[FLINK-8410] Kafka consumer's commitedOffsets gauge metric is prematurely set

• Type: Bug
• Status: Reopened
• Priority: Not a Priority
• Resolution: Unresolved
• Affects Version/s: 1.3.2, 1.4.0, 1.5.0
• Component/s: Connectors / Kafka, Runtime / Metrics
• Assignee: Unassigned
• Reporter: Tzu-Li (Gordon) Tai
• Created: 11/Jan/18 01:32
• Updated: 29/Jul/21 07:38

Description

The committedOffset metric gauge value is set too early. It is set here:
https://github.com/apache/flink/blob/master/flink-connectors/flink-connector-kafka-0.10/src/main/java/org/apache/flink/streaming/connectors/kafka/internal/Kafka010Fetcher.java#L224

This sets the committed offset before the actual commit happens, which varies depending on whether the commit mode is auto periodically, or committed on checkpoints. Moreover, in the 0.9+ consumers, the KafkaConsumerThread may choose to supersede some commit attempts if the commit takes longer than the commit interval.

While the committed offset back to Kafka is not a critical value used by the consumer, it will be best to have more accurate values as a Flink-shipped metric.