Brendan Chang created MESOS-2896:
------------------------------------
Summary: Can't mock queued_tasks in slave tests
Key: MESOS-2896
URL: https://issues.apache.org/jira/browse/MESOS-2896
Project: Mesos
Issue Type: Improvement
Components: slave, test
Reporter: Brendan Chang
Priority: Minor
The issue of tracking queued_tasks in an executor resulted from
[MESOS-2598|https://issues.apache.org/jira/browse/MESOS-2598].
When frameworks are run, tasks are queued and then immediately launched on the
executor, so it is difficult to isolate the slave state when tasks are still
queued (i.e. tasks are in queued_tasks[] and not tasks[] or completed_tasks[]
in state.json).
In slave.cpp:_runTask(), tasks are queued on the executor only when the
executor has not yet registered or it is currently running:
{code}
case Executor::REGISTERING:
// Checkpoint the task before we do anything else.
if (executor->checkpoint) {
executor->checkpointTask(task);
}
// Queue task if the executor has not yet registered.
LOG(INFO) << "Queuing task '" << task.task_id()
<< "' for executor " << executorId
<< " of framework '" << frameworkId;
executor->queuedTasks[task.task_id()] = task;
break;
case Executor::RUNNING: {
// Checkpoint the task before we do anything else.
if (executor->checkpoint) {
executor->checkpointTask(task);
}
// Queue task until the containerizer is updated with new
// resource limits (MESOS-998).
LOG(INFO) << "Queuing task '" << task.task_id()
<< "' for executor " << executorId
<< " of framework '" << frameworkId;
executor->queuedTasks[task.task_id()] = task;
{code}
It is difficult to catch the executor in either of these states with queued
tasks.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
