Igor De SouzaKafka Compute Is Cheap. Network Is Not
https://dev.to/eeldor/kafka-compute-is-cheap-network-is-not-2bdh
Why Diskless Kafka must be 100% compatible with Kafka
HabrПовторная обработка сообщений в Kafka Consumer
Привет! Меня зовут Дмитрий Михеев, я ведущий разработчик в MAGNIT OMNI — бизнес-группе ритейлера «Магнит», которая отвечает за развитие омниканального опыта для клиентов. В своих сервисах для межсервисных коммуникаций помимо gRPC-запросов мы используем брокер сообщений Kafka. Если описывать его в двух словах, Kafka — это распределённый журнал событий (event log), через который сервисы обмениваются данными в реальном времени. Не буду подробно останавливаться на устройстве Kafka — это хорошо описано в документации. В этой статье хочу подсветить один неочевидный момент, который может привести к проблемам при работе с consumer’ами — повторную обработку сообщений (retry).
https://habr.com/ru/companies/magnit/articles/1043068/
#kafka #apachekafka #kafkaconsumer #retry #повторнаяобработка #идемпотентность #высоконагруженныесистемы #java
Kafka to Iceberg: Build a queryable data lake with ClickHouse
Kafka® to Iceberg: Build a queryable data lake with ClickHouse®
That’s ClickHouse below querying an Iceberg table on S3 within 0.31 seconds to read metadata and return the first rows. No Spark job, no data movement, and no separate warehouse layer to manage. By the end of this article, you’ll have the full pipeline running and understand why each component exists—not just how to configure......
Super interesting idea ...
BlobShuffle
Cost-effective repartitioning for Kafka Streams via cloud object storage
Kafka Schema Registry doesn’t guarantee compatibility (and what actually does)
The Bad PlaceTimes of India | Snowflake announces tools to help businesses adopt agentic AI systems: All details
AI generated summary, Read the full article for complete information.
Snowflake unveiled a suite of new AI‑focused tools at its 2026 summit aimed at enabling enterprises to move from experimentation to fully governed, agentic AI systems. The updates expand its AI coding assistant (CoCo) to platforms such as Slack, VS Code, Claude Code, mobile devices and Excel, and introduce Snowflake Datastream for real‑time Kafka‑based data pipelines. Enhancements to Snowflake CoWork provide context‑aware research, personalisation, and AI‑driven workflows for knowledge workers, while Cortex Training offers managed infrastructure for fine‑tuning foundation models. Governance gains new features through Horizon Catalog—including Horizon Context, Agent Identity, and an upgraded Trust Center—plus Adaptive Compute that auto‑scales resources for AI workloads. Interoperability is boosted with Apache Iceberg v3 support, Snowflake Storage for Iceberg tables, external engine access management, and the Iceberg REST Scan Plan API, delivering a single governed data layer across clouds. Finally, Automatic Data Agents and Agent Sharing let users transform shared datasets into conversational AI agents while maintaining strict governance controls.
#Snowflake #Microsoft #ApacheKafka #ApacheIceberg #SnowflakeSummit
Kafka’s quiet observability superpower — Kafka Interceptors
https://thej3.com/kafkas-quiet-observability-superpower-kafka-interceptors-aca88c33867e
KroxyliciousWe promised more details on how we got those Kroxylicious numbers https://kroxylicious.io/benchmarking/performance/2026/05/28/benchmarking-the-proxy.html
Shortest possible summary: we built a second cluster because the first one was lying to us.
The full story — catching that lie, and how complicated reliable numbers really are — is here: https://kroxylicious.io/benchmarking/performance/engineering/2026/06/03/benchmarking-the-proxy-under-the-hood.html
Does my proxy look big in this cluster?
Every good benchmarking story starts with a hunch. Mine was that Kroxylicious is cheap to run — I’d stake my career on it, in fact — but it turns out that “trust me, I wrote it” is not a widely accepted unit of measurement. People want proof. Sensibly.
