[
https://issues.apache.org/jira/browse/MESOS-4567?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15125224#comment-15125224
]
Vinod Kone commented on MESOS-4567:
-----------------------------------
{quote}
Further, looks like no where in the source code, we are setting the
TASK_STARTING state.
{quote}
This is not surprising considering this is expected to be set by the executor.
To provide some context, TASK_STARTING was designed to be used by a custom
executor to convey to the scheduler that it has received the task launch
command. This is because it might take a non-trivial amount of time before an
executor can actually run the task (e.g., it might need to fetch the packages
for a task from a remote data store) and hence send TASK_RUNNING.
At least one framework I know of (Apache Aurora) uses this task state as a hand
off between scheduler and executor. There might well be other frameworks that
depend/utilize this state.
So I don't think we have a good reason to deprecate this state.
> Deprecate TASK_STARTING state
> -----------------------------
>
> Key: MESOS-4567
> URL: https://issues.apache.org/jira/browse/MESOS-4567
> Project: Mesos
> Issue Type: Task
> Reporter: Kapil Arya
> Labels: mesosphere
>
> We currently have the following task stages:
> * TASK_STAGING -> set by slave
> * TASK_STARTING -> set by the executor (?)
> * TASK_RUNNING -> set by the executor when the task is running
> * TASK_XXX -> task termination statuses
> The confusion here is about TASK_STARTING. This is the state between
> TASK_STAGING and TASK_RUNNING and is somewhat non-intuitive for the reader.
> Further, looks like no where in the source code, we are setting the
> TASK_STARTING state.
> Why shouldn't we just deprecate/remove it?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
