[MESOS-380] Command Executor doesn't send TASK_KILLED for killed tasks.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.14.0
• Assignee: Vinod Kone
• Reporter: Benjamin Mahler
• Created: 07/Mar/13 00:27
• Updated: 29/Jul/13 21:23
• Resolved: 29/Jul/13 21:23

Description

Currently the command executor will send either TASK_FAILED or TASK_FINISHED as a result of a killTask.

The fix is non-trivial since it requires sharing some state with the waiter thread:

static void waiter(pid_t pid, const TaskID& taskId, ExecutorDriver* driver)
{
int status;
while (wait(&status) != pid || WIFSTOPPED(status));

CHECK(WIFEXITED(status) || WIFSIGNALED(status));

std::cout << "Waited on process " << pid
<< ", returned status " << status << std::endl;

TaskStatus taskStatus;
taskStatus.mutable_task_id()->MergeFrom(taskId);

// THIS BIT NEEDS TO KNOW WHETHER THE TASK WAS KILLED!
if (WIFEXITED(status) && WEXITSTATUS(status) == 0)

{ taskStatus.set_state(TASK_FINISHED); }

else

{ taskStatus.set_state(TASK_FAILED); }

Try<string> message = WIFEXITED(status)
? strings::format("Command exited with status %d", WEXITSTATUS(status))
: strings::format("Command terminated with signal '%s'",
strsignal(WTERMSIG(status)));

if (message.isSome())

{ taskStatus.set_message(message.get()); }

driver->sendStatusUpdate(taskStatus);

// A hack for now ... but we need to wait until for the status
// update to get sent to the slave before we shut ourselves down.
sleep(1);
driver->stop();
}