Remove the withdraw sessions
diff --git a/content/sessions/agenticcoding-1212139.md b/content/sessions/agenticcoding-1212139.md deleted file mode 100644 index 47f3224..0000000 --- a/content/sessions/agenticcoding-1212139.md +++ /dev/null
@@ -1,23 +0,0 @@ ---- -title: "Agentic Debugging: Using LLMs to Auto-Diagnose Impala Query Profiles" -date: "" -track: "agenticcoding" -presenters: "Surya Hebbar" -stype: "English Session" ---- - -Large Language Models show incredible promise for code generation, but debugging distributed systems is a different beast. If you feed a raw 400MB database query profile to an LLM, it will either choke on the token limit or hallucinate wildly. To turn agents into dependable collaborators in production, we have to fix the data pipeline feeding them. - -This session shares hands-on experience from Apache Impala on making "Agentic Debugging" a reality. We will explore how transitioning from verbose execution forests to Aggregated Runtime Profiles solves the LLM context problem. By dense-packing telemetry and safely structuring missing events into token-efficient JSON payloads, we empower AI agents to autonomously and accurately diagnose execution bottlenecks. Attendees will learn repeatable patterns for bridging the gap between massive system logs and strict LLM context windows. - -### Speakers: - - -<img src="https://cdn.sessionize.com/image/6b92-400o400o1-sPSKNvSRCLEreK26PfKD7w.jpg" width="200" /><br/> - -Surya Hebbar: Core Contributor to Apache Impala | MS Researcher at The University of Tokyo - -Surya Hebbar is a High-Performance Computing Engineer and a Core Contributor to the Apache Impala project. Working as a Software Engineer at Cloudera, his work has focused heavily on Impala's C++ backend and distributed systems optimization. He architected the "Aggregated Runtime Profiles" initiative, which fundamentally re-engineered the profile structure from sparse instance-level counters into dense, memory-efficient arrays to eliminate bottlenecks in high-concurrency clusters. He also led the modernization of Impala's WebUI and observability tools to handle petabyte-scale query execution graphs without locking up the browser. - -Surya is currently transitioning to pursue his MS in Earth and Planetary Science at The University of Tokyo. Working under Prof. Masaki Satoh, he will be leveraging exascale supercomputers like Fugaku and Miyabi to run high-resolution climate models, applying his expertise in parallel processing to the field of atmospheric fluid dynamics. -
diff --git a/content/sessions/agenticcoding-1212139.zh.md b/content/sessions/agenticcoding-1212139.zh.md deleted file mode 100644 index d064ee6..0000000 --- a/content/sessions/agenticcoding-1212139.zh.md +++ /dev/null
@@ -1,22 +0,0 @@ ---- -title: "Agentic 调试：用 LLM 自动诊断 Impala 查询 Profile" -date: "" -track: "agenticcoding" -presenters: "Surya Hebbar" -stype: "英文演讲" ---- - -大语言模型在代码生成方面展现出惊人的潜力，但调试分布式系统完全是另一回事。如果你把一份原始的 400MB 数据库查询 profile 喂给 LLM，它要么会被 token 上限噎住，要么会疯狂产生幻觉。要让 agent 在生产环境中成为可靠的协作者，我们必须先修好喂给它们的数据管道。 - -本次演讲分享来自 Apache Impala 的实战经验，讲述如何让"Agentic 调试"成为现实。我们将探讨从冗长的执行树（execution forest）转向聚合运行时 profile（Aggregated Runtime Profiles）如何解决 LLM 的上下文问题。通过对遥测数据进行高密度打包，并把缺失事件安全地结构化为节省 token 的 JSON 负载，我们让 AI agent 能够自主且准确地诊断执行瓶颈。听众将学到可复用的模式，用以弥合海量系统日志与严格的 LLM 上下文窗口之间的鸿沟。 - -### 讲师: - - -<img src="https://cdn.sessionize.com/image/6b92-400o400o1-sPSKNvSRCLEreK26PfKD7w.jpg" width="200" /><br/> - -Surya Hebbar：Apache Impala 核心贡献者 | 东京大学硕士研究者 - -Surya Hebbar 是一名高性能计算工程师，也是 Apache Impala 项目的核心贡献者。作为 Cloudera 的软件工程师，他的工作主要集中在 Impala 的 C++ 后端与分布式系统优化。他主导设计了"Aggregated Runtime Profiles"项目，从根本上对 profile 结构进行了重新工程化——从稀疏的实例级计数器改为密集的、内存高效的数组，以消除高并发集群中的瓶颈。他还牵头对 Impala 的 WebUI 和可观测性工具进行了现代化改造，使其能够处理 PB 级的查询执行图而不会卡死浏览器。 - -Surya 目前正转入东京大学的地球与行星科学系攻读硕士学位。他将在 Masaki Satoh 教授指导下，借助 Fugaku、Miyabi 等百亿亿次（exascale）超级计算机运行高分辨率气候模型，把他在并行处理方面的专长应用到大气流体动力学领域。
diff --git a/content/sessions/datalake-1203260.md b/content/sessions/datalake-1203260.md deleted file mode 100644 index 9b5ec20..0000000 --- a/content/sessions/datalake-1203260.md +++ /dev/null
@@ -1,26 +0,0 @@ ---- -title: "From database to lakehouse in real-time: CDC, Kafka, Apicurio, and Apache Iceberg" -date: "" -track: "datalake" -presenters: "Carles Arnal" -stype: "English Session" ---- - -Batch ETL runs nightly. Your analysts query stale data. Your ML models train on yesterday's features. The streaming-first lakehouse replaces all of that with a single, real-time pipeline — and you can build it entirely with open-source tools. - -In this talk, I'll demo a complete pipeline: Debezium captures row-level changes from PostgreSQL, streams them through Kafka with schemas enforced by Apicurio Registry (a CNCF sandbox project), and lands them in Apache Iceberg tables — queryable within seconds via Trino or Spark. I'll cover how the Flink Dynamic Iceberg Sink leverages schema registries for automatic schema evolution, eliminating the manual DDL changes that plague traditional data lakes. - -Attendees will leave with: -- A deployable CDC-to-Iceberg pipeline architecture using only open-source components on Kubernetes -- Practical patterns for handling schema evolution across the Kafka-to-Iceberg boundary -- A clear understanding of when this approach replaces batch ETL and where hybrid patterns still make sense - -### Speakers: - - -<img src="https://cdn.sessionize.com/image/ebeb-400o400o1-nXA4VjZFgbUQrzL5Hm1DKV.jpg" width="200" /><br/> - -Carles Arnal: Principal Software Engineer at Red Hat - -Carles Arnal is a Principal Software Engineer at IBM and a core maintainer of Apicurio Registry (CNCF sandbox project) working in the AI and Data Streaming space. He's also an associate professor at BarcelonaTech and he's an active committer of Quarkus with over 10 years of industry experience. -
diff --git a/content/sessions/datalake-1203260.zh.md b/content/sessions/datalake-1203260.zh.md deleted file mode 100644 index e3f975e..0000000 --- a/content/sessions/datalake-1203260.zh.md +++ /dev/null
@@ -1,25 +0,0 @@ ---- -title: "从数据库到实时湖仓：CDC、Kafka、Apicurio 与 Apache Iceberg" -date: "" -track: "datalake" -presenters: "Carles Arnal" -stype: "英文演讲" ---- - -批处理 ETL 在夜间运行。你的分析师查询的是过时数据。你的 ML 模型基于昨天的特征来训练。而流式优先（streaming-first）的湖仓用一条实时管道取代了这一切——而且你可以完全用开源工具来构建它。 - -在本次演讲中，我将演示一条完整的管道：Debezium 从 PostgreSQL 捕获行级变更，通过 Kafka 进行流转（其 schema 由 Apicurio Registry——一个 CNCF sandbox 项目——强制约束），最终落入 Apache Iceberg 表中——几秒钟内即可通过 Trino 或 Spark 查询。我还会讲解 Flink Dynamic Iceberg Sink 如何借助 schema registry 实现自动 schema 演进，从而消除困扰传统数据湖的手工 DDL 变更。 - -听众将带走： -- 一套可部署的、仅使用 Kubernetes 上开源组件的 CDC-to-Iceberg 管道架构 -- 处理 Kafka 到 Iceberg 边界处 schema 演进的实用模式 -- 清晰地理解这种方法何时能取代批处理 ETL，以及哪些场景下混合模式仍然合理 - -### 讲师: - - -<img src="https://cdn.sessionize.com/image/ebeb-400o400o1-nXA4VjZFgbUQrzL5Hm1DKV.jpg" width="200" /><br/> - -Carles Arnal：Red Hat 软件首席工程师 - -Carles Arnal 是 IBM 的软件首席工程师，也是 Apicurio Registry（CNCF sandbox 项目）的核心维护者，活跃于 AI 与数据流领域。他还是 BarcelonaTech 的副教授，以及 Quarkus 的活跃 committer，拥有 10 年以上的行业经验。
diff --git a/content/sessions/datalake-1206024.md b/content/sessions/datalake-1206024.md deleted file mode 100644 index 357f6e7..0000000 --- a/content/sessions/datalake-1206024.md +++ /dev/null
@@ -1,19 +0,0 @@ ---- -title: "First-Class Constraint Metadata in Iceberg: Portable Data Quality Across Engines" -date: "" -track: "datalake" -presenters: "Huaxin Gao" -stype: "English Session" ---- - -Modern lakehouse deployments increasingly need reliable, portable data quality guarantees across the ecosystem. Today, constraints are handled inconsistently across engines and connectors, and Iceberg does not yet provide an engine-agnostic constraint model. This session proposes adding constraint support to Apache Iceberg as first-class metadata: stable IDs, binding by field IDs for schema-evolution safety, and consistent cross-engine introspection. Phase 1 is pragmatic: NOT NULL stays enforced via required fields; CHECK constraints are stored and exposed by Iceberg and enforced on writes by engines that support it; UNIQUE/PRIMARY KEY/FOREIGN KEY are informational. Phase 1 also delivers an end-to-end Spark DSv2 integration (CREATE/ALTER with constraints, validation on ADD CHECK, and write-time CHECK enforcement where supported). For ADD CHECK, engines validate existing data on a specific table version and Iceberg commits only if the current snapshot/version still matches the validated token (strict CAS), preventing races and stale validations. - -### Speakers: - - -<img src="https://cdn.sessionize.com/image/be04-400o400o1-P3f2LM1B3ad2TjS4zYxdvG.jpg" width="200" /><br/> - -Huaxin Gao: Software engineer at Snowflake - -Huaxin Gao is a software engineer at Snowflake and an Apache Spark committer and PMC member. She is also a committer for Apache Iceberg and Apache DataFusion Comet, with contributions spanning query engines, table formats, and distributed data systems. -
diff --git a/content/sessions/datalake-1206024.zh.md b/content/sessions/datalake-1206024.zh.md deleted file mode 100644 index dea8163..0000000 --- a/content/sessions/datalake-1206024.zh.md +++ /dev/null
@@ -1,18 +0,0 @@ ---- -title: "Iceberg 中的一等公民约束元数据：跨引擎的可移植数据质量" -date: "" -track: "datalake" -presenters: "Huaxin Gao" -stype: "英文演讲" ---- - -现代湖仓部署越来越需要跨生态的、可靠且可移植的数据质量保证。如今，不同引擎和连接器对约束的处理并不一致，而 Iceberg 也尚未提供与引擎无关的约束模型。本次演讲提议在 Apache Iceberg 中将约束支持作为一等公民（first-class）元数据加入：稳定的 ID、按字段 ID 绑定以保障 schema 演进安全，以及跨引擎一致的元信息查询（introspection）。第一阶段以务实为主：NOT NULL 仍通过 required 字段来强制；CHECK 约束由 Iceberg 存储并暴露，由支持它的引擎在写入时强制执行；UNIQUE/PRIMARY KEY/FOREIGN KEY 则是信息性的。第一阶段还交付了端到端的 Spark DSv2 集成（带约束的 CREATE/ALTER、ADD CHECK 时的校验，以及在受支持情况下的写入时 CHECK 强制执行）。对于 ADD CHECK，引擎会在某个特定的表版本上校验已有数据，而 Iceberg 仅在当前快照/版本仍与已校验的 token 匹配时才提交（严格 CAS），从而防止竞态与过时校验。 - -### 讲师: - - -<img src="https://cdn.sessionize.com/image/be04-400o400o1-P3f2LM1B3ad2TjS4zYxdvG.jpg" width="200" /><br/> - -Huaxin Gao：Snowflake 软件工程师 - -Huaxin Gao 是 Snowflake 的软件工程师，也是 Apache Spark 的 committer 和 PMC 成员。她还是 Apache Iceberg 和 Apache DataFusion Comet 的 committer，贡献横跨查询引擎、表格式和分布式数据系统。