Object Storage

4 posts · Nov 2019 – Feb 2026

What this trail covers

Cloudflare R2 object storage now offers Local Uploads in open beta, allowing object data to be written to a storage location close to the client first, then asynchronously copied to the bucket's region. This significantly reduces upload request duration (TTLB) by up to 75% for cross-region uploads. The feature utilizes Cloudflare Queues for asynchronous replication tasks and a pull model for queue consumers. Data remains strongly consistent and immediately accessible after the initial write. Local Uploads are now generally available.

All Cloudflare trails

Feb 2026Improve global upload performance with R2 Local UploadsIntroduces 'Local Uploads' for R2 object storage, enabling writes to a client-local storage location before asynchronous replication to the bucket's primary region. This feature reduces upload Time to Last Byte (TTLB) by up to 75% for cross-region uploads. It leverages Cloudflare Queues for managing replication tasks and a pull model for queue consumers. The implementation atomically stores object metadata, creates pending replica keys, and schedules replication tasks. It also details the architecture of R2 Gateway Worker, Durable Object Metadata Service, and Distributed Storage Infrastructure in the context of Local Uploads.
Aug 2025Redesigning Workers KV for increased availability and faster performanceThis post details the redesign of Cloudflare Workers KV to address availability and performance issues stemming from its reliance on third-party object storage providers. The new architecture uses a distributed database for small objects (hundreds of billions, <1KB) and Cloudflare R2 for larger objects (up to 25 MiB), creating a hybrid storage solution. It also reintroduces dual-provider capabilities by racing writes to both the new database backend and a restored third-party provider, with background reconciliation for failed writes. The KV Storage Proxy (KVSP) was built to bridge HTTP communication from Workers to the database clusters, handling sharding and routing. This redesign eliminates single points of failure and improves performance.
Apr 2025Making Super Slurper 5x faster with Workers, Durable Objects, and QueuesThis post details the rearchitecture of Cloudflare's Super Slurper data migration tool using Cloudflare Workers, Durable Objects, and Queues. The new architecture addresses performance bottlenecks of the previous Kubernetes-based system by distributing compute and bandwidth across the global network. Key improvements include using Queues for auto-scaling asynchronous object transfers, Workers for lightweight, globally distributed compute, SQLite-backed Durable Objects for distributed state management (handling billions of objects with sharding and cleanup), and Hyperdrive for accessing legacy PostgreSQL data. This resulted in up to a 5x performance improvement for large-scale data migrations to R2. The post also details strategies for handling duplicate messages and failures.
Oct 2024Building Vectorize, a distributed vector database, on Cloudflare’s Developer PlatformThis post details the architecture and implementation of Vectorize, a distributed vector database built on Cloudflare's Developer Platform. It explains the challenges of vector databases, the use of Inverted File Index (IVF) for pruning search space, Product Quantization (PQ) for vector compression, and the trade-offs of approximate nearest neighbor search with result refinement. It also describes the eventual consistency model using R2 for storage, manifest files for versioning, and SQLite in Durable Objects for the write-ahead log (WAL) to coordinate distributed writes.
May 2022A New Hope for Object Storage: R2 enters open betaThis post announces the open beta of Cloudflare R2, an object storage solution. It details the two APIs (In-Worker and S3-compatible), authentication mechanisms (SigV4), core features (CRUD operations, object size limits, strong consistency), pricing model (storage and operation costs, free tier), and future plans for performance improvements, global distribution, and expanded features like TTLs and public buckets.
Advertisement
Jun 2020UtahFS: Encrypted File StorageThis post introduces UtahFS, a proof-of-concept end-to-end encrypted file storage system. It leverages object storage as the underlying layer and implements a data layer using linked allocation with skip lists for efficient partial file access. The encryption layer uses AES-GCM for confidentiality and Argon2 for password hashing. The integrity layer employs Merkle Trees for data verification. It also introduces Path ORAM to hide access patterns from the storage provider, defending against traffic analysis attacks.
Nov 2019What’s new with Workers KV?This post details several new features and enhancements for Cloudflare Workers KV: 1. Bulk upload and delete APIs for multiple key/value pairs via `api.cloudflare.com`, including support for base64 encoded binary values. 2. Key listing functionality via `list()` from within a Worker or via the API, with support for pagination using `cursor` and filtering by `prefix`. 3. Increased the limit of namespaces per account from 20 to 100. 4. Increased the maximum value size limit from 2MB to 10MB. 5. Introduced a KV browser in the dashboard for managing keys and values (add, edit, upload, download). 6. Enhanced Wrangler CLI integration for creating namespaces, putting/getting keys, and managing key/value pairs.