[
https://issues.apache.org/jira/browse/FLINK-16903?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17073322#comment-17073322
]
Jingsong Lee edited comment on FLINK-16903 at 4/2/20, 2:39 AM:
---------------------------------------------------------------
There are these ways:
# Planner can read {{sink.parallelism}} property from TableConfig, it will be
global.
# Planner can obtains table properties, so it can read this
{{sink.parallelism}} property.
# Have a special {{SupportsCustomParallelism}} ability, but also we have
{{sink.parallelism}} property in connector to let users have the ability to
specify.
Maybe the option 3 is the best one
* first, It looks more natural than reading a specific key.
* second, it can support custom parallelism inference. The property can not do
this.
What do you think?
was (Author: lzljs3620320):
There are these ways:
# Planner can read {{sink.parallelism}} property from TableConfig, it will be
global.
# Planner can obtains table properties, so it can read this
{{sink.parallelism}} property.
# Have a special {{SupportsCustomParallelism}} ability
Maybe the option 3 is the best one
* first, It looks more natural than reading a specific key.
* second, it can support custom parallelism inference. The property can not do
this.
What do you think?
> Add sink.parallelism for file system factory
> --------------------------------------------
>
> Key: FLINK-16903
> URL: https://issues.apache.org/jira/browse/FLINK-16903
> Project: Flink
> Issue Type: Sub-task
> Components: Table SQL / Runtime
> Reporter: Jingsong Lee
> Priority: Major
> Fix For: 1.11.0
>
>
> A single task may be writing multiple files at the same time. If the
> parallelism is too high, it may lead to a large number of small files. If the
> parallelism is too small, the performance is not enough. This requires that
> the user can specify parallelism.
> * Default is the same as upstream transformation
> * Users can specify parallelism too.
> |‘connector.sink.parallelism’ = ...|
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
