azagrebin commented on a change in pull request #11545:
[FLINK-16742][runtime][dist] Extend and use BashJavaUtils to start JM JVM
process and pass JVM memory args
URL: https://github.com/apache/flink/pull/11545#discussion_r400749399
##########
File path:
flink-container/src/main/java/org/apache/flink/container/entrypoint/StandaloneJobClusterEntryPoint.java
##########
@@ -116,6 +107,20 @@ public static void main(String[] args) {
ClusterEntrypoint.runClusterEntrypoint(entrypoint);
}
+ public static StandaloneJobClusterConfiguration
getEntrypointClusterConfiguration(String[] args) {
+ final CommandLineParser<StandaloneJobClusterConfiguration>
commandLineParser = new CommandLineParser<>(new
StandaloneJobClusterConfigurationParserFactory());
Review comment:
The actual code to create `Configuration` is the same in all cases: TM / JM
session / JM job mode. It uses same `CONFIG_DIR_OPTION` and
`DYNAMIC_PROPERTY_OPTION`. Basically, `TaskManagerRunner#loadConfiguration`
would work for all cases atm.
I agree that different Flink processes may theoretically do more magic for
the memory options in the `Configuration` and parse it differently in future.
Not sure though how probable it is and that we want to address this now. The
reason that there are various `ParserFactories` is that they inject some
specific cmdline options but not memory related. The entry point specific
options are not even passed to the bash utils. Therefore, maybe, a simpler
solution for now is to just abstract `TaskManagerRunner#loadConfiguration` into
some util and use it for all purposes in one `BashJavaUtils`?
As I understand, introducing `StandaloneJobClusterBashJavaUtils` tries to
overcome the problem that `BashJavaUtils` cannot access `flink-container`
module from `flink-runtime`. If the previously described simple solution is not
enough for now, what about moving `BashJavaUtils` into a separate module? This
may be even a better abstraction of it as it is a separate artefact to build by
maven.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
