aurora-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Renan DelValle (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (AURORA-1734) Configurable Metadata prefix
Date Fri, 19 Aug 2016 01:07:20 GMT

    [ https://issues.apache.org/jira/browse/AURORA-1734?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15427421#comment-15427421
] 

Renan DelValle edited comment on AURORA-1734 at 8/19/16 1:06 AM:
-----------------------------------------------------------------

So I've been thinking about this quite a bit. And it feels to me that we're jumping the gun
a bit on a future "what-if" regarding the name collision.

As an alternative, I'd like to propose that we reserve the `org.apache.aurora` namespace for
future use and remove the prefix altogether (we can even go as far as rejecting a task that
includes a label key with this prefix). I'm interested in hearing what everyone's opinion
on this would be.

The reason I've come to be against every label key having the prefix is that we, the docker
compose executor devs, pass labels to our containers. As such, if lables have the prefix,
we have to create a special "aurora editon" of the docker compose executor to filter out the
prefix. (This will be true of any future Mesos executor that wants to make use of labels as
well).

Configuring the metadata from the scheduler is an acceptable solution as well, however, it's
less flexible for executor devs. For example, if the community decides to make `environment`
or `role` in the future, we would have to filter these out in the executor on a case by case
basis (blacklist), instead of a  filter for any task beginning with the prefix `org.apache.aurora`.
As a bonus, the patch to do this is less messy :).

Would like to know the community's thoughts on this before moving forward with the patch.


was (Author: rdelvalle):
So I've been thinking about this quite a bit. And it feels to me that we're jumping the gun
a bit on a future "what-if" regarding the name collision.

As an alternative, I'd like to propose that we reserve the `org.apache.aurora` namespace for
future use and remove the prefix altogether (we can even go as far as rejecting a task that
includes a label key with this prefix). I'm interested in hearing what everyone's opinion
on this would be.

The reason I've come to be against every label key having the prefix is that we need to pass
labels to our containers with the compose executor. As such, if we have the prefix, we have
to create a special "aurora ediiton" of the docker compose executor to filter out the prefix.
(This will be true of any future Mesos executor that wants to make use of labels as well).

Configuring the metadata from the scheduler is an acceptable solution as well, however, it's
less flexible for executor devs. For example, if the community decides to make `environment`
or `role` in the future, we would have to filter these out in the executor on a case by case
basis (blacklist), instead of a  filter for any task beginning with the prefix `org.apache.aurora`.
As a bonus, the patch to do this is less messy :).

Would like to know the community's thoughts on this before moving forward with the patch.

> Configurable Metadata prefix
> ----------------------------
>
>                 Key: AURORA-1734
>                 URL: https://issues.apache.org/jira/browse/AURORA-1734
>             Project: Aurora
>          Issue Type: Task
>          Components: Scheduler
>            Reporter: Renan DelValle
>            Assignee: Renan DelValle
>            Priority: Trivial
>
> Currently, a prefix ("org.apache.aurora.metadata.") is injected into the metadata key
in the scheduler. It would be beneficial to allow users to set their own metadata prefix (including
an empty string).



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message