Hi Nix, The PythonDriverOptionsParserFactory still doesn't support parsing arguments other than "--python" and "-pym", see [1] for more details. Regarding how to work around your current requirements, I guess you could use configurations for now as all the CLI options have corresponding configurations, see [2] for more details. For example, you could use configuration `python.files` for `pyFiles` arguments.
Regards, Dian [1] https://github.com/apache/flink/blob/cae5fb4d3b6d9e0c10c3539ea4994fc1ad463b70/flink-python/src/main/java/org/apache/flink/client/python/PythonDriverOptionsParserFactory.java#L36 [1] https://nightlies.apache.org/flink/flink-docs-release-1.20/docs/dev/python/dependency_management/ On Tue, Aug 5, 2025 at 8:08 PM Nikola Milutinovic <n.milutino...@levi9.com> wrote: > > Hi all. > > PROBLEM: We cannot run a PyFlink job with our custom libraries, using K8s > Operator. > > We have a Kubernetes deployment of Flink 1.20.2, session cluster mode. > > We have a bunch of PyFlink jobs, which have a structure: > > Entrypoint Python script: defines the job. > Core library: core Python classes, shared by all environments. > Client library: client-specific Python classes. > > > We can successfully launch those jobs from command line, even from a > specialized deployer pod. The command we use is: > > bin/flink run \ > --pyFiles core-1.1.0-…whl \ > --pyFiles clinet1-1.2.1…whl \ > --python workflows/SCHED/entrypoint.py \ > --lookup Name > > And this is fine. However, when we try to launch them with K8s operator, we > are using the following snippet: > > apiVersion: flink.apache.org/v1beta1 > > kind: FlinkSessionJob > > metadata: > > name: client-shed-job > > spec: > > deploymentName: client1-main-flink > > job: > > args: > > - --python > > - /opt/flink/workflows/SCHED/entrypoint.py > > entryClass: org.apache.flink.client.python.PythonDriver > > parallelism: 1 > > state: running > > upgradeMode: savepoint > > restartNonce: 1 > > > If we try to add pyFiles arguments, PythonRunner complains about that. And, > looking into the source, there is no such option in the PythonDriver. > > So, how is it intended to be used? Do we have to have our libraries > pre-installed in the Docker image? > > Nix. >
