[
https://issues.apache.org/jira/browse/FLINK-15470?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yang Wang updated FLINK-15470:
------------------------------
Description:
Currently, when we start a session, flink client will write a yarn properties
under \{java.io.tmpdir}/.yarn-properties-\{user-name}. So the user could use a
simple command {{./bin/flink run ./examples/batch/WordCount.jar}} to submit a
job to the session.
Even if the session cluster has finished or kill by {{yarn application -kill}},
it will still submit a job to the non-existing session. Some users has run into
this problem in the ML. So i suggest to remove the properties file and always
use {{-yid application_id}} to submit job an existing session.
[https://lists.apache.org/thread.html/b0e0e388a16d1891d1de651bec027a3820130f28aaaeb843440cb679%40%3Cuser.flink.apache.org%3E]
We also have some discussion under FLINK-15179. The cli options will be
deprecated in the future and we hope {{FlinkYarnSessionCli}} could use
{{ExecutorCli}}. So we will try to clean up the code, i think the magic
properties file is the case.
cc [~tison] [~kkl0u]
was:
Currently, when we start a session, flink client will write a yarn properties
under \{java.io.tmpdir}/.yarn-properties-\{user-name}. So the user could use a
simple command {{./bin/flink run ./examples/batch/WordCount.jar}} to submit a
job to the session.
Even if the session cluster has finished or kill by {{yarn application -kill}},
it will still submit a job to the non-existing session. Some users has run into
this problem in the ML. So i suggest to remove the properties file and always
use {{-yid application_id}} to submit job an existing session.
[https://lists.apache.org/thread.html/b0e0e388a16d1891d1de651bec027a3820130f28aaaeb843440cb679%40%3Cuser.flink.apache.org%3E]
We also have some discussion under
[FLINK-15179|https://issues.apache.org/jira/browse/FLINK-15179]. The cli
options will be deprecated in the future and we hope {{FlinkYarnSessionCli}}
could use {{ExecutorCli}}. So we will try to clean up the code, i think the
magic properties file is the case.
> Remove YARN magic properties file
> ---------------------------------
>
> Key: FLINK-15470
> URL: https://issues.apache.org/jira/browse/FLINK-15470
> Project: Flink
> Issue Type: Task
> Components: Command Line Client, Deployment / YARN
> Reporter: Yang Wang
> Priority: Minor
>
> Currently, when we start a session, flink client will write a yarn properties
> under \{java.io.tmpdir}/.yarn-properties-\{user-name}. So the user could use
> a simple command {{./bin/flink run ./examples/batch/WordCount.jar}} to submit
> a job to the session.
>
> Even if the session cluster has finished or kill by {{yarn application
> -kill}}, it will still submit a job to the non-existing session. Some users
> has run into this problem in the ML. So i suggest to remove the properties
> file and always use {{-yid application_id}} to submit job an existing session.
> [https://lists.apache.org/thread.html/b0e0e388a16d1891d1de651bec027a3820130f28aaaeb843440cb679%40%3Cuser.flink.apache.org%3E]
>
> We also have some discussion under FLINK-15179. The cli options will be
> deprecated in the future and we hope {{FlinkYarnSessionCli}} could use
> {{ExecutorCli}}. So we will try to clean up the code, i think the magic
> properties file is the case.
>
> cc [~tison] [~kkl0u]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
