Improve error handling in org.apache.beam.sdk.extensions.ml.CloudVision
Open
extensions
gcp
java
new feature
P3
- Dominant language
- Java
- Stars
- 8.7k
- Forks
- 4.7k
- Avg merge
- 1d 20h
- Merged PRs (30d)
- 196
Description
CloudVision invokes batchAnnotateImages API ((at AnnotateImages class). Any exception thrown will propagate to the pipeline. Because of the Vision API's quota processing large number of images with a large number of workers will often result in ResourceExhaustedException and subsequent reprocessing of the complete bundle and pipeline errors.
By catching that exception in the transform and providing an exponential backoff reprocessing and excessive error reporting can be significantly minimized.
Imported from Jira [BEAM-10693](https://issues.apache.org/jira/browse/BEAM-10693). Original Jira may contain additional context.
Reported by: slilichenko.
Contributor guide
Assessment
This issue has not been assessed yet.