[
https://issues.apache.org/jira/browse/BEAM-1393?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15856294#comment-15856294
]
Aljoscha Krettek commented on BEAM-1393:
----------------------------------------
Yes, there are definitely some changes that need to be done while moving to
Flink 1.2.
1. This should be covered by BEAM-1394.
2. I think we can remove the checks and use {{SourceFunction.SourceContext}}
instead.
3. Definitely true, this will get us rescalability.
4. I think it should be {{GlobalConfiguration.loadConfiguration()}}, AFAIK the
dynamic properties is only used in conjunction with Mesos code so far
5. true
6. I think we have to get a bit creative here and only call super when we know
that the watermark should advance. Casting doesn't seem wise since we won't
always be guaranteed that the timer service is in fact a
{{HeapInternalTimerService}}.
7. What BROADCAST does is collect all checkpointed states into one "list" and
then send out that list to all parallel subtasks when restoring. The way we
would use it is to only checkpoint anything from the operator with subtask
index 0 because we assume that the state is the same on all parallel instances
of the operator.
That last part is still a bit vague, I think we can tackle that once the other
parts are figured out.
> Update Flink Runner to Flink 1.2.0
> ----------------------------------
>
> Key: BEAM-1393
> URL: https://issues.apache.org/jira/browse/BEAM-1393
> Project: Beam
> Issue Type: Improvement
> Components: runner-flink
> Reporter: Aljoscha Krettek
> Assignee: Jingsong Lee
>
> When we update to 1.2.0 we can use the new internal Timer API that is
> available to Flink operators: {{InternalTimerService}} and also use broadcast
> state to store side-input data.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
