CockroachDB Engineering engineering blog

Isaac Wong examines how orbital data centers would reshape distributed database design, covering moving topologies, Raft consensus, latency, replica placement, storage economics, node failure, and data sovereignty. The post identifies practical capabilities future orbital-aware databases would need, including dynamic localities and autonomous decommissioning.

Alejandro Infanzon demonstrates image-to-image vehicle search using CLIP embeddings, Python, and CockroachDB. The walkthrough shows how to combine vector similarity with relational filters, explore embeddings with PCA and K-means, and run similarity queries alongside metadata in SQL.

CockroachDB engineers explain how Pebble’s value separation stores large values in blob files outside SSTables, reducing compaction read and write amplification. The post details blob-file formats, locality and space-amplification trade-offs, rewrite compactions, and benchmark results showing substantial throughput gains.

CockroachDB engineers describe C-SPANN, a distributed vector-indexing design for billions of fresh, searchable embeddings. The post explains its hierarchical K-means layout, incremental partition maintenance, RaBitQ quantization, reranking, tenant-aware indexing, and multi-region locality trade-offs.

Jessie Lin explains how Retrieval Augmented Generation grounds LLM responses in current, domain-specific data and demonstrates a knowledge-base chatbot built with CockroachDB. The tutorial covers vector search, unified storage for source and embedding data, distributed scalability, consistency, and access control.

Cockroach Labs introduces “Performance under Adversity,” a benchmarking methodology that evaluates database behavior during failures such as disk stalls, network partitions, node restarts, and regional outages. The post explains how admission control, lease architecture, range rebalancing, and multi-disk WAL failover help CockroachDB preserve throughput, latency, and consistency under operational stress.

CockroachDB 25.2 delivers a reported 50% average throughput increase, faster restores, and preview buffered writes that reduce transaction round trips. The release also adds distributed C-SPANN vector indexing, row-level security, enhanced observability, and new cloud deployment options.

CockroachDB introduces a database benchmarking methodology that measures sustained performance and recovery during realistic failures, including disk stalls, network partitions, node restarts, zone outages, and regional outages. The post explains its seven-phase test design, throughput and latency metrics, and reproducible setup for evaluating database resilience under adversity.

Harsh Shah presents a deployment architecture for running CockroachDB on Red Hat OpenShift Virtualization, covering VM placement, networking, storage, service discovery, security, and cluster initialization. The post also reports high-availability and TPC-C-style validation results, with guidance for monitoring, upgrades, and VMware migration planning.

Cockroach Labs explains how MOLT Verify validates database migrations through schema checks, row-level comparisons, streaming batches, and live-data retries. The post covers its architecture, concurrency controls, filtering options, and operational best practices for verifying large migrations before cutover.