Apache Kafka follower replicas are often described as both synchronous and asynchronous, causing confusion. The distinction lies in two separate questions: how replication physically works (pull-based, followers can lag — asynchronous) versus when a write is considered committed (ISR followers must acknowledge before acks=all succeeds — synchronous). In Confluent Platform Multi-Region Clusters, this distinction is made explicit: normal replicas in the ISR are synchronous replicas, while observers are asynchronous replicas that don't participate in the ISR-based acknowledgement path. The min.insync.replicas configuration controls the minimum ISR size required for a write to succeed with acks=all, making it critical for durability guarantees in stretched and 2.5 DC architectures.

9m read timeFrom softwaremill.com
Post cover image
Table of contents
So which answer is correct?The source of the confusionWhat really happens during a Kafka writeWhy followers are often called asynchronousWhy Confluent 2.5 DC documentation calls followers synchronousThe 2.5 data center stretched cluster contextSummaryDocumentation links
2.1K Impressions