May 2025 Summaries

Seemplicity, a RemOps platform, faced challenges as it scaled due to Postgres becoming a bottleneck. The company needed a new foundation for real-time analytics at scale, which led them to rebuild their backend with ClickHouse Cloud and PeerDB (now part of ClickPipes). By decoupling analytical workloads from transactional layers, Seemplicity achieved sub-second query performance across billions of rows, improved data compression, and increased transparency. The migration delivered big-time performance gains, simplified analytics, and gave the team confidence that their analytics stack wouldn't break under pressure. With a faster and more scalable analytics engine in place, Seemplicity aims to improve precision and control, knowing exactly what's happening under the hood.

ClickHouse, a leader in real-time analytics and data warehousing, has raised $350 million in Series C financing led by Khosla Ventures. The funding will be used to scale product development, support global expansion, and deepen partnerships with customers and technology providers building AI-native applications. ClickHouse has grown over 300% in the past year and now serves over 2,000 customers across various industries, including fintech, transportation, consumer, and healthcare. The company's high-performance analytics capabilities are designed to meet the demands of the agentic era, where enterprises need real-time data platforms that can support both traditional analytics and AI-native workloads. With this funding, ClickHouse is poised to become a default engine for next-generation intelligent data products, enabling fast, scalable, and cost-efficient analytics in an increasingly complex market landscape.

The Postgres CDC connector for ClickPipes is now generally available, allowing customers to easily replicate their Postgres databases to ClickHouse Cloud. This integration represents a major milestone and supports enterprise-grade Postgres use cases. The connector has undergone an extensive beta phase, evolving rapidly based on user feedback, and now supports hundreds of mission-critical workloads, moving 100TB+ of data per month into ClickHouse. Reference customers include Ashby, Seemplicity, and AutoNation, who have seen significant improvements in their real-time analytics and reporting capabilities. The connector offers a suite of features, including 10x faster initial loads and resyncs, replication latencies as low as 10 seconds, automatic replication of schema changes, and secure connectivity. Pricing for the connector is highly competitive, with a metered monthly billing structure starting September 1st, 2025. Customers can try the Postgres CDC to ClickHouse integration by following links provided in the announcement.

Today, ClickStack is announced as a new open-source observability solution built on ClickHouse. It delivers a complete and out-of-the-box experience for logs, metrics, traces, and session replay, powered by the performance and efficiency of ClickHouse. This modern approach breaks away from traditional "three pillars" models and eliminates complexity in stitching together siloed telemetry sources. With ClickStack, teams can enjoy fast search, fast aggregations, and high-cardinality data handling without sacrificing cost or scalability. The solution is built on HyperDX, an open-source observability layer purpose-built on ClickHouse, which brings standards-based data collection, open-source first philosophy, and a UI tightly coupled to the engine for fast performance. ClickStack's modular design allows each layer to scale independently, making it easy to grow with your data and team without overhauling the stack. The solution is now available for everyone, including teams that previously relied on general-purpose visualization tools or third-party proprietary platforms built on ClickHouse.

The new distributed cache for ClickHouse Cloud is a shared, low-latency layer that gives all compute nodes fast access to hot data. It solves the challenge of maintaining continuity of hot data in decoupled cloud databases and provides instant access to cached hot data across elastic, stateless compute nodes. The cache uses a network service that encapsulates the same caching logic as the local filesystem cache but makes it available to all compute nodes, bringing consistency, elasticity, and low latency to hot data. Benchmarks show that the distributed cache delivers SSD-grade performance in full-table scans and outperforms traditional setups by up to 4 times, with cold queries reaching sub-60 ms memory-speed latency without relying on local disks. The cache currently supports S3 and GCS and is now available in a private preview for early access and feedback.

ClickHouse, renowned for its speed as an analytics database, has further enhanced its performance with the introduction of delayed materialization, an optimization that postpones reading column data until it's actually needed. This approach is particularly effective for workloads with large datasets and top-N queries often found in observability and general analytics. In a striking example, a query execution time was reduced from 219 seconds to just 139 milliseconds, representing a 1,576-fold speed increase, all without altering a single line of SQL. ClickHouse achieves this by implementing a layered stack of I/O optimizations, including columnar storage, sparse primary indexes, secondary data skipping indexes, projections, and PREWHERE filters, which collectively minimize unnecessary data processing. Delayed materialization completes this stack by deferring reading of columns until subsequent operations like sorting and applying LIMIT clauses necessitate it, thereby dramatically enhancing performance, especially when only a few rows are required from large columns. This method is made possible by ClickHouse's column-oriented storage, allowing it to outperform traditional row-oriented databases that would otherwise need to read all columns simultaneously.

ClickPipes has simplified the process of setting up robust continuous data pipelines to ClickHouse Cloud, allowing users to connect various sources with just a few clicks. The tool has successfully moved over 20 trillion rows and now introduces features such as a revamped UI for greater clarity and efficiency, support for editing running pipelines, dedicated API endpoints and a Terraform provider for automated management of pipeline configurations, AWS PrivateLink self-service for secure connectivity, notifications for real-time alerts about pipeline activity, Prometheus exporter for observability, and centralized system tables for monitoring pipeline status. These new capabilities aim to provide a complete ingestion platform for ClickHouse Cloud with automation, visibility, security, and developer-friendliness.

Chartmetric, a music analytics platform, struggled to keep up with its growing data volumes as it scaled. To address this challenge, the company turned to ClickHouse Cloud, an open-source relational database engine that provides high-performance time-series queries and supports complex, relational workloads. By migrating from Postgres RDS to ClickHouse Cloud, Chartmetric was able to improve performance, reduce latency, and scale its analytics capabilities without increasing operational complexity. The company's experience with ClickHouse Cloud has been overwhelmingly positive, allowing it to handle massive datasets, run complex queries faster, and provide real-time insights into music industry trends. With ClickHouse Cloud, Chartmetric can now focus on delivering value to its customers, rather than being bogged down by data management challenges.

Monitoring ClickHouse can be challenging, especially when navigating between system tables and ad hoc queries. The ClickHouse Support team previously provided essential monitoring queries to help users monitor their instance using powerful system tables. In this post, a centralized, reusable dashboard is created in ClickHouse Cloud, providing real-time visibility in just a few clicks. A query on `system.parts` is used to provide a global overview of the cluster, showing the biggest tables in terms of rows, data, and primary key size. This query is saved and added to a new dashboard, which includes additional queries to display average query duration, number of requests, and user activity. The dashboard is interactive, allowing viewers to filter data by time period, and is easily shareable via its custom URL. With this dashboard, users can optimize performance and reduce the time spent hunting down queries.

Kami, a New Zealand-based edtech company, faced significant growth challenges in the wake of the COVID-19 pandemic, with weekly event volumes increasing by 23 times from 2019 to 2020. Their original Postgres-based data architecture struggled under the load, leading to slow queries and rendering analytics dashboards nearly useless. In response, Kami migrated to ClickHouse, an open-source columnar storage system that enabled efficient compression and lightning-fast queries on massive datasets. The migration allowed for real-time event processing without overwhelming infrastructure, and has since focused on bringing in more contextual data for deeper insights. With their new system, Kami can now scale even further, with plans to move to PeerDB for improved reliability and optimize ingestion processes.

ClickHouse Cloud on Microsoft Azure has been well-received by developers, data engineers, and architects since its launch last year. The platform offers a turnkey cloud-native database that can power interactive data-driven applications with high-performance analytical workloads. Weights & Biases launched Weave, an LLM observability tool, built on ClickHouse Cloud, while Astronomer launched Astro Observe, a ClickHouse-powered observability platform. To further support Azure customers, we have doubled down on Azure ecosystem support. The integration of Azure OpenAI with ClickHouse offers powerful real-time analytics and natural language interaction capabilities. Additionally, ClickHouse now supports Azure Data Lake, allowing users to read and write over 70 different data formats. New connectors for Azure Blob Storage, Event Hubs, and Azure Data Factory have been introduced, making it easier for users to integrate ClickHouse with their existing workflows. The platform also offers a Power BI connector, chDB .NET bindings, and enterprise-grade security features. With the launch of ClickHouse Cloud on Azure Marketplace, customers can easily access and deploy the solution, taking advantage of the MACC program for cost-effectiveness.

Boundary's BAML programming language aims to bring engineering discipline and type safety to LLM workflows by defining prompts as typed functions with clear inputs, outputs, and schemas. This allows developers to test, version, and debug prompts like any other code, enabling a hot reload loop for LLMs and reliable extraction of structured data from model output. By integrating BAML with ClickHouse, the Boundary team provides a complete analytics backbone for prompt engineering, offering schema-aware logging, real-time observability, and fast query capabilities to transform the way developers build and maintain LLM applications.

AI agents are increasingly becoming essential players in real-time analytics, functioning as autonomous entities that interact with data systems to optimize business metrics. These agents, enhanced by large language models (LLMs) like the GPT series, have evolved beyond simple digital assistants and now possess the ability to understand context, make decisions, and execute tasks with varying levels of autonomy. Real-time analytics databases, essential for processing and providing timely insights, are adapting to accommodate these AI agents, treating them as new user personas. The integration of AI agents into data systems has prompted the exploration of new database optimizations and features to support their complex workloads, such as enhanced SQL generation, improved data exploration capabilities, and LLM-friendly documentation. These developments highlight the shift towards AI-driven data interactions, presenting both opportunities and challenges in ensuring efficient and responsive systems. As organizations continue to deploy AI agents at scale, the relationship between these agents and real-time databases is expected to evolve, leading to further innovations and optimizations in the data analytics landscape.

In the wake of Anthropic's 2024 release of the MCP protocol, ClickHouse's integration team demonstrated a basic integration of Anthropic's Sonnet model with the ClickHouse database, prompting the creation of Dwaine, an internal LLM designed to assist company users in querying data for insights like revenue and customer behavior. This initiative led to the development of AgentHouse, an interactive demo environment showcasing the seamless integration of ClickHouse's real-time analytics with large language models, accessible without account creation or data upload. AgentHouse utilizes Anthropic's Sonnet model for its advanced understanding of structured data and integrates with the open-source LibreChat UI for user-friendly interaction. It features the ClickHouse MCP server to bridge data transfer and query optimization between databases and LLMs, providing access to various public datasets for testing and exploration through natural language queries. Users can experience real-time analysis and automated data visualization, emphasizing ClickHouse's suitability as a backend for LLM applications.

ClickPipes implements Change Data Capture (CDC) for MySQL databases, providing real-time data integration with ClickHouse. The system leverages MySQL's native replication capabilities and implements intelligent processing pipelines to balance reliability, performance, and ease of use. It supports both GTID-based and binlog position-based replication methods, handles various data types, and includes features like transaction batching, idle timeouts, parallelism, and backoff strategies for optimal performance. The system is designed to handle high-frequency transactions, large bulk operations, and realistic mixed workloads, with sub-30s latency under optimal conditions. However, it has some limitations, including schema changes, data type restrictions, replication requirements, and support for TRUNCATE operations.

March's ClickHouse newsletter highlights several key developments and community events. Notably, the Postgres CDC connector for ClickPipes is now in public beta, enabling seamless data replication from PostgreSQL to ClickHouse Cloud. The newsletter introduces Matteo Pelati, co-founder of LangDB, as the featured community member, and details his contributions to integrating AI functions with ClickHouse. Various upcoming events, including the Open House User Conference in San Francisco, offer opportunities for both new and experienced users to gain insights into ClickHouse applications. The release of ClickHouse 25.2 brings performance enhancements and new features such as Parquet Bloom filters and a backup database engine. The newsletter also discusses the integration of ClickHouse with Grafana for handling high cardinality metrics and explores ClickHouse's evolving role in data lake ecosystems. Furthermore, AWS's general availability of the ClickHouse BYOC (Bring Your Own Cloud) model is announced, allowing companies to run ClickHouse Cloud within their AWS VPC environments.

The April 2025 edition of the ClickHouse newsletter highlights several key developments and events related to its community and technological advancements. Noteworthy updates include the introduction of a new query condition cache in version 25.3, enhancing query performance for dashboards and observability use cases. The newsletter also covers the experience of CloudQuery using ClickHouse as a database backend for six months, emphasizing its performance and scalability, despite some challenges. ClickHouse's acquisition of HyperDX aims to bolster open-source observability solutions, combining HyperDX's user-friendly interface with ClickHouse's robust database capabilities. Additionally, the newsletter features Julian LaNeve from Astronomer, who discusses their choice of ClickHouse Cloud for Astro Observe due to its fast query performance and low maintenance requirements. Upcoming events include the ClickHouse User Conference in San Francisco, featuring discussions on the future of data infrastructure for large-scale AI. The newsletter also mentions a strategic shift to the "Make Before Break" scaling approach in ClickHouse Cloud, significantly reducing downtime during scaling operations. Various community contributions and blog posts explore using ClickHouse in different contexts, such as real-time data pipelines and .NET applications, showcasing its versatility across different platforms and use cases.

The newsletter for February 2025 highlights the release of JSONBench, a benchmark suite for JSON analysis, alongside various insights from experts like Ryadh Dahimene on agent-oriented analytics and Shahar Gvirtz on the benefits of ClickHouse, particularly its data compression capabilities. It features Chris Lawrence as a notable community member, detailing his contributions to transforming data pipelines at AMP using ClickHouse Cloud. The newsletter also covers upcoming global events, including training workshops and meetups, and explores recent developments in ClickHouse Cloud, such as compute separation and improved auto-scaling. Additionally, it discusses the integration of ClickHouse with Cribl for optimized data ingestion and analysis, and reviews the 25.1 release's enhancements, including improvements in join operations and CLI usability. The newsletter also introduces various ClickHouse-based projects and shares video content demonstrating database synchronization methods and real-time solutions utilizing ClickHouse and Kafka.

The first ClickHouse newsletter of 2025 highlights the 24.12 release, introducing Apache Iceberg REST catalog and schema evolution support, along with various features like improved Enum usability and experimental table sorting functions. Key discussions include the use of ClickHouse in building product analysis solutions, as detailed by Chloé Carasso, and optimization strategies for bulk inserts into partitioned tables shared by Jesse Grodman. The newsletter also covers the infrastructural evolution of Langfuse, an open-source LLM observability platform, and examines the implementation of medallion architecture in data warehousing with ClickHouse. Highlighted community member Jason Anderson, from Skool, shared insights on transitioning from Postgres to ClickHouse for enhanced data processing speed. Additionally, the newsletter outlines upcoming global events and workshops, and features contributions and insights from various ClickHouse users and experts on topics such as data integration, array processing functions, and efficient data pipeline construction.

The May 2025 ClickHouse newsletter highlights various advancements and community contributions related to the ClickHouse database management system. It delves into the implementation of "lazy materialization," which optimizes query performance by delaying column data reading until necessary, and covers Microsoft Clarity's choice of ClickHouse for its superior handling of large-scale data analytics. The newsletter also introduces AgentHouse, a chat-based environment using large language models for ClickHouse datasets, and discusses strategies for handling large data inserts with UUID range bucketing. Additionally, the newsletter outlines upcoming global events, training sessions, and meetups, while featuring community member Can Tian and their contributions to the dbt-clickhouse project. It also explores various case studies and implementations, such as Vimeo's use of ClickHouse for advanced viewer retention analytics, as well as new features and updates, including the private preview of the MySQL CDC connector in ClickPipes.

We're excited to announce the release of our new ClickHouse Cloud Prometheus/Grafana mix-in, designed to make monitoring your ClickHouse Cloud services as easy as possible. This mix-in leverages our existing Prometheus-compatible API endpoint to scrape ClickHouse metrics into your existing Prometheus and Grafana setup, providing real-time visibility into your services' health and performance with a pre-configured dashboard. Monitoring is crucial for maintaining the health and performance of your ClickHouse deployment, and this mix-in solves problems associated with scraping individual instances or managing a large number of metrics. The mix-in provides an optional filtered_metrics parameter to pare down the 1000+ available metrics to a more manageable 125 'mission critical' metrics. To get started, update your Prometheus config to scrape metrics from all services in your organization, restart Prometheus, and then add the Prometheus data source to Grafana and import the pre-configured dashboard.

The author recently organized a data visualization hackathon at their company's all-company offsite event. The dataset used for the challenge was Foursquare Places, which contains information about places on a map, including shops, restaurants, parks, playgrounds, and monuments. The author created a ClickHouse database to handle the dataset and built a visualization tool using a small command-line tool that provides a full ClickHouse engine. The visualization tool is able to load the data in 42 seconds and take up 11 GB of space, and it provides a beautiful and interactive experience for users. The author compared their tool with other similar projects, including Foursquare Studio, which loads much slower and has a low resolution, and Kepler.gl, which is limited by its ability to process large datasets without lagging. Overall, the author concludes that ClickHouse is a good option for analytics on large-scale geographical datasets due to its fast query processing capabilities and scalability.

ClickHouse is a fast query engine that can run on Parquet files directly without ingestion, outperforming many databases when querying their own native formats. It has been optimized for Parquet for years and its current reader applies parallelism across every layer of the query execution, using metadata like min/max statistics and Bloom filters to skip unnecessary work. A new native Parquet reader is on the way, bringing support for dictionary-based filtering, page-level min/max stats, and ClickHouse-specific optimizations like PREWHERE and lazy materialization. When benchmarked against other file formats and a purpose-built table engine, ClickHouse's performance over Parquet came closest to the engine. This makes ClickHouse a solid foundation for Lakehouse architectures, not just fast for Parquet but already there.

The latest release of ClickHouse version 25.4 brings numerous exciting features and improvements, including lazy materialization, which defers reading column data until it's actually needed, resulting in significant performance gains; Apache Iceberg time travel, allowing users to run queries based on previous snapshots; correlated subqueries for the `EXISTS` clause, enabling more complex querying capabilities; persistent databases in clickhouse-local, ensuring database persistence even when exiting and relaunching the CLI; default compression codecs for tables and globally for all tables via a config file; and support for the SSH protocol, allowing users to connect directly to the ClickHouse Server. These features are designed to enhance performance, simplify querying, and provide more flexibility in managing data.

ClickHouse has become "lazier" with the introduction of lazy materialization, which defers reading column data until needed, resulting in faster queries. The new feature is accompanied by CPU slot scheduling and improved quality-of-life updates. Microsoft Clarity chose ClickHouse as a solution for its free analytics tool due to its scalability and performance issues with the original Elasticsearch and Spark setup. AgentHouse, a chat-based demo environment, allows users to interact with ClickHouse datasets using natural language processing. CloudQuery solved the problem of handling large batches of data in ClickHouse by developing an "Insert-Splitter" algorithm that breaks up inserts into smaller chunks based on UUID ranges. The MySQL CDC connector for ClickHouse is now available in private preview. Bootstrapping statistical algorithms with ClickHouse led to a successful trial-and-error process using native functions and Python user-defined functions. Vimeo built a sophisticated retention analytics system using ClickHouse, tracking viewing patterns by recording changes and using window functions to calculate cumulative views at each second. The company also developed an AI-powered insights layer to generate concise, actionable insights about viewer engagement patterns. Various videos were shared on topics such as adopting ClickHouse for test analytics, observability challenges and solutions, Model Context Protocol, backfilling materialized views, querying Apache Iceberg tables via AWS Glue catalog, and explaining ClickHouse's JSON data type.

In this interview, Martijn Witteveen from Anlytic.com shares their experience with ClickHouse and its impact on their data analytics. They were previously using Postgres but switched to ClickHouse due to its exceptional performance, scalability, and flexibility. The switch has enabled them to remove the data accessibility barrier by distributing data directly around an organization, reducing duplication of work and improving communication. Anlytic.com is creating a new category in the market with their product as a substitute for traditional BI tools. They are using ClickHouse's on-demand scaling, performance, and flexibility to drive innovation and efficiency in their organization. The company plans to add managed ClickHouse options and data connectors to their platform, aiming to create predefined data hubs that connect various applications and services.