// AI news · 2026-11-14 · Databricks Blog
Spark Real-Time Mode Achieves Millisecond Latency
Spark Real-Time Mode achieves millisecond latency in streaming workloads, matching or beating Flink on p99 latency for stateless and stateful operations.
Published 2026-11-14 15:02 (Beijing time)Source Databricks BlogAuto-summarised, fact-checked against source
01Summary
It runs on existing Structured Streaming code with no rewrite, enabling unified batch and stream processing in one engine. Production use cases show significant latency reduction and reduced operational complexity.
Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.
02Key points
- Spark Real-Time Mode matches or beats Flink on p99 latency for key workloads.
- Real-Time Mode uses existing Structured Streaming code without rewriting.
- Production deployments report up to 500ms P99 latency and 44ms P50 latency.
- One engine supports batch, streaming, SQL, and ML, reducing duplication and governance overhead.
03Source
Apache Spark vs Apache Flink: which should you use?
Databricks Blog · published 2026-11-14 15:02 Beijing time
05Related news
- Same source · OpenAI and Databricks Enable Enterprise AI on Governed Data2026-11-13 · Databricks Blog
- Same source · Lakebase LTAP Architecture Debunks Myths2026-11-13 · Databricks Blog
- Same topic · The Warehouse Alleged AI Monitors Employee Bathroom Time2026-11-14 · IT之家
- Same topic · Routing with NVIDIA NeMo Switchyard on Amazon Bedrock2026-11-14 · AWS Machine Learning Blog
- Same day · Microsoft posts AI question during service outage2026-11-14 · IT之家
From AIMap AI news: collected from 1,100+ official and media sources. Summaries only restate the original article; follow the link for the full story. All stories of 2026-11-14 · Archive by date · All AI news · Today