redis

Bilingual

<antirez>
2 months ago
- Redis creator announces Gopher protocol implementation for its 10th anniversary.
- Gopher is a text-only protocol that emphasizes content without modern web tracking, comments, or heavy scripting.
- Gopher is intentionally 'uncool' to provide an alternative space for focused, long-lasting content.
- Implementation is only 100 lines of code, using Redis inline protocol for requests starting with '/' or empty.
- Content is served from Redis keys via Gopher; normal Redis commands remain functional.
- Authoring tool 'gopher2redis' is provided to create Gopher holes easily.
- Example Gopher hole available at gopher://gopher.antirez.com.
- To enable, use Redis unstable branch with 'gopher-enabled yes' and password protect the instance.
- Creator hopes to foster a community outside the chaos of the modern Internet.
<antirez>
2 months ago
- Maintaining a popular open source project is psychologically taxing due to the overwhelming volume of messages, issues, and pull requests—a phenomenon called the 'flood effect'.
- Attempts to manage the flood with 'inbox zero' tactics often fail because they don't allow for the thoughtful processing that community feedback deserves.
- The maintainer's role shifts from designing and coding to mostly reviewing contributions, which reduces time for creative work and leads to fewer organic new features.
- Working continuously on the same project for years is mentally draining; the author used to work in bursts and found that forced daily schedules reduce productivity.
- There is constant fear of losing technological leadership due to misalignment between the maintainer's vision and user expectations.
- Dealing with difficult people (assholes) in the community adds significant stress.
- A sense of futility can arise from the belief that software is ephemeral and will be replaced, contrasting with the desire to work on 'big ideas'.
<antirez>
2 months ago
- Redis 6.0.0 stable includes SSL, ACLs, RESP3, client side caching, threaded I/O, diskless replication, cluster support, and more.
- Client side caching was redesigned: caching slot approach replaced by key names, broadcasting mode added, and opt-in/opt-out support.
- RDB files used for replication can be immediately deleted if no longer useful.
- ACL improvements: new ACL LOG command for debugging, and ACL GENPASS reimplemented using SHA256 HMAC.
- PSYNC2 replication protocol improved for more frequent partial resynchronization.
- Timeout commands now accept decimal numbers and have better resolution.
- RDB loading is 20-30% faster, and INFO command is faster with many clients.
- New STRALGO command implements LCS (longest common subsequence) algorithm.
- Redis 6 is the biggest release ever; users are advised to test carefully before production use.
<antirez>
2 months ago
- Redis was born in a playful, problem-solving environment during the creator's early web 2.0 career.
- Over time, the creator's role shifted from building Redis to maintaining it for reliability, security, and community contributions.
- The creator views coding as artistic expression and prefers to be remembered as a 'bad artist' rather than a 'good programmer'.
- Stepping back as Redis maintainer to focus on idea generation at Redis Labs and personal freedom, with no fixed plans.
- Yossi Gottlieb and Oran Agra will take over as maintainers, tasked with shaping a sustainable development model.
- The creator expresses gratitude to the community, companies (Redis Labs, VMware, Pivotal), and hopes his alternative vision for software development endures.
- Future plans include exploring hobbies, writing blog posts, and creating educational videos in Italian.
<antirez>
2 months ago
- The author left Redis 4.44 years ago and avoided all related work, but returned after feeling a lack of purpose and concern over community fragmentation.
- They contacted Redis CEO Rowan Trollope for a part-time role as a 'bridge' between the company and community, focusing on demos, design, and documentation.
- The author argues the license switch from BSD to SSPL is understandable, as it prevents cloud providers from using Redis without revenue sharing.
- They see vector capabilities as a natural fit for Redis, introducing 'Vector Sets' (e.g., VSIM command) for K-NN searches using embeddings.
- The author emphasizes Redis as 'Lego for programmers,' leaving design flexibility to users rather than enforcing rigid patterns.
- AI tools like Claude have helped the author improve coding and writing efficiency, and they believe vector search is crucial for Redis's future.
<antirez>
2 months ago
- The author rejoined Redis and advocated for a switch from SSPL to AGPL license, finding the SSPL was not accepted by the OSI and the open source community.
- The author wanted the new Vector Sets data type to be released under an open source license, driven by a deep-rooted commitment to open source software.
- Returning to an open source license like AGPLv3 is seen as essential for Redis's coherence, user acceptance, and contribution to the broader open source community.
- The author expresses happiness that Redis is now open source under AGPLv3, and commits to writing high-quality code for Vector Sets and other improvements.
- Redis 8, the first version under the new AGPL license, was released as GA with many new features and speed improvements.
<antirez>
2 months ago
- The author went for a 4-hour walk on Mount Etna on May 1st, Italy's workers holiday, and enjoys walking to clear his mind while coding.
- He released a blog post about switching Redis to the AGPL license and learned that license understanding and community trust matter more than just legal clauses.
- He was touched by Simon Willison's similar perspective on open source, highlighting how shared values in the community drive open source success.
- Despite his early Linux contributions, the author moved away from Linux due to fragmentation and binary compatibility issues, favoring zero-dependency software distribution.
- He believes distributions should offer both Redis and ValKey as they have diverged, allowing users to choose based on features or licensing policies.
- The author appreciates the kindness of the community, despite some harsh critiques about license decisions.
- He notes that many users are indifferent to license debates as long as software works, emphasizing that practical use often outweighs licensing expertise.
<antirez>
2 months ago
- The author took a break from HNSW development to write a blog post sharing advanced findings, especially on making HNSWs fast enough for a low-latency, high-performance system like Redis.
- The original HNSW paper lacks key details like deletion support and the necessity of multiple layers; the author modified HNSW to support actual removal of entries and questions if the 'H' (hierarchical) aspect is always beneficial.
- Memory scaling is addressed by using 8-bit quantization per vector (computing max absolute value) which gives ~4x speedup, 4x reduction in vector size, and almost no recall loss; full precision and binary quantization options are also available.
- For speed, threads are used for reads and partial writes; a per-node epoch array prevents revisiting visited nodes during concurrent searches, and writes split into a reading-half and commit-half to minimize lock contention.
- To reclaim memory on deletion, links are forced to be bidirectional, and a distance matrix is computed among neighbors of a deleted node to reconnect them greedily, preserving graph quality.
- HNSWs are exposed as a Redis data structure (Vector Sets) with commands VADD, VREM, VSIM, allowing composability across keys/instances and per-user/per-item indexing; this also simplifies scaling by hashing elements modulo N.
- Loading time is accelerated by serializing the entire graph structure (nodes and neighbor pointers) so that rebuilding is allocation only, achieving a 100x speedup; security checks ensure validity even with corrupted RDB files.
- A JSON filter mechanism allows mixed search (e.g., vector similarity plus metadata constraints) within the greedy search loop, letting users specify effort and filter by JSON attributes.
- In-memory HNSWs are acceptable for many use cases (e.g., 3M Word2Vec entries take 3GB RAM); the author believes HNSWs will remain relevant and encourages further research.
Redis Patterns for Coding Agents
3 months ago
- Optimized website for developers and AI agents with Redis patterns, practices, and commands.
- Human-friendly browsing with clickable sections and machine-readable index (llms.txt).
- AI agents start by fetching llms.txt, then individual .md files with Markdown and Redis commands.
- Local mirror of official Redis documentation including command index and detailed docs.
- Core architectural patterns, community-developed use cases, and real-world scale examples available.
We put a Redis server inside our runtime
3 months ago
- Encore enables running identical backend code across local development, testing, and production environments, with infrastructure declared directly in the code and provisioned per environment.
- For local development, Encore's runtime includes an in-memory Redis server built directly into the process, avoiding external dependencies like Docker containers or mocks that might behave differently.
- Originally implemented in Go using alicebob/miniredis for Go applications, Encore ported this to Rust to embed within the Rust runtime for TypeScript apps, totaling around 25,000 lines and supporting key Redis data types and protocols.
- The embedded server mimics real Redis behavior, including time-based expiry via a mock clock and key pruning to prevent unlimited growth, ensuring local development closely matches production behavior.
- Applications declare caches in code, and the runtime automatically uses the embedded server locally or connects to a managed Redis in production based on configuration flags, with no changes needed in application code.
- To ensure parity, the Rust server is tested against the original Go miniredis suite, comparing raw RESP responses byte-for-byte to catch subtle differences in command handling or TLS implementation.
- The in-memory server is intended only for development and testing; in production, Encore provisions scalable managed Redis instances, maintaining consistency without requiring manual infrastructure setup locally.

redis

<antirez>

<antirez>

<antirez>

<antirez>

<antirez>

<antirez>

<antirez>

<antirez>

Redis Patterns for Coding Agents

We put a Redis server inside our runtime