Hi all! Recently I've been moving my jobs that use FlinkRunner to the latest 2.56.0 version from the 2.49.0. I've noticed the following:
- Busy time with 2.49.0: N/A literal - probably the 100% load wasn't happening before and wasn't being reported due to lack of metrics? - Busy time with 2.56.0: 100% even with no load. I'm using a 1 shard configuration and the simplest Kinesis configuration and still, 100% load shown. I've got the same pipelines but using Kafka as a Source, and there is no issue on that side of things, the load reports normal numbers and updated accordingly. This has become an issue when implementing autoscaling with flink-operator in the jobs since with that stage constantly under load I can't make it to downscale the number of paralellism and thus the number of Task Managers. How are those metrics being reported within IO classes? Is this something related to being translated into Flink or the Source connector itself?
