In case you missed it, we need to update the scheduler to notice the
TASK_ERROR state. I've filed
https://issues.apache.org/jira/browse/AURORA-1001 to track the work for
this on our end.
---------- Forwarded message ----------
From: Adam Bordelon <a...@mesosphere.io>
Date: Wed, Jan 7, 2015 at 1:59 PM
Subject: Re: TaskStatus source and reason fields
To: "u...@mesos.apache.org" <u...@mesos.apache.org>, dev <
d...@mesos.apache.org>
FYI, Mesos wasn't actually considering TASK_ERROR a terminal state until
now (0.22). No impact if you weren't using it yet.
commit 1c80d845431a57dd8c20e636ab7fc313602e4b49
Author: Connor Doyle <con...@mesosphere.io>
Date: Wed Jan 7 13:33:16 2015 -0800
TASK_ERROR is terminal
TASK_ERROR was introduced as a new terminal state, but was not handled
as such in mesos::internal::protobuf::isTerminalState.
Review: https://reviews.apache.org/r/29610
On Mon, Nov 10, 2014 at 2:15 PM, Dominic Hamon <dha...@twopensource.com>
wrote:
> I have a patch ready to land for MESOS-1143
> <https://issues.apache.org/jira/browse/MESOS-1143>. This adds TASK_ERROR
> to the possible states that a framework might receive as a status update.
> The semantics of TASK_LOST vs TASK_ERROR are simple: TASK_LOST means that
> attempting to reschedule the task should succeed. TASK_ERROR means that
any
> attempt to reschedule the task will fail. This allows frameworks to make
> better decisions.
>
> Before it lands, I'd like to solicit feedback as this changes the
> semantics for frameworks (for the better!).
>
> Any thoughts or reservations?
>
> --
> Dominic Hamon | @mrdo | Twitter
> *There are no bad ideas; only good ideas that go horribly wrong.*
>
