[
https://issues.apache.org/jira/browse/FLINK-17012?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17085785#comment-17085785
]
Stephan Ewen commented on FLINK-17012:
--------------------------------------
I am against an "initialize" phase. There is a lot of existing complexity
because the tasks don't eagerly initialize in the constructor, but at a later
stage.
That means after invokable instance is created, it is incomplete. That leads to
a bunch or issues, for example can the instance already by "cancel()"-ed, which
needs to be extra careful because of incomplete initialization. Or checkpoints
see a task but need to take extra care that the task in not actually fully
initialized. Lot's of extra status checking, easy to forget, it was a frequent
source of bugs over the years.
I would suggest to move more towards "eager initialization", rather than
hard-wiring the delayed initialization with an "initialize()" method.
As a general thought: My take is that almost always (with very very few
exceptions), "initialize()" or "configure()" methods should be seen as
anti-patterns and be avoided if possible.
> Expose stage of task initialization
> -----------------------------------
>
> Key: FLINK-17012
> URL: https://issues.apache.org/jira/browse/FLINK-17012
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Metrics, Runtime / Task
> Reporter: Wenlong Lyu
> Priority: Major
>
> Currently a task switches to running before fully initialized, does not take
> state initialization and operator initialization(#open ) in to account, which
> may take long time to finish. As a result, there would be a weird phenomenon
> that all tasks are running but throughput is 0.
> I think it could be good if we can expose the initialization stage of tasks.
> What to you think?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
