You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Xintong Song (Jira)" <ji...@apache.org> on 2021/01/15 07:03:01 UTC
[jira] [Updated] (FLINK-17372) SlotManager should expose total
required resources
[ https://issues.apache.org/jira/browse/FLINK-17372?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Xintong Song updated FLINK-17372:
---------------------------------
Parent: FLINK-20988
Issue Type: Sub-task (was: Improvement)
> SlotManager should expose total required resources
> --------------------------------------------------
>
> Key: FLINK-17372
> URL: https://issues.apache.org/jira/browse/FLINK-17372
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Affects Versions: 1.11.0
> Reporter: Till Rohrmann
> Priority: Major
> Fix For: 1.13.0
>
>
> Currently, the {{SlotManager}} exposes the set of required resources which have not been fulfilled via {{SlotManager.getRequiredResources}}. The idea of this function is to allow the {{ResourceManager}} to decide whether new pods/containers need to be started or not.
> The problem is that once a resource has been registered at the {{SlotManager}} it will decrease the set of required resources. If now a pod/container fails, then the {{ResourceManager}} won't know whether it needs to restart the container or not.
> In order to simplify the interaction, I propose to let the {{SlotManager}} announce all of its required resources (pending + registered resources). That way the {{ResourceManager}} only needs to compare the set of required resources with the set of pending and allocated containers/pods.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)