You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@yunikorn.apache.org by "Craig Condit (Jira)" <ji...@apache.org> on 2024/02/20 15:44:00 UTC

[jira] [Comment Edited] (YUNIKORN-1351) Define policy for when to use annotations vs. labels

    [ https://issues.apache.org/jira/browse/YUNIKORN-1351?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17818872#comment-17818872 ] 

Craig Condit edited comment on YUNIKORN-1351 at 2/20/24 3:43 PM:
-----------------------------------------------------------------

Some thoughts on the first design doc rev:

1) Everything we define moving forward should have a namespace ({*}{{yunikorn.apache.org}}{*} for public properties; *{{internal.yunikorn.apache.org}}* for internal ones). We should also standardize on hyphenated lowercase identifiers for consistency (we currently have a mix of camelCase, dot.case and hypen-case identifiers). *However* we should not rename existing already-namespaced identifiers, as this will just cause further confusion.

2) We need to be careful about the "deprecated" status – we have lots of third-party components that use the older forms; it could be a very long time (if ever) before these get updated. It would be better to define a "canonical" representation (i.e. the one that should be used moving forward).

3) In the case where we have multiple representations moving forward, we should define a policy that any of them are acceptable; however if more than one is specified, their values *must* be the same. It should be considered invalid (for example) to have two different application IDs defined using two different labels or annotations.

4) We should NOT complicate matters by changing annotations on namespaces, only on pods.

 

As for the some of the values themselves, here is what I would propose as the canonical representations:
 * Application ID
 ** Currently in use: applicationId (label), yunikorn.apache.org/app-id (annotation)
 ** Proposed canonical representation: *yunikorn.apache.org/app-id (label)*
 ** Why: The namespaced value is good, however application ID is often used as a selector.
 * Queue
 ** Currently in use: queue (label), yunikorn.apache.org/queue (annotation)
 ** Proposed canonical representation: *yunikorn.apache.org/queue (label)*
 ** Why: The namespaced value is good, however selecting applications by queue is useful.

 


was (Author: ccondit):
Some thoughts on the first design doc rev:

1) Everything we define moving forward should have a namespace ({*}{{yunikorn.apache.org}}{*} for public properties; *{{internal.yunikorn.apache.org}}* for internal ones). We should also standardize on hyphenated lowercase identifiers for consistency (we currently have a mix of camelCase, dot.case and hypen-case identifiers).

2) We need to be careful about the "deprecated" status – we have lots of third-party components that use the older forms; it could be a very long time (if ever) before these get updated. It would be better to define a "canonical" representation (i.e. the one that should be used moving forward).

3) In the case where we have multiple representations moving forward, we should define a policy that any of them are acceptable; however if more than one is specified, their values *must* be the same. It should be considered invalid (for example) to have two different application IDs defined using two different labels or annotations.

4) We should NOT complicate matters by changing annotations on namespaces, only on pods.

 

As for the some of the values themselves, here is what I would propose as the canonical representations:
 * Application ID
 ** Currently in use: applicationId (label), yunikorn.apache.org/app-id (annotation)
 ** Proposed canonical representation: *yunikorn.apache.org/app-id (label)*
 ** Why: The namespaced value is good, however application ID is often used as a selector.
 * Queue
 ** Currently in use: queue (label), yunikorn.apache.org/queue (annotation)
 ** Proposed canonical representation: *yunikorn.apache.org/queue (label)*
 ** Why: The namespaced value is good, however selecting applications by queue is useful.

 

> Define policy for when to use annotations vs. labels
> ----------------------------------------------------
>
>                 Key: YUNIKORN-1351
>                 URL: https://issues.apache.org/jira/browse/YUNIKORN-1351
>             Project: Apache YuniKorn
>          Issue Type: Improvement
>          Components: shim - kubernetes
>            Reporter: Wilfred Spiegelenburg
>            Assignee: Yu-Lin Chen
>            Priority: Major
>              Labels: newbie, pull-request-available
>         Attachments: [Design Doc] Define Usage of Labels and Annotations in YuniKorn.pdf
>
>
> Currently, we have very inconsistent use of labels vs. annotations in YuniKorn. Additionally, some values are namespaced under yunikorn.apache.org while some are not. We will likely need to keep legacy values around for backwards compatibility but we should define a strategy for defining canonical representations of this metadata as well as a developer-oriented pollcy around when to use labels vs. annotations.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@yunikorn.apache.org
For additional commands, e-mail: issues-help@yunikorn.apache.org