[
https://issues.apache.org/jira/browse/FLINK-23425?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17387129#comment-17387129
]
Yao Zhang commented on FLINK-23425:
-----------------------------------
Hi [~tartarus],
The problem here is the default parallelism of the execution environment is
determined by the number of available processors, which will vary in different
test environments.
However, one of the conditions whether two StreamEdge is chainable or not is
the parallelisms of upstream vertex and the downstream vertex. if they can be
chainable, the parallelisms of upstream vertex and the downstream vertex must
be equal. Whether the source and map vertex can be chainable affects the number
of job vertex.
The parallelism of fromElements is always 1 while the counterpart of map
function is configurable(default is the number of available processors).
I will fix it by assigning the parallelism explicitly. Could you please assign
this to me?
> The impact of cpu cores on test results for
> StreamingJobGraphGeneratorTest#testSlotSharingResourceConfigurationWithDefaultSlotSharingGroup
>
> -------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-23425
> URL: https://issues.apache.org/jira/browse/FLINK-23425
> Project: Flink
> Issue Type: Bug
> Components: Tests
> Reporter: tartarus
> Priority: Major
>
> By default, LocalStreamEnvironment will use the number of cpu
> cores(Runtime.getRuntime().availableProcessors()) as the default parallelism;
> In our company, we will use a 1-core docker container to run the test.
> Then
> StreamingJobGraphGeneratorTest#testSlotSharingResourceConfigurationWithDefaultSlotSharingGroup
> failed.
> because the parallelism of map operator is 1, so the number of Vertex is 1,
> not expected 2 anymore.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
