Flink relies on its checkpointing and savepointing feature for resilience and recovery. It does not work with the Kafka offsets or consumer groups for resilience. The offset commits during checkpoints are only for observability purposes, as documented on https://nightlies.apache.org/flink/flink-docs-master/docs/connectors/datastream/kafka/#consumer-offset-committing

If you don't recover from a checkpoint or savepoint, you will get undeterministic results with regards to data recovery.