January 2026 Summaries

Since 2019, MongoDB has supported distributed, multi-document transactions, enabling ACID compliance and cross-shard transactions with snapshot isolation. The text explores how MongoDB's distributed transactions protocol operates, emphasizing recent advancements in formal modeling and verification using compositional TLA+ specifications to ensure correctness and high-level isolation guarantees. The protocol leverages a multi-version, transactional key-value storage engine called WiredTiger and employs a two-phase commit process across sharded clusters to manage transactions. Notably, the text introduces the concept of "permissiveness," a metric assessing a protocol's efficiency in implementing isolation levels, allowing for a nuanced analysis beyond binary correctness. By using TLA+ specifications, MongoDB validates the protocol's isolation guarantees and explores potential optimizations to enhance concurrency. This rigorous approach ensures developers that MongoDB’s transactions are not only correct but also efficient, impacting throughput and system availability positively. The post serves as a companion to a more technical exploration in a VLDB ’25 paper and highlights the utility of formal methods in both correctness verification and performance analysis.

JSON's prevalence in modern applications has led to questions about whether a single database can handle both relational and document-style workloads, with PostgreSQL's JSONB and MongoDB's BSON often being compared. Despite their apparent similarities, these systems differ significantly in their internal storage models and execution paths, particularly under realistic workloads involving frequent updates and concurrency. PostgreSQL's JSONB stores JSON-like objects in a binary format within relational tables, offering flexibility but potentially experiencing performance bottlenecks due to its MVCC model during heavy update workloads. In contrast, MongoDB's BSON, integral to its document-oriented design, supports efficient field-level updates without reconstructing full documents, maintaining stable performance under similar conditions. Controlled experiments reveal that while MongoDB sustains a high rate of updates with steady CPU usage, PostgreSQL's throughput declines over time as CPU utilization increases, highlighting the influence of architectural choices on performance under sustained update pressure. Both databases offer mechanisms for managing JSON-style data, but their performance differences are rooted in the architectural trade-offs of update handling.

As search systems evolve to accommodate natural language queries and multimodal data, a single retrieval strategy is insufficient to capture the diverse user intents. This tutorial demonstrates building an agentic video search system using MongoDB's capabilities and Voyage AI's voyage-multimodal-3.5 model, which supports text, images, and videos. It focuses on implementing a system that searches cooking videos, adapting its strategy based on the query type, whether visual or textual. The system utilizes a large language model (LLM) as a router to decide between vector and hybrid search methods, based on the user's query. The vector search examines joint embeddings of video segments and captions, while the hybrid search combines vector search with full-text search on captions. The tutorial guides through setting up necessary libraries, embedding video segments, creating search indexes, and defining search functions, culminating in an adaptive search architecture that highlights the potential of the LLM-as-a-router design pattern in handling diverse queries.

Automated Embedding in MongoDB Vector Search introduces a new feature designed to simplify the building of AI-powered applications by integrating seamless vector search capabilities directly into MongoDB. This innovation, now in public preview, is bolstered by the acquisition of Voyage AI, which enhances MongoDB's offerings with state-of-the-art embedding models. By automating the generation of vector embeddings, MongoDB addresses previous challenges such as manual embedding generation, synchronization overheads, and complex API management, thereby reducing development complexity and operational overhead. This streamlined process allows developers to focus on application functionality rather than embedding logistics, offering benefits such as improved retrieval speed and relevance. The feature supports a variety of use cases, from generative AI to e-commerce and content management, and provides a low-friction path for enterprises and startups alike to adopt AI-driven capabilities. Additionally, MongoDB's integrated approach eliminates the need for multiple systems and external models, enhancing performance and reliability while future-proofing AI applications through easy model lifecycle management.

MongoDB Atlas Resource Policies provide a structured approach to enforce consistent database configurations and standards across an organization, addressing common issues like configuration drift and inefficient resource allocation. By implementing these policies, organizations can prevent scenarios such as overprovisioned storage, inappropriate cluster topology choices, and under-sharded cluster configurations, which can lead to unnecessary costs and operational complexity. The policies allow for the specification of disk size limits, cluster topologies, and minimum shard counts, ensuring that resources are optimized and aligned with organizational goals. Enforcing these standards from the outset not only aids in accurate capacity planning and cost management but also reduces the need for costly migrations and operational overhead by maintaining a standardized infrastructure. Through a combination of these policies, organizations can create a comprehensive governance framework that transforms database provisioning from a reactive to a proactive process, ultimately leading to a more efficient and predictable database environment.

AWS Kinesis Data Streams now seamlessly integrate with MongoDB Atlas Stream Processing, eliminating the need for custom code and ongoing maintenance to connect these platforms for data streaming. This integration allows users to capture events from various sources like applications, IoT devices, and logs, and process them in MongoDB Atlas for operational queries, while also enabling the enrichment of Kinesis streams with data from MongoDB collections. With this enhancement, users can build transformation pipelines that read from Kinesis, enrich the data with MongoDB Atlas, and write back to Kinesis, facilitating low-latency insights and complex data pipeline construction. The integration supports advanced security features such as IAM Assume Role and PrivateLink endpoints, ensuring secure, intra-network connections. This development simplifies the process of ingesting, validating, and transforming event streams between Kinesis and MongoDB Atlas, paving the way for advanced analytics and application queries without the necessity of custom integration solutions.

In 2025, the European manufacturing sector faced numerous high-profile cybersecurity incidents, with manufacturing becoming the primary global target for ransomware due to the critical nature of industrial operations. As manufacturers increasingly rely on real-time data from their factory floors for strategic decision-making, MongoDB has emerged as a crucial infrastructure, offering flexibility and seamless management of diverse data types. To address the unique challenges of industrial environments, MongoDB collaborates with HackRTU to enhance security and performance, adhering to standards such as IEC 62443. The blog post by HackRTU co-founders Aarón Flecha Menéndez and Víctor Bello Cuevas outlines key strategies for leveraging MongoDB's security features, including access control, network restrictions, and encryption, to improve resilience against threats. They emphasize the importance of a security-by-design approach, focusing on availability, performance, and integrity to protect critical operational data, with MongoDB proving to be a trusted platform for manufacturers like Dongwha. The ongoing evolution of the industry, particularly with the rise of AI and cloud systems, underscores the need for robust security measures to mitigate cyber risks and support innovation.

The Teach & Learn blog series highlights how educators worldwide are integrating MongoDB into their curricula to revolutionize education and prepare tech professionals. One featured educator is Dr. Mahesh Chaudhari from the University of San Francisco, who integrates MongoDB into the Master of Science in Data Science program, emphasizing hands-on practice and real-world applications. He incorporates MongoDB in courses on relational and distributed data systems, providing students with the practical skills needed for careers in data science, AI, and data engineering. MongoDB's support, through resources like Atlas credits and technical guidance, enhances the learning experience, preparing students for industry roles and enabling them to work on impactful projects. Dr. Chaudhari's efforts have led to successful student projects, some of which resulted in published conference papers, showcasing the effectiveness of MongoDB in building robust data platforms. The MongoDB for Educators program is recommended for other educators seeking to integrate MongoDB into their teaching to equip students with in-demand skills.