[ 
https://issues.apache.org/jira/browse/FLINK-15179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16994623#comment-16994623
 ] 

Kostas Kloudas commented on FLINK-15179:
----------------------------------------

Hi [~fly_in_gis], 

To start a session cluster, for now each deployment target has its own CLI 
(yarnSessionCLI and KubernetesSessionCLI). In the future, it may make sense to 
unify them as well, but this cannot be done for 1.10 so we can chat on that for 
the next release. Actually, I think that it could be interesting to see if we 
can unify them.

For the flink run -h, there are some options that are common for all deployment 
targets, and some that are specific (e.g. in per-job mode). Given that we need 
to include both in a nice help message, it is inevitable that part of the 
message will be executor-specific (picked up by the specified -e option-no need 
for a prefix). 

One thing that it is worth discussing is if we want to keep the "magic" of the 
"YARN properties file", where a hidden file is written when creating a cluster, 
so that future calls to {{flink run}} can pick up the cluster id. But in any 
case, I think that partially the help message has to be executor-specific.

> Kubernetes should not have a CustomCommandLine.
> -----------------------------------------------
>
>                 Key: FLINK-15179
>                 URL: https://issues.apache.org/jira/browse/FLINK-15179
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Deployment / Kubernetes
>    Affects Versions: 1.10.0
>            Reporter: Kostas Kloudas
>            Assignee: Kostas Kloudas
>            Priority: Major
>              Labels: pull-request-available
>          Time Spent: 0.5h
>  Remaining Estimate: 0h
>
> As part of FLIP-73, all command line options are mapped to config options. 
> Given this 1-to-1 mapping, the Kubernetes command line could simply forward 
> the command line arguments to ConfigOptions directly, instead of introducing 
> new command line options. In this case, the user is expected to simply write:
>  
> {\{bin/run -e (or --executor) kubernetes-session-cluster -D 
> kubernetes.container.image=MY_IMAGE ...}} 
> and the CLI will parse the -e to figure out the correct 
> {{ClusterClientFactory}} and {{ExecutorFactory}} and then forward to that the 
> config options specified with {{-D}}. 
> For this, we need to introduce a {{GenericCustomCommandLine}} that simply 
> forward the specified parameters to the executors.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to