[ 
https://issues.apache.org/jira/browse/YUNIKORN-2208?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17791478#comment-17791478
 ] 

Wilfred Spiegelenburg commented on YUNIKORN-2208:
-------------------------------------------------

The "pod bind Time" is not well defined and not consistently set. It is set to 
the time of recovery not when the pod started running. We know the real start 
time of the pod as it comes from the k8shim. It really should always be the one 
and the same value.

When we use the event system do not need it at all. We should rely on the 
original event that was created when the pod was allocated. That was the time 
usage started. We should not even have bind time in code at all I think as the 
current value is ambiguous.

Finish time is the pod removal. For the restart case: any pod node etc that 
existed before the restart and no longer exists after the restart the startup 
time of the scheduler should be used by the aggregation service as the finish 
time.

Adding the instance type as a field to the message seems like a huge overkill. 
It will only be set on an event:
 * Type = node
 * ChangeType = Add
 * ChangeDetail = NONE

The node instance type is really is a reference to an external list of instance 
types... Need to think if we can integrate it like that.

> Extend the EventRecord to support aggreated resources tracking
> --------------------------------------------------------------
>
>                 Key: YUNIKORN-2208
>                 URL: https://issues.apache.org/jira/browse/YUNIKORN-2208
>             Project: Apache YuniKorn
>          Issue Type: Sub-task
>          Components: scheduler-interface
>            Reporter: Qi Zhu
>            Assignee: Qi Zhu
>            Priority: Major
>              Labels: pull-request-available
>
> Before we create a separate tracking service to consume EventRecord, we need 
> to extend the EventRecord to cover the resource aggreated support, such as:
> 1. Node instance type
> 2. pod bind Time
> 3. pod finish time, we can reuse the event sending time here, so we don't 
> need to add new field for this
> 4. etc



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to