[
https://issues.apache.org/jira/browse/FLINK-14330?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16945280#comment-16945280
]
Zhu Zhu edited comment on FLINK-14330 at 10/7/19 6:40 AM:
----------------------------------------------------------
[~trohrmann] [~GJL] I think it's worthy doing it in Flink 1.10.
One reason is that it's a prerequisite to avoid code duplication when
supporting building regions from JobGraph in FLINK-14312.
Another reason is that the earlier we do it, the easier it can be. Future
introduced dependencies on those topologies will make the change even harder.
Especially after the pluggable scheduling strategy feature is released.
Here's the [detailed
doc|https://docs.google.com/document/d/1f88luAOfUQ6Pm4JkxYexLXpfH-crcXJdbubi1pS2Y5A/edit?usp=sharing]
for the motivation and design. A
[POC|https://github.com/zhuzhurk/flink/commits/poc_unified_topology_interfaces]
is also done for this change.
What's your opinion?
was (Author: zhuzh):
[~trohrmann] [~GJL] I think it's worthy doing it in Flink 1.10.
One reason is that it's a prerequisite to avoid code duplication when
supporting building regions from JobGraph in FLINK-14312.
Another reason is that the earlier we do it, the easier it can be. Future
introduced dependencies on those topologies will make the change even harder.
Here's the [detailed
doc|https://docs.google.com/document/d/1f88luAOfUQ6Pm4JkxYexLXpfH-crcXJdbubi1pS2Y5A/edit?usp=sharing]
for the motivation and design. A
[POC|https://github.com/zhuzhurk/flink/commits/poc_unified_topology_interfaces]
is also done for this change.
What's your opinion?
> Introduce a unified topology interface
> --------------------------------------
>
> Key: FLINK-14330
> URL: https://issues.apache.org/jira/browse/FLINK-14330
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Affects Versions: 1.10.0
> Reporter: Zhu Zhu
> Priority: Major
> Fix For: 1.10.0
>
>
> When working on FLINK-14312 to partition {{JobGraph}} into logical pipelined
> regions, I found that we can hardly reuse the existing util
> {{PipelinedRegionComputeUtil#computePipelinedRegions(..)}} to do it since
> it's based on the {{FailoverTopology}}.
> To avoid code duplication, we need a unified topology base for
> {{FailoverTopology}} and {{JobGraph/LogicalTopology}}.
> Besides that, the inconsistency of {{FailoverTopology}} and
> {{SchedulingTopology}} is also causing troubles for development and
> performance.
> That's why I'd propose to unify the interfaces all these topologies. More
> details can be found in this [design
> doc|https://docs.google.com/document/d/1f88luAOfUQ6Pm4JkxYexLXpfH-crcXJdbubi1pS2Y5A/edit?usp=sharing].
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
