Understand this story
On November 14, 2026, Databricks released a streaming benchmark comparison between Spark Real-Time Mode and Apache Flink. The test ran at 10,000 records per second for 30 minutes, measuring end-to-end latency through Kafka; Spark was deployed on Databricks, and Flink was deployed on AWS Managed Service for Apache Flink 1.18. Databricks claims that Spark Real-Time Mode has matched or outperformed Flink in p99 latency on stateless, aggregation, and join workloads, with the official summary stating it is up to 92% faster. It also noted that Flink remains suitable for scenarios such as very large keyed state and complex event processing; Spark's stateful operators will be open-sourced in Apache Spark 4.3 and are currently available on Databricks.
AI-generated from public reports
Latest development · Report
Apache Spark vs. Apache Flink Stream Processing Comparison: Spark Real-Time…Developments
1 confirmed developments
Databricks releases streaming benchmark comparison of Spark and Flink
Source:Databricks Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 14
Databricks BlogFirst-party
Apache Spark vs. Apache Flink Stream Processing Comparison: Spark Real-Time…
Databricks benchmarks show that Spark Real-Time Mode has caught up with or surpassed Apache Flink in p99 latency on stateless, aggregation, and join workloads, with the official summary stating it is up to 92% faster. The test ran at 10,000 records/second for 30 minutes, measuring Kafka end-to-end latency, with Spark running on Databricks and Flink running on AWS Managed Service for Apache Flink 1.18. Flink remains suitable for scenarios such as very large keyed state and complex event processing; Spark's stateful operators will be open-sourced in Apache Spark 4.3 and are currently available on Databricks.