[
https://issues.apache.org/jira/browse/FLINK-32255?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17730013#comment-17730013
]
Christophe Bornet commented on FLINK-32255:
-------------------------------------------
Sure. I'll open the discussion.
I don't know if the semantics should be changed since when it's used for fast
recovery and not for fast startup, it makes sense to not start TMs when there
are no running jobs.
Anyway, let's discuss on the ML.
> Pre-deployed idle Task Managers
> -------------------------------
>
> Key: FLINK-32255
> URL: https://issues.apache.org/jira/browse/FLINK-32255
> Project: Flink
> Issue Type: New Feature
> Components: Kubernetes Operator
> Reporter: Christophe Bornet
> Priority: Major
>
> In the kubernetes operator, it would be nice to have the possibility to
> define a number of pre-deployed spare Task Managers. The operator would
> ensure that there are always a number of ready-to-be-used TMs that can accept
> new incoming jobs.
> The goal would be to reduce the startup time of jobs, especially in the
> session cluster mode.
> See for inspiration
> [https://www.decodable.co/blog/flink-deployments-at-decodable] -> Custom
> Flink Session Clusters for Preview Jobs: "To further reduce the preview
> response time, we also modified Flink to pre-deploy some idle TaskManagers.
> When a TaskManager is used, a new TaskManager is immediately created so there
> is always a pool of idle TaskManagers."
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
