[
https://issues.apache.org/jira/browse/FLINK-16943?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wei Zhong updated FLINK-16943:
------------------------------
Description:
Since flink-1.10.0 released, many users have complained that PyFlink is
inconvenient when loading external jar packages. For local execution, users
need to copy the jar files to the lib folder under the installation directory
of PyFlink, which is hard to locate. For job submission, users need to merge
their jars into one, as `flink run` only accepts one jar file. It may be easy
for Java users but difficult for Python users if they haven't touched Java.
We intend to add a `add_jars` interface on PyFlink TableEnvironment to solve
this problem. It will add the jars to the context classloader of Py4j gateway
server and add to the `PipelineOptions.JARS` of the configuration of
StreamExecutionEnviornment/ExecutionEnviornment.
Via this interface, users could add jars in their python job. The jars will be
loaded immediately, and users could use it even on the next line of the Python
code. Submitting a job with multiple external jars won't be a problem anymore
because all the jars in `PipelineOptions.JARS` will be added to the JobGraph
and upload to the cluster.
As it is not a big change I'm not sure whether it is necessary to create a FLIP
to discuss this. So I created a JIRA first for flexibility. What do you think
guys?
was:
Since flink-1.10.0 released, many users have complained that PyFlink is
inconvenient when loading external jar packages. For local execution, users
need to copy the jar files to the lib folder under the installation directory
of PyFlink, which is hard to locate. For job submission, users need to merge
their jars into one, as `flink run` only accepts one jar file. It may be easy
for Java users but difficult for Python users if they haven't touched Java.
We intend to add a `add_jars` interface on PyFlink TableEnvironment to solve
this problem. It will add the jars to the context classloader of Py4j gateway
server and add to the `PipelineOptions.JARS` of the configuration of
StreamExecutionEnviornment/ExecutionEnviornment.
Via this interface, users could add jars in their python job. The jars will be
loaded immediately, and users could use it even on the next line of the Python
code. Submitting a job with multiple external jars won't be a problem anymore
because all the jars in `PipelineOptions.JARS` will be added to the JobGraph
and upload to the cluster.
As it seems not a big change I'm not sure whether it is necessary to create a
FLIP to discuss this. So I created a JIRA first for flexibility. What do you
think guys?
> Support adding jars in PyFlink
> ------------------------------
>
> Key: FLINK-16943
> URL: https://issues.apache.org/jira/browse/FLINK-16943
> Project: Flink
> Issue Type: Improvement
> Components: API / Python
> Reporter: Wei Zhong
> Priority: Major
>
> Since flink-1.10.0 released, many users have complained that PyFlink is
> inconvenient when loading external jar packages. For local execution, users
> need to copy the jar files to the lib folder under the installation directory
> of PyFlink, which is hard to locate. For job submission, users need to merge
> their jars into one, as `flink run` only accepts one jar file. It may be easy
> for Java users but difficult for Python users if they haven't touched Java.
> We intend to add a `add_jars` interface on PyFlink TableEnvironment to solve
> this problem. It will add the jars to the context classloader of Py4j gateway
> server and add to the `PipelineOptions.JARS` of the configuration of
> StreamExecutionEnviornment/ExecutionEnviornment.
> Via this interface, users could add jars in their python job. The jars will
> be loaded immediately, and users could use it even on the next line of the
> Python code. Submitting a job with multiple external jars won't be a problem
> anymore because all the jars in `PipelineOptions.JARS` will be added to the
> JobGraph and upload to the cluster.
> As it is not a big change I'm not sure whether it is necessary to create a
> FLIP to discuss this. So I created a JIRA first for flexibility. What do you
> think guys?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
