Jie Yu created MESOS-1417:
-----------------------------
Summary: Slave should not send terminal status update before
containerizer update is finished
Key: MESOS-1417
URL: https://issues.apache.org/jira/browse/MESOS-1417
Project: Mesos
Issue Type: Bug
Reporter: Jie Yu
Priority: Critical
This starts to become a problem when we introduce the network isolator. Say a
port P is initially allocated to task A. When task A finishes, if we don't wait
for containerizer update to finish and send TASK_FINISHED to the master, port P
might be allocated to some other task B on the same host before it is even
freed!
{noformat}
void Slave::statusUpdate(const StatusUpdate& update, const UPID& pid)
{
...
if (protobuf::isTerminalState(status.state()) &&
(executor->queuedTasks.contains(status.task_id()) ||
executor->launchedTasks.contains(status.task_id()))) {
executor->terminateTask(status.task_id(), status.state());
// Tell the isolator to update the resources.
// TODO(idownes): Wait until this completes.
CHECK_SOME(executor->resources);
containerizer->update(executor->containerId, executor->resources.get());
}
if (executor->checkpoint) {
// Ask the status update manager to checkpoint and reliably send the update.
statusUpdateManager->update(
update,
info.id(),
executor->id,
executor->containerId)
.onAny(defer(self(),
&Slave::_statusUpdate,
lambda::_1,
update,
pid));
} else {
...
}
}
{noformat}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
