This is a snippet from my flink prometheus reporter. It showed that kafka current offsets and committed offsets metric names changed after I migrated my flink job from 1.6.0 to 1.6.3.
The origin metrics name should not contain partition index in metric name, i.e. the metric name should be flink_taskmanager_job_task_operator_KafkaConsumer_topic_partition_currentOffsets and flink_taskmanager_job_task_operator_KafkaConsumer_topic_partition_committedOffsets.
After digging into the source code, I found that the incompatibility started from this PR, because it overloaded a new getLogicalScope(CharacterFilter, char, int) and didn't override in GenericValueMetricGroup class.
When the tail metric group from a metric is GenericValueMetricGroup and this new getLogicalScope is called, i.e. calling FrontMetricGroup#getLogicalScope, the value group name will not be ignored, but it should be in previous released version.