June 2026 Summaries

Tinybird has introduced a new lightweight delete functionality for ClickHouse, improving upon its previous data deletion method that required costly and time-consuming mutations. The new endpoint, POST /v1/datasources/{name}/delete, utilizes a feature in ClickHouse that materializes a hidden _row_exists mask, allowing for faster deletion acknowledgment and reduced resource consumption by reusing unchanged column files instead of rewriting affected parts. This ensures that deleted rows are filtered out during query read time, with the physical removal occurring later during background merges. The system supports both synchronous and asynchronous modes, providing flexibility for different use cases such as GDPR compliance, error removal, and partition-scoped cleanups. Users can implement this new functionality via the Tinybird CLI or API, with detailed instructions available in their Data Sources API reference, allowing for efficient data management without incurring the extensive costs associated with traditional methods.

ClickHouse is a high-performance analytical database optimized for handling large datasets through its unique architectural choices like columnar storage, vectorized execution, sparse indexes, and materialized views. These features collectively enhance performance as data scales, enabling efficient query execution that would otherwise be cumbersome on traditional row-oriented databases. ClickHouse allows for precise schema design with sort keys that minimize data scanning, along with approximate aggregation functions that maintain accuracy with reduced memory usage. Materialized views facilitate pre-aggregation, significantly reducing query times by summarizing data upon ingestion. Additional features like skip indexes, effective partitioning strategies, and automated data lifecycle management further improve performance and manageability. Tinybird, a managed ClickHouse service, simplifies the setup and maintenance of these features, offering seamless integration for large dataset workloads with streaming ingestion and real-time querying capabilities. By implementing these strategies, ClickHouse achieves sub-second query latencies across expansive datasets, turning what is typically a challenge for traditional databases into an expected performance standard.

Marketing dashboards often experience slowdowns due to the accumulation of vast event logs from campaigns, resulting in inefficient standard reporting queries. To address this, ClickHouse offers a solution through its columnar storage and vectorized execution, enabling faster processing of marketing queries without relying on pre-computed reports. The text discusses various techniques, such as event schema design, attribution models, and funnel analysis, to enhance real-time marketing dashboards, ensuring they remain efficient as event volumes increase. Additionally, Tinybird, a managed ClickHouse service, provides a streamlined approach for creating real-time marketing dashboards by offering SQL endpoints that maintain sub-second query response times, even with massive data volumes, thus transforming dashboards into current, actionable tools rather than mere historical reports.

Modern logistics operations depend heavily on real-time data to manage shipments, inventory, and fleet operations efficiently, with ClickHouse playing a crucial role in processing the high event volume that traditional databases struggle to handle. Logistics platforms generate continuous data streams, such as shipment status updates and vehicle telemetry, which require real-time analytical querying to prevent operational issues like shipment delays or stockouts. ClickHouse's MergeTree engine facilitates real-time insights into shipment tracking, inventory levels, and fleet performance by leveraging append-only event histories and optimized query patterns. The platform's capabilities extend to supply chain visibility and lead time analytics, allowing logistics teams to evaluate carrier performance, manage inventory risks, and respond promptly to operational exceptions. Tinybird enhances this ecosystem by serving as a managed ClickHouse solution with streaming event APIs and SQL endpoints, enabling logistics control towers to access unified, up-to-date data without the need for additional infrastructure management. By bridging various data sources such as carrier APIs and ERP systems, Tinybird ensures that logistics teams can maintain operational speed, ensuring that dashboards and alerts reflect the latest information, thus supporting more informed decision-making in a fast-paced environment.

Machine learning pipelines face challenges in bridging the gap between training on static snapshots and serving features computed over live data with low latency, leading to feature skew and infrastructure complexity. ClickHouse serves as an efficient analytical layer that computes features over event streams, storing raw events and materializing feature aggregations for online serving. This approach enables ML teams to build feature engineering patterns, online feature store architecture, and model monitoring systems using ClickHouse. By leveraging event logs as feature sources, users can perform rapid computations over large datasets for tasks such as training data extraction, model monitoring, and drift detection. The integration of ClickHouse with Tinybird facilitates seamless feature serving without the need for extensive infrastructure, providing SQL-driven pipelines that handle both online and offline data processing. Tinybird's platform offers a unified environment for managing the full ML data loop, from raw event ingestion to feature serving and model monitoring, significantly reducing the complexity and latency associated with traditional ML infrastructure setups.

IoT deployments generate vast amounts of time-series data that can overwhelm traditional databases, necessitating specialized solutions like ClickHouse® for efficient handling. ClickHouse is tailored for high-throughput, append-only writes, utilizing columnar storage and vectorized execution to swiftly process and compress time-series data, enabling rapid query responses even when dealing with billions of rows. The document discusses the application of ClickHouse for managing IoT data, highlighting key features such as the device telemetry schema, time-series aggregation, and anomaly detection, which are essential for maintaining performance and accuracy in IoT dashboards. It also covers data lifecycle management through TTL rules and the use of materialized views to optimize query performance, illustrating how these practices support effective fleet-wide analytics and geospatial tracking. Furthermore, the text introduces Tinybird as a complementary solution for real-time data ingestion and querying, offering a seamless integration with ClickHouse for building real-time applications with low latency and maintaining data isolation through parameterized SQL endpoints.

Postgres and SQLite are both SQL-speaking databases with support for transactions, indexes, and relational data modeling, but they are designed for different environments and use cases, much like comparing a local file system to a network file server. SQLite, an embedded C library, is perfect for applications that require zero-configuration deployment, making it ideal for local tools, mobile apps, and single-client scenarios due to its simple setup and predictable performance for single-writer workloads. In contrast, Postgres is a client-server database that excels in handling concurrent writes from multiple clients, providing a rich type system, and supporting complex SQL features, making it suitable for web applications, APIs, and systems requiring robust data integrity and concurrency. Both databases, however, are not optimized for analytical queries over large datasets; Postgres teams often stream data to columnar stores like ClickHouse for OLAP needs. Managed services like Tinybird offer integration with Postgres for real-time analytics, enabling sub-second queries without needing a separate data warehouse.

Healthcare organizations face challenges in managing vast volumes of clinical event data, often overwhelming traditional analytics systems. A 500-bed hospital, for instance, generates continuous streams of data like ADT events, lab results, and medication records, scaling up to hundreds of millions of events monthly at the health system level. ClickHouse, a real-time data processing solution, effectively manages this volume, supporting various healthcare analytics needs such as real-time patient monitoring, anomaly detection in lab results, and quality metrics like 30-day readmission rates. The system uses a schema that logs clinical events as append-only records, enabling efficient queries for operational dashboards and compliance reporting. Tinybird, a managed ClickHouse service, further enhances healthcare analytics with real-time data ingestion and query capabilities, offering solutions for clinical monitoring and population health management. It uses SQL-based endpoints to deliver timely insights without requiring infrastructure management, making it suitable for dynamic healthcare environments that demand both rapid data processing and strict compliance with data regulations.

Postgres and SQL Server are both robust databases capable of managing production-level relational workloads, each with distinct advantages and limitations. Postgres is open source with a cost-effective licensing model and boasts an extensive extension ecosystem, offering strong SQL standard compliance, JSONB support for semi-structured data, and flexibility for geospatial and AI workloads. SQL Server, on the other hand, provides enterprise-grade high availability, seamless integration with Microsoft and Azure environments, and significant advantages for mixed OLTP and analytical workloads through features like columnstore indexes. Despite their strengths, both databases face challenges with large-scale analytics, where a dedicated analytical solution like ClickHouse may be more effective. Tinybird offers a managed solution by connecting Postgres and SQL Server to ClickHouse, enabling real-time analytics without complex infrastructure management, allowing teams to focus on writing SQL and deploying features efficiently.

Games generate a vast amount of event data, requiring robust analytics systems to process and interpret this information in real-time, particularly for live operations and player behavior analysis. ClickHouse, with its columnar storage and vectorized execution, is favored for its ability to efficiently handle high-cardinality event streams and billion-row queries, making it suitable for tasks like real-time leaderboards and player analytics in game studios. The text explores various aspects of game analytics, including telemetry ingestion, player lifecycle, economy analytics, and matchmaking, emphasizing the importance of pre-aggregated rollup tables for visualization and the use of LowCardinality columns to optimize query speed. Tinybird, a managed ClickHouse service, offers a streaming Events API that facilitates real-time data ingestion and query processing without the need for complex infrastructure management, allowing game studios to focus on feature development while maintaining efficient and scalable analytics.

The comparison between Postgres and MongoDB often sparks debate, focusing on their operational capabilities and suitability depending on factors like data structure, team expertise, and project requirements. Postgres, a relational database, excels in scenarios requiring consistent data structure, multi-table joins, strict data integrity, and SQL-based analytics due to its inherent schema enforcement and ACID-compliant transactions. On the other hand, MongoDB, as a document database, is advantageous for applications requiring flexible, variable document structures, deep nesting, and rapid schema changes, especially during prototyping or when horizontal scaling is needed at extreme write scales. However, neither database is optimized for large-scale analytical queries, as both Postgres and MongoDB are designed for OLTP rather than OLAP workloads. To address this, teams often integrate with OLAP databases like ClickHouse, which efficiently handles analytical queries by separating storage and processing of data, thus allowing real-time analytics without overburdening operational databases. Tinybird facilitates the integration by providing a managed ClickHouse service that seamlessly connects with Postgres and MongoDB, using change data capture to keep data fresh and enabling fast, scalable analytics through HTTP endpoints.

ClickHouse is specifically designed to handle the data latency and high-throughput requirements of financial applications, addressing critical issues such as fraud detection, compliance, and risk scoring in real-time. It supports schema patterns optimal for append-only financial event data, allowing for efficient query processing over billions of rows, which is essential for neobanks, payment processors, and trading platforms. ClickHouse's capabilities include executing in-database fraud detection queries, performing real-time anomaly detection using Z-scores, and supporting compliance and audit queries through its columnar storage. It also accommodates data retention needs across various regulatory requirements by employing TTL rules for data storage tiers. Tinybird, a managed ClickHouse service, offers a platform for building real-time analytics by turning complex SQL queries into parameterized HTTP endpoints, ensuring low-latency responses, and eliminating the need for custom consumer code or separate stream processing layers. This integration allows fintech applications to perform real-time analytics, fraud detection, and compliance reporting efficiently, with Tinybird handling the infrastructure, scaling, and SQL endpoint management, enabling teams to focus solely on SQL development.

Postgres and MySQL, the two leading open-source relational databases, each have their own strengths and weaknesses depending on specific OLTP requirements, with the choice between them often influenced by factors like SQL compliance, ecosystem maturity, and extension support. Both databases are mature and widely supported, excelling in ACID transactions, standard SQL with vendor extensions, and offering cloud-managed options. Postgres stands out for its SQL compliance, advanced querying capabilities, and a robust extension ecosystem, making it ideal for complex queries, JSON workloads, and geospatial data through extensions like PostGIS. MySQL, on the other hand, excels in replication and high availability, speed for simple queries, ease of operations, and full-text search, benefiting from a longer track record as a managed service. Both databases, however, hit limitations when handling analytics at scale, as they are primarily designed for OLTP workloads, leading to inefficiencies in large-scale analytical queries. For analytics, the common solution is to use a columnar OLAP database like ClickHouse, which integrates with Postgres and MySQL through change data capture (CDC) or event streaming, providing a scalable and efficient analytical layer. Tinybird offers a managed ClickHouse service that simplifies real-time analytics integration with Postgres and MySQL, providing low-latency analytical queries through HTTP endpoints without requiring extensive infrastructure changes.

Ecommerce analytics often involve handling massive event volumes and require real-time querying capabilities that traditional databases like Postgres and MySQL struggle to support, especially for analytical queries. ClickHouse®, with its ability to manage large-scale data ingestion and execute sub-second analytical queries, is well-suited for ecommerce data infrastructure. The text emphasizes the importance of separating behavioral data from transactional data, suggesting that orders should be handled by OLTP databases while behavioral events, such as product impressions and operational dashboards, should be managed by ClickHouse. It discusses various schema patterns and queries essential for ecommerce, including conversion funnel analysis, real-time inventory tracking, customer segmentation, A/B testing, and in-product analytics. Additionally, it highlights the use of ClickHouse for real-time product performance analysis, customer behavior segmentation, and inventory management. Tinybird, a managed ClickHouse solution with streaming ingestion and SQL endpoints, is proposed as a way to facilitate real-time ecommerce analytics without extensive infrastructure, allowing for efficient data ingestion and querying for both internal dashboards and customer-facing features.

Product analytics involves querying event logs to understand user actions, sequences, frequency, and session termination, with ClickHouse® serving as a robust database engine supporting major platforms like Mixpanel, Amplitude, and PostHog due to its columnar storage and efficient processing capabilities. The text discusses best practices for designing event tracking schemas using ClickHouse, particularly emphasizing the use of the MergeTree engine and JSON properties to handle flexible event data. It outlines methodologies for session analysis, including pre-computation in materialized views for efficiency, as well as conducting funnel and retention analyses to track user progression and return rates, respectively. The document also covers calculating active user metrics such as DAU, WAU, and MAU, underscoring the advantage of pre-aggregating metrics for real-time dashboard queries and real-time alerting through materialized views. Finally, it highlights the use of Tinybird for managing ClickHouse, allowing rapid event ingestion and exposure of analytics as real-time API endpoints, ensuring low-latency data queries essential for real-time applications and product feature development.

ClickHouse is a powerful columnar database designed to optimize data science workflows by addressing database performance constraints rather than algorithmic ones. It is particularly suited for analytical queries over large datasets, enabling efficient data processing and feature computation for machine learning models. The integration of ClickHouse with Python is facilitated by two main clients, clickhouse-driver and clickhouse-connect, which support operations in Jupyter notebooks and production pipelines, respectively. ClickHouse offers a range of built-in statistical functions, allowing data scientists to perform complex analyses like correlation, regression, and A/B testing directly in the database, saving time and computational resources. Feature engineering is significantly enhanced by ClickHouse's ability to compute features at query time from raw data, which can be exposed through parameterized endpoints using tools like Tinybird. Furthermore, ClickHouse's SAMPLE clause allows for efficient exploratory data analysis by providing representative subsets of data, while its window functions and cohort analysis capabilities support sophisticated time series and retention analyses. For those seeking managed services, Tinybird offers a platform that leverages ClickHouse's capabilities with a Python SDK and HTTP API, enabling seamless integration with data science workflows and rapid iteration from development to production.

ClickHouse is a database system known for its speed, which is contingent on the proper design of schema and queries. Key performance improvements in ClickHouse are primarily achieved through thoughtful schema decisions during table creation, particularly by optimizing the sort key and table engine. The text outlines several impactful strategies to enhance query speed, such as designing an appropriate sort key, using LowCardinality for dimension columns, avoiding functions on sort key columns, utilizing the PREWHERE clause for selective filters, and replacing COUNT(DISTINCT) with the uniq() function for faster approximations. Pre-aggregating data with materialized views also significantly reduces query load, especially for dashboards. Diagnostic tools like system.query_log, EXPLAIN PIPELINE, and Tinybird's managed ClickHouse service can help diagnose and optimize query performance. The article emphasizes avoiding common pitfalls associated with BI tools generating inefficient SQL queries and suggests using dedicated views or rollup tables to maintain query efficiency.

The restructured Tinybird documentation has been revamped to align more closely with the developer's workflow, featuring a streamlined sidebar with a maximum of two levels of indentation to simplify navigation. This new organization focuses on core areas like quickstart, development workflow, core concepts, actions, and guides, each designed to reflect the user's journey through the product. The documentation now integrates SDK examples for TypeScript and Python alongside CLI and HTTP calls within relevant sections, eliminating the need for a separate reference appendix. Visual enhancements have been made, including a higher-contrast dark mode and a refreshed navigation system that minimizes distractions. The previous Classic/Forward toggle has been replaced with a method selector for Tinybird CLI, TypeScript SDK, and Python SDK, allowing users to choose their preferred development path and maintain consistency across documentation. The changes aim to provide a more intuitive, efficient user experience, while the Classic docs remain available for those who still use them.

Tinybird's Birdwatcher platform represents a significant shift in how the company leverages its internal data for analytics and growth insights, transitioning from relying on external tools to using its own infrastructure. Initially developed as an internal observability tool for various departments, Birdwatcher evolved to integrate growth analytics by building dashboards, reports, and an agent on top of the operational data already present within Tinybird's ecosystem. This integration allowed the platform to answer complex questions spanning multiple data sources, such as product usage and user behavior, without the need to duplicate data into external analytics silos. By adding an agent that operates through Slack, Birdwatcher became more accessible to non-technical staff, enabling teams to query data and gain insights without needing SQL or specific dashboard knowledge. The platform now serves as a comprehensive interface connecting operational and growth data, allowing different teams to ask diverse questions within the same system. This approach not only improved internal analytics capabilities but also provided Tinybird with valuable insights into enhancing its own analytics product for customers.