Top 3 this week
PostgreSQL fixes 52 security vulnerabilities and ends support for version 14
The quarterly update (18.7, 17.12, 16.16, 15.20 and 14.25) fixes 52 CVEs (published security vulnerabilities), 32 of them rated 8.8 out of 10 and mostly letting a signed-in database user run code on the server, plus more than 140 bugs. Some bugs could corrupt GIN and pg_trgm GiST indexes or incremental backups, so the notes ask you to reindex and take a fresh full backup after upgrading. Version 14.25 is the last release of PostgreSQL 14, and PostgreSQL 19 is planned for Nov 19.
Why it matters: Patch every PostgreSQL server this cycle, follow the reindex and backup steps, and put a date on getting off version 14.
Fivetran and dbt Labs now go by one name: Fivetran
The founders of Fivetran and dbt Labs published a joint letter, and the dbt site now says that Fivetran plus dbt Labs is Fivetran. They say the dbt brand, products and open-source community stay central, and they commit to open source, open standards and portability across platforms. The letter also outlines plans for AI agent infrastructure but gives no product dates.
Why it matters: If dbt and Fivetran are both in your stack, you now have one vendor relationship to manage, so watch pricing and the dbt Core roadmap closely.
Redpanda unveils a new engine and ends source availability for its streaming platform
Redpanda, a Kafka-compatible streaming platform, introduced R2, a new core engine written from scratch that will power its streaming and SQL products starting with release 26.3.1. The post says that starting with that release, Redpanda Streaming will no longer be source-available. A free Community Edition remains, limited to 50 cores per organization.
Why it matters: Teams running Redpanda should check whether the new license and the 50-core limit affect them before upgrading to 26.3.1.
Data Engineering
Polars says its distributed engine beat Databricks Photon on a TPC-H style benchmark
Polars ran its distributed engine, Polars Nanook, against Databricks Spark with Photon on PDS-H, a benchmark based on TPC-H, at 1 TB scale on Parquet files in S3. Polars reports Nanook was faster on 17 of 22 queries and about 1.3 times faster overall, and about 6 times faster than open-source Spark. These are the vendor's own numbers.
Amazon Redshift adds multi-statement transactions on Iceberg tables
Redshift can now run several INSERT, UPDATE, DELETE or MERGE statements against one Apache Iceberg table inside a single BEGIN and COMMIT block. The whole batch lands as one Iceberg snapshot, so other engines see all of the change or none of it. All statements must target the same table, and it needs Redshift patch 206 or later.
AWS Glue can keep Spark jobs on the latest runtime automatically
AWS Glue, the managed ETL (extract, transform, load) service, now offers an Auto version setting for Spark jobs. Before each upgrade, Glue checks compatibility, asks for consent, and compares the upgraded job's output to its usual output. Jobs that need code changes stay where they are and get a list of fixes.
Streaming and Real-Time
Debezium 3.6.4 fixes missed Oracle change events
Debezium, the open-source change data capture (CDC) project, shipped a maintenance release that resolves 35 issues. Several fixes address Oracle change events that could be missed, especially on Oracle RAC, and LogMiner queries are faster. It also upgrades the MySQL driver to address CVE-2025-30706, and MySQL 8.0 is no longer in that driver's documented supported range.
Amazon's managed Flink service can now move application state across apps and Regions
Amazon Managed Service for Apache Flink can now capture, export and restore application state between applications and AWS Regions. That makes blue/green deployments, moving savepoints over from self-managed Flink, and cross-Region recovery possible. It requires Flink 1.20 or later.
Analytics and Semantic Modeling
Snowflake semantic views can now model many-to-many relationships
Snowflake semantic views, its governed layer of metrics and relationships, can now model many-to-many relationships through a bridge table or directly, in preview. Snowflake finds the join path and removes duplicate rows so metrics are not double counted. Queries that cross these relationships cannot yet use level-of-detail or multi-path metrics.
Cube brings its semantic layer into ChatGPT
Cube launched a ChatGPT plugin that answers questions from Cube's semantic layer, using each person's own row-level permissions, and returns charts and dashboards in the chat. Missing metrics can be added on a development branch for review before they are merged. The plugin is live in ChatGPT's plugin directory.
Data Delivery
Sigma makes Workbooks and Reports as Code generally available
Every Sigma workbook and pixel-perfect report now has a JSON or YAML code version that can be read, written, versioned and validated through the REST API, including by AI coding agents. Sigma points to migrating dashboards from older BI tools and managing reports with Git and CI as the main uses. Native Git integration is planned next.
Snowflake's coding agent can convert Tableau workbooks into Streamlit apps
A preview skill in Cortex Code, Snowflake's AI coding assistant, turns a Tableau workbook into a multipage Streamlit in Snowflake app. Each dashboard becomes a page and calculations become Snowflake SQL. It builds against synthetic data first and switches to real tables only after the user approves.
Data Governance and Privacy
Italy fines Meta 805 million euros over Facebook's default access to Android photo galleries
Italy's data protection authority fined Meta 805 million euros (about $887 million) over a Facebook for Android feature that was on by default and reached users' photos, videos and their metadata through one broad permission. Meta has 60 days to turn it off by default and ask for separate consent. Meta calls the fine disproportionate and plans to appeal.
Five more states join the FTC's privacy case against Hims & Hers
Arizona, Florida, Nebraska, Ohio and Wisconsin joined the Federal Trade Commission's amended complaint against the telehealth company. The complaint alleges Hims & Hers shared consumers' sensitive health information with ad platforms, including Meta and Snap, while saying it kept that information private. It also alleges hard-to-cancel subscriptions.
Snowflake adds role grants that expire on their own
Admins can now grant a role for a set time, such as 24 hours of on-call access, or until a set timestamp, and Snowflake removes it automatically, in preview. SHOW GRANTS and the grants views gain an expires_on column. It makes least-privilege access, where people only hold the access they currently need, much easier to run.
Data as a Product
Databricks Pages, governed definitions of business concepts, are generally available
Pages are governed definitions of business concepts, organized by business domain on the Databricks Discover page. Curators can certify or deprecate them, and a new MANAGE PAGES permission controls who can publish them in each domain. In data mesh terms, it gives each domain a place to own what its terms mean.
AI in the Data Stack
Databricks rebuilds Genie as an AI coworker that runs multi-step tasks
Genie, Databricks' natural-language analytics assistant, can now run long multi-step tasks, ask for missing inputs and request approval before acting. Databricks also introduced Genie Ontology, a business-context layer built from existing dashboards and queries that people then govern, and Genie Search, which searches business apps while respecting each source's permissions.
A domain expert's AI analysis shows why checking the output still matters
Simon Spati walks through an analysis built with Claude and MotherDuck by someone outside the data team. Reviewing the output with business knowledge caught a date-parsing bug that had silently dropped 61 percent of the 2025 rows. It is a practical reminder that AI-built analysis still needs someone who knows what the numbers should look like.
Platforms
Databricks launches Lakehouse//RT for real-time analytics on lakehouse tables
Lakehouse//RT is a Databricks warehouse option for fast, high-concurrency queries run directly on governed lakehouse tables, pitched as a replacement for separate serving systems like Apache Druid or Apache Pinot. Databricks cites under 100 millisecond latency at 12,000 queries per second, which are vendor figures. It is in beta with a 30 percent discount through January 2027.
ClickHouse can now read and write Iceberg tables on Amazon S3 Tables
ClickHouse can connect to Amazon S3 Tables, AWS's managed Apache Iceberg storage, through the Iceberg REST catalog, query tables in place and write results back as Iceberg tables other engines can read. It needs self-managed ClickHouse 26.8 or later, and the write path is not yet generally available in ClickHouse Cloud.
Standards
Parquet adds a FINITE_VECTOR type for AI embeddings
The Parquet community approved FINITE_VECTOR, a new logical type for fixed-length vectors of numbers such as AI embeddings, with no votes against. On Nov 10 the spec change was confirmed merged into the Parquet format, and a pull request to document the type on the Parquet website was opened.
Apache Polaris 1.9.0 improves its Iceberg catalog and outside identity support
Polaris is an open-source catalog for Apache Iceberg tables. Version 1.9.0 moves to Iceberg 1.12.0, turns on paging for list calls by default, and keeps newly written metadata when a commit's outcome is unknown. It also fully supports users managed outside Polaris when an external identity provider and policy engine such as Open Policy Agent or Ranger is set up.
Open Source
Apache Spark votes to bring GraphFrames into the main project
A Spark Project Improvement Proposal to move GraphFrames, the DataFrame-based graph processing library, into Apache Spark passed with 14 binding and 9 non-binding votes in favor and none against. That should make graph analysis a built-in part of Spark rather than a separate package.
Data Security
Apify, a web-scraping platform, confirms customer data was accessed
An intruder used one compromised credential to collect more credentials and reached Apify customer accounts on Nov 9 and 10. Apify confirmed that some customers' personal information and stored data were accessed, rotated internal secrets and revoked some OAuth connections, including GitHub. Customers are told to reconnect their GitHub integrations.
sqlparse fixes two bugs and announces its last release
sqlparse is a widely used Python library for parsing and formatting SQL. Version 0.6.1 fixes a code injection flaw (CVE-2026-112307) in its PHP output format and a denial-of-service bug in format(). Its sole volunteer maintainer says this is the final release and no more security fixes are planned, so projects that depend on it should plan for a fork or replacement.
rsyslog's ClickHouse output could let log senders inject SQL
rsyslog's optional module for shipping logs into ClickHouse ignored ClickHouse's backslash escapes (CVE-2026-93404, rated High). Anyone whose log messages are routed to ClickHouse could change the SQL to forge stored log records or run expressions with the ClickHouse account's privileges. It is fixed in the Nov 11 daily-stable build, and Linux distributions need to update their own packages.