyFrom(offer.slave_id());
task.mutable_resources()->CopyFrom(offer.resources());
CommandInfo command;
command.set_value("test -f " + path::join(flags.sandbox_directory, "foo"));
ContainerInfo containerInfo;
containerInfo.set_type(ContainerInfo::DOCKER);
// TODO(tnachen): Use local image to test if possible.
ContainerInfo::DockerInfo dockerInfo;
dockerInfo.set_image("alpine");
containerInfo.mutable_docker()->CopyFrom(dockerInfo);
task.mutable_command()->CopyFrom(command);
task.mutable_container()->CopyFrom(containerInfo);
vector<TaskInfo> tasks;
tasks.push_back(task);
Future<ContainerID> containerId;
EXPECT_CALL(containerizer, launch(_, _, _, _, _, _, _, _))
.WillOnce(DoAll(FutureArg<0>(&containerId),
Invoke(&containerizer,
&MockDockerContainerizer::_launch)));
Future<TaskStatus> statusRunning;
Future<TaskStatus> statusFinished;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&statusRunning))
.WillOnce(FutureArg<1>(&statusFinished))
.WillRepeatedly(DoDefault());
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY_FOR(containerId, Seconds(60));
AWAIT_READY_FOR(statusRunning, Seconds(60));
EXPECT_EQ(TASK_RUNNING, statusRunning.get().state());
AWAIT_READY_FOR(statusFinished, Seconds(60));
EXPECT_EQ(TASK_FINISHED, statusFinished.get().state());
Future<containerizer::Termination> termination =
containerizer.wait(containerId.get());
driver.stop();
driver.join();
AWAIT_READY(termination);
Future<list<Docker::Container>> containers =
docker.get()->ps(true, slave::DOCKER_NAME_PREFIX);
AWAIT_READY(containers);
// Cleanup all mesos launched containers.
foreach (const Docker::Container& container, containers.get()) {
AWAIT_READY_FOR(docker.get()->rm(container.id, true), Seconds(30));
}
}
// 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();
}
// Test executor environment decorator hook and remove executor hook
// for slave. We expect the environment-decorator hook to create a
// temporary file and the remove-executor hook to delete that file.
TEST_F(HookTest, VerifySlaveLaunchExecutorHook)
{
master::Flags masterFlags = CreateMasterFlags();
Try<Owned<cluster::Master>> master = StartMaster(masterFlags);
ASSERT_SOME(master);
slave::Flags slaveFlags = CreateSlaveFlags();
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.get().size());
// Launch a task with the command executor.
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->CopyFrom(offers.get()[0].slave_id());
task.mutable_resources()->CopyFrom(offers.get()[0].resources());
task.mutable_executor()->CopyFrom(DEFAULT_EXECUTOR_INFO);
EXPECT_CALL(exec, registered(_, _, _, _));
EXPECT_CALL(exec, launchTask(_, _))
.WillOnce(SendStatusUpdateFromTask(TASK_RUNNING));
// Executor shutdown would force the Slave to execute the
// remove-executor hook.
EXPECT_CALL(exec, shutdown(_));
Future<TaskStatus> status;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&status))
.WillRepeatedly(Return());
// On successful completion of the "slaveLaunchExecutorHook", the
// test hook will send a HookExecuted message to itself. We wait
// until that message is intercepted by the testing infrastructure.
Future<HookExecuted> hookFuture = FUTURE_PROTOBUF(HookExecuted(), _, _);
driver.launchTasks(offers.get()[0].id(), {task});
AWAIT_READY(status);
driver.stop();
driver.join();
// The scheduler shutdown from above forces the executor to
// shutdown. This in turn should force the Slave to execute
// the remove-executor hook.
// Here, we wait for the hook to finish execution.
AWAIT_READY(hookFuture);
}
// 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();
}
// This test verifies that an authorized task launch is successful.
TEST_F(MasterAuthorizationTest, AuthorizedTask)
{
// Setup ACLs so that the framework can launch tasks as "foo".
ACLs acls;
mesos::ACL::RunTasks* acl = acls.add_run_tasks();
acl->mutable_principals()->add_values(DEFAULT_FRAMEWORK_INFO.principal());
acl->mutable_users()->add_values("foo");
master::Flags flags = CreateMasterFlags();
flags.acls = acls;
Try<PID<Master> > master = StartMaster(flags);
ASSERT_SOME(master);
// Create an authorized executor.
ExecutorInfo executor; // Bug in gcc 4.1.*, must assign on next line.
executor = CREATE_EXECUTOR_INFO("test-executor", "exit 1");
executor.mutable_command()->set_user("foo");
MockExecutor exec(executor.executor_id());
Try<PID<Slave> > slave = StartSlave(&exec);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _))
.Times(1);
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.get().size());
// Create an authorized task.
TaskInfo task;
task.set_name("test");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
task.mutable_executor()->MergeFrom(executor);
vector<TaskInfo> tasks;
tasks.push_back(task);
EXPECT_CALL(exec, registered(_, _, _, _))
.Times(1);
EXPECT_CALL(exec, launchTask(_, _))
.WillOnce(SendStatusUpdateFromTask(TASK_RUNNING));
Future<TaskStatus> status;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&status));
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(status);
EXPECT_EQ(TASK_RUNNING, status.get().state());
EXPECT_CALL(exec, shutdown(_))
.Times(AtMost(1));
driver.stop();
driver.join();
Shutdown(); // Must shutdown before 'containerizer' gets deallocated.
}
virtual void resourceOffers(
SchedulerDriver* driver,
const vector<Offer>& offers)
{
static const Try<Resources> TASK_RESOURCES = Resources::parse(resources);
if (TASK_RESOURCES.isError()) {
cerr << "Failed to parse resources '" << resources
<< "': " << TASK_RESOURCES.error() << endl;
driver->abort();
return;
}
foreach (const Offer& offer, offers) {
if (!launched &&
Resources(offer.resources()).contains(TASK_RESOURCES.get())) {
TaskInfo task;
task.set_name(name);
task.mutable_task_id()->set_value(name);
task.mutable_slave_id()->MergeFrom(offer.slave_id());
task.mutable_resources()->CopyFrom(TASK_RESOURCES.get());
CommandInfo* commandInfo = task.mutable_command();
commandInfo->set_value(command);
if (environment.isSome()) {
Environment* environment_ = commandInfo->mutable_environment();
foreachpair (const std::string& name,
const std::string& value,
environment.get()) {
Environment_Variable* environmentVariable =
environment_->add_variables();
environmentVariable->set_name(name);
environmentVariable->set_value(value);
}
}
if (uri.isSome()) {
task.mutable_command()->add_uris()->set_value(uri.get());
}
if (dockerImage.isSome()) {
ContainerInfo containerInfo;
if (containerizer == "mesos") {
containerInfo.set_type(ContainerInfo::MESOS);
ContainerInfo::MesosInfo mesosInfo;
Image mesosImage;
mesosImage.set_type(Image::DOCKER);
mesosImage.mutable_docker()->set_name(dockerImage.get());
mesosInfo.mutable_image()->CopyFrom(mesosImage);
containerInfo.mutable_mesos()->CopyFrom(mesosInfo);
} else if (containerizer == "docker") {
containerInfo.set_type(ContainerInfo::DOCKER);
ContainerInfo::DockerInfo dockerInfo;
dockerInfo.set_image(dockerImage.get());
containerInfo.mutable_docker()->CopyFrom(dockerInfo);
} else {
cerr << "Unsupported containerizer: " << containerizer << endl;;
driver->abort();
return;
}
task.mutable_container()->CopyFrom(containerInfo);
}
vector<TaskInfo> tasks;
tasks.push_back(task);
driver->launchTasks(offer.id(), tasks);
cout << "task " << name << " submitted to slave "
<< offer.slave_id() << endl;
launched = true;
} else {
// This test runs a command _with_ the command user field set. The
// command will verify the assumption that the command is run as the
// specified user. We use (and assume the precense) of the
// unprivileged 'nobody' user which should be available on both Linux
// and Mac OS X.
TEST_F(SlaveTest, DISABLED_ROOT_RunTaskWithCommandInfoWithUser)
{
// TODO(nnielsen): Introduce STOUT abstraction for user verification
// instead of flat getpwnam call.
const string testUser = "******";
if (::getpwnam(testUser.c_str()) == NULL) {
LOG(WARNING) << "Cannot run ROOT_RunTaskWithCommandInfoWithUser test:"
<< " user '" << testUser << "' is not present";
return;
}
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
// Need flags for 'executor_registration_timeout'.
slave::Flags flags = CreateSlaveFlags();
flags.isolation = "posix/cpu,posix/mem";
Try<MesosContainerizer*> containerizer =
MesosContainerizer::create(flags, false);
CHECK_SOME(containerizer);
Try<PID<Slave> > slave = StartSlave(containerizer.get());
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _))
.Times(1);
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.get().size());
// Launch a task with the command executor.
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
CommandInfo command;
command.set_value("test `whoami` = " + testUser);
command.set_user(testUser);
task.mutable_command()->MergeFrom(command);
vector<TaskInfo> tasks;
tasks.push_back(task);
Future<TaskStatus> statusRunning;
Future<TaskStatus> statusFinished;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&statusRunning))
.WillOnce(FutureArg<1>(&statusFinished));
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(statusRunning);
EXPECT_EQ(TASK_RUNNING, statusRunning.get().state());
AWAIT_READY(statusFinished);
EXPECT_EQ(TASK_FINISHED, statusFinished.get().state());
driver.stop();
driver.join();
Shutdown(); // Must shutdown before 'containerizer' gets deallocated.
}
TYPED_TEST(IsolatorTest, Usage)
{
Try<PID<Master> > master = this->StartMaster();
ASSERT_SOME(master);
TypeParam isolator;
slave::Flags flags = this->CreateSlaveFlags();
Try<PID<Slave> > slave = this->StartSlave(&isolator, flags);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), 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.get().size());
TaskInfo task;
task.set_name("isolator_test");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
Resources resources(offers.get()[0].resources());
Option<Bytes> mem = resources.mem();
ASSERT_SOME(mem);
Option<double> cpus = resources.cpus();
ASSERT_SOME(cpus);
const std::string& file = path::join(flags.work_dir, "ready");
// This task induces user/system load in a child process by
// running top in a child process for ten seconds.
task.mutable_command()->set_value(
#ifdef __APPLE__
// Use logging mode with 30,000 samples with no interval.
"top -l 30000 -s 0 2>&1 > /dev/null & "
#else
// Batch mode, with 30,000 samples with no interval.
"top -b -d 0 -n 30000 2>&1 > /dev/null & "
#endif
"touch " + file + "; " // Signals that the top command is running.
"sleep 60");
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);
EXPECT_EQ(TASK_RUNNING, status.get().state());
// Wait for the task to begin inducing cpu time.
while (!os::exists(file));
ExecutorID executorId;
executorId.set_value(task.task_id().value());
// We'll wait up to 10 seconds for the child process to induce
// 1/8 of a second of user and system cpu time in total.
// TODO(bmahler): Also induce rss memory consumption, by re-using
// the balloon framework.
ResourceStatistics statistics;
Duration waited = Duration::zero();
do {
Future<ResourceStatistics> usage =
process::dispatch(
(Isolator*) &isolator, // TODO(benh): Fix after reaper changes.
&Isolator::usage,
frameworkId.get(),
executorId);
AWAIT_READY(usage);
statistics = usage.get();
// If we meet our usage expectations, we're done!
if (statistics.cpus_user_time_secs() >= 0.125 &&
statistics.cpus_system_time_secs() >= 0.125 &&
statistics.mem_rss_bytes() >= 1024u) {
break;
}
os::sleep(Milliseconds(100));
waited += Milliseconds(100);
} while (waited < Seconds(10));
EXPECT_GE(statistics.cpus_user_time_secs(), 0.125);
EXPECT_GE(statistics.cpus_system_time_secs(), 0.125);
EXPECT_EQ(statistics.cpus_limit(), cpus.get());
EXPECT_GE(statistics.mem_rss_bytes(), 1024u);
EXPECT_EQ(statistics.mem_limit_bytes(), mem.get().bytes());
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&status));
driver.killTask(task.task_id());
AWAIT_READY(status);
EXPECT_EQ(TASK_KILLED, status.get().state());
driver.stop();
driver.join();
this->Shutdown(); // Must shutdown before 'isolator' gets deallocated.
}
TEST_F(SlaveTest, ShutdownUnregisteredExecutor)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
// Need flags for 'executor_registration_timeout'.
slave::Flags flags = CreateSlaveFlags();
// Set the isolation flag so we know a MesoContainerizer will be created.
flags.isolation = "posix/cpu,posix/mem";
Try<MesosContainerizer*> containerizer =
MesosContainerizer::create(flags, false);
CHECK_SOME(containerizer);
Try<PID<Slave> > slave = StartSlave(containerizer.get());
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _))
.Times(1);
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.get().size());
// Launch a task with the command executor.
TaskInfo task;
task.set_name("");
task.mutable_task_id()->set_value("1");
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
CommandInfo command;
command.set_value("sleep 10");
task.mutable_command()->MergeFrom(command);
vector<TaskInfo> tasks;
tasks.push_back(task);
// Drop the registration message from the executor to the slave.
Future<process::Message> registerExecutor =
DROP_MESSAGE(Eq(RegisterExecutorMessage().GetTypeName()), _, _);
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(registerExecutor);
Clock::pause();
Future<TaskStatus> status;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&status));
// Ensure that the slave times out and kills the executor.
Future<Nothing> destroyExecutor =
FUTURE_DISPATCH(_, &MesosContainerizerProcess::destroy);
Clock::advance(flags.executor_registration_timeout);
AWAIT_READY(destroyExecutor);
Clock::settle(); // Wait for Containerizer::destroy to complete.
// Now advance time until the reaper reaps the executor.
while (status.isPending()) {
Clock::advance(Seconds(1));
Clock::settle();
}
AWAIT_READY(status);
ASSERT_EQ(TASK_FAILED, status.get().state());
Clock::resume();
driver.stop();
driver.join();
Shutdown(); // Must shutdown before 'containerizer' gets deallocated.
}
// The purpose of this test is to ensure that when slaves are removed
// from the master, and then attempt to re-register, we deny the
// re-registration by sending a ShutdownMessage to the slave.
// Why? Because during a network partition, the master will remove a
// partitioned slave, thus sending its tasks to LOST. At this point,
// when the partition is removed, the slave will attempt to
// re-register with its running tasks. We've already notified
// frameworks that these tasks were LOST, so we have to have the slave
// slave shut down.
TEST_F(PartitionTest, PartitionedSlaveReregistration)
{
Try<PID<Master> > master = StartMaster();
ASSERT_SOME(master);
// Allow the master to PING the slave, but drop all PONG messages
// from the slave. Note that we don't match on the master / slave
// PIDs because it's actually the SlaveObserver Process that sends
// the pings.
Future<Message> ping = FUTURE_MESSAGE(Eq("PING"), _, _);
DROP_MESSAGES(Eq("PONG"), _, _);
MockExecutor exec(DEFAULT_EXECUTOR_ID);
StandaloneMasterDetector detector(master.get());
Try<PID<Slave> > slave = StartSlave(&exec, &detector);
ASSERT_SOME(slave);
MockScheduler sched;
MesosSchedulerDriver driver(
&sched, DEFAULT_FRAMEWORK_INFO, master.get(), DEFAULT_CREDENTIAL);
EXPECT_CALL(sched, registered(&driver, _, _));
Future<vector<Offer> > offers;
EXPECT_CALL(sched, resourceOffers(&driver, _))
.WillOnce(FutureArg<1>(&offers))
.WillRepeatedly(Return());
driver.start();
AWAIT_READY(offers);
ASSERT_NE(0u, offers.get().size());
// Launch a task. This is to ensure the task is killed by the slave,
// during shutdown.
TaskID taskId;
taskId.set_value("1");
TaskInfo task;
task.set_name("");
task.mutable_task_id()->MergeFrom(taskId);
task.mutable_slave_id()->MergeFrom(offers.get()[0].slave_id());
task.mutable_resources()->MergeFrom(offers.get()[0].resources());
task.mutable_executor()->MergeFrom(DEFAULT_EXECUTOR_INFO);
task.mutable_executor()->mutable_command()->set_value("sleep 60");
vector<TaskInfo> tasks;
tasks.push_back(task);
// Set up the expectations for launching the task.
EXPECT_CALL(exec, registered(_, _, _, _));
EXPECT_CALL(exec, launchTask(_, _))
.WillOnce(SendStatusUpdateFromTask(TASK_RUNNING));
Future<TaskStatus> runningStatus;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&runningStatus));
Future<Nothing> statusUpdateAck = FUTURE_DISPATCH(
slave.get(), &Slave::_statusUpdateAcknowledgement);
driver.launchTasks(offers.get()[0].id(), tasks);
AWAIT_READY(runningStatus);
EXPECT_EQ(TASK_RUNNING, runningStatus.get().state());
// Wait for the slave to have handled the acknowledgment prior
// to pausing the clock.
AWAIT_READY(statusUpdateAck);
// Drop the first shutdown message from the master (simulated
// partition), allow the second shutdown message to pass when
// the slave re-registers.
Future<ShutdownMessage> shutdownMessage =
DROP_PROTOBUF(ShutdownMessage(), _, slave.get());
Future<TaskStatus> lostStatus;
EXPECT_CALL(sched, statusUpdate(&driver, _))
.WillOnce(FutureArg<1>(&lostStatus));
Future<Nothing> slaveLost;
EXPECT_CALL(sched, slaveLost(&driver, _))
.WillOnce(FutureSatisfy(&slaveLost));
Clock::pause();
// Now, induce a partition of the slave by having the master
// timeout the slave.
uint32_t pings = 0;
while (true) {
AWAIT_READY(ping);
pings++;
if (pings == master::MAX_SLAVE_PING_TIMEOUTS) {
break;
}
ping = FUTURE_MESSAGE(Eq("PING"), _, _);
Clock::advance(master::SLAVE_PING_TIMEOUT);
Clock::settle();
}
Clock::advance(master::SLAVE_PING_TIMEOUT);
Clock::settle();
// The master will have notified the framework of the lost task.
AWAIT_READY(lostStatus);
EXPECT_EQ(TASK_LOST, lostStatus.get().state());
// Wait for the master to attempt to shut down the slave.
AWAIT_READY(shutdownMessage);
// The master will notify the framework that the slave was lost.
AWAIT_READY(slaveLost);
Clock::resume();
// We now complete the partition on the slave side as well. This
// is done by simulating a master loss event which would normally
// occur during a network partition.
detector.appoint(None());
Future<Nothing> shutdown;
EXPECT_CALL(exec, shutdown(_))
.WillOnce(FutureSatisfy(&shutdown));
shutdownMessage = FUTURE_PROTOBUF(ShutdownMessage(), _, slave.get());
// Have the slave re-register with the master.
detector.appoint(master.get());
// Upon re-registration, the master will shutdown the slave.
// The slave will then shut down the executor.
AWAIT_READY(shutdownMessage);
AWAIT_READY(shutdown);
driver.stop();
driver.join();
Shutdown();
}