Details
-
New Feature
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
Description
Currently, Flink assumes all resources are locally accessible in the pods, which requires users to prepare the resources by mounting storages, downloading resources with init containers, or rebuilding images for each execution.
We could make things much easier by introducing a built-in file distribution mechanism based on Flink-supported filesystems. It's implemented in two steps:
1. KubernetesClusterDescripter uploads all local resources to remote storage via Flink filesystem (skips if the resources are already remote).
2. KubernetesApplicationClusterEntrypoint and KubernetesTaskExecutorRunner download the resources and put them in the classpath during startup.
The 2nd step is mostly done by FLINK-28915, thus this issue is focused on the upload part.
Attachments
Issue Links
- blocks
-
FLINK-26541 SQL Client should support submitting SQL jobs in application mode
- Open
- links to