Hi Experts,
When submitting a Flink program to Yarn, the app jar( a fat jar about
200M with Flink dependencies ) will be uploaded to Yarn, which will take a lot
of time. I check the code in CliFrontend, and found that there is a config item
named “yarn.per-job-cluster.include-user-jar”, I try to set the config item
value to “DISABLED”, the fat jar will not be uploaded, but I have to set the
CLASSPATH environment of the Yarn container, to point to a route on HDFS. Also
I found a config item starts with “containerized.master.env.”, I set the
“containerized.master.env.CLASSPATH” value to
"hdfs://xm-hdfs-test-01/henry/flink-application-0.0.1-SNAPSHOT-uber.jar”, but
with no luck. I also found that the
ConfigConstants.CONTAINERIZED_MASTER_ENV_PREFIX variable is marked as
deprecated, so maybe I can not use it.
So how could this requirement be implemented? Because I think it will
save a lot of time to start the Flink program.
Best
Henry
