Redis
Redis (Remote Dictionary Server) is an open-source, in-memory key-value data store supporting multiple data structures including strings, lists, hashes, sets, and sorted sets. It is commonly used for caching, session management, message queuing, and real-time leaderboards. Redis is known for high performance and low latency, and supports data persistence, primary-replica replication, and cluster modes. It is widely adopted in internet backend architectures to improve system response speed and throughput.
Core Facts
Agent的记忆系统通常分为三个层次：工作记忆（当前对话上下文窗口）、短期记忆（近期交互历史）和长期记忆（通过向量数据库持久化存储）
Dify's core architecture adopts a microservices design, using Docker Compose to decouple and deploy components such as the API service, Worker, Web frontend, PostgreSQL, Redis, and vector database.
Enterprise-grade Agent systems typically introduce external memory modules (such as vector databases) to store long-term memory, in order to overcome context window limitations.
Timeline (last 90 days)
A Redis cluster always uses a fixed 16384 hash slots, and the modulus operand is fixed at 16384 rather than the current number of servers.
The hash tag mechanism at the Redis protocol level wraps the part of the key that needs to share a slot in curly braces, so that the hash function only computes the content inside the braces, thereby ensuring that all keys of the same user land in the same slot.
Dragonfly is a drop-in replacement for the Redis protocol; with an ordinary ioredis client, migration only requires changing the connection URL.
Slot assignment in a Redis cluster uses the CRC16 hash function, not the SHA family.
The comparison data cited by the blogger shows that Dragonfly's QPS reaches 25 times that of Redis.
Pipeline is a batch submission mechanism at the Redis protocol level that writes multiple commands to the TCP connection at once, and the server executes them in order and then packages and returns the results, but it is not equivalent to a transaction and does not guarantee atomicity.
Idempotent design ensures that repeated calls with the same request ID return the same result without repeatedly deducting inventory or generating multiple orders; common implementations include database unique constraints, Redis deduplication keys (such as SETNX) and request fingerprint comparison.
Common Checkpointer implementations include InMemorySaver (valid only within the process, suitable for testing) and persistence implementations based on SQLite/Postgres/Redis (suitable for production).
In multi-instance deployments, the nonce cache requires shared storage (such as Redis) or sticky routing, otherwise the same frame may be accepted once on each of different instances, creating a protection gap.
durable-harness writes to the World interface, so the same agent can migrate from one World to another without code changes: use Vercel World on Vercel, local files during development, and Postgres, MongoDB, Redis, or Turso when self-hosting.
40 more timeline events
All Facts (20)
Agent的记忆系统通常分为三个层次：工作记忆（当前对话上下文窗口）、短期记忆（近期交互历史）和长期记忆（通过向量数据库持久化存储）
80%VerifiedDify's core architecture adopts a microservices design, using Docker Compose to decouple and deploy components such as the API service, Worker, Web frontend, PostgreSQL, Redis, and vector database.
80%VerifiedEnterprise-grade Agent systems typically introduce external memory modules (such as vector databases) to store long-term memory, in order to overcome context window limitations.
75%VerifiedThe WebSocket protocol establishes a persistent bidirectional communication channel through which the server can proactively push data to the client, with latency as low as milliseconds.
70%VerifiedMongoDB is a document-oriented NoSQL database that stores data in a JSON-like BSON format.
70%VerifiedSession Service manages short-term memory at the conversation level (message history, tool call results, and intermediate reasoning states), while Memory Bank provides long-term memory capabilities across sessions.
65%VerifiedIn an Agent framework, Checkpointing serializes the current execution state and persists it to external storage (a database or Redis), while suspending the execution thread and resuming after human input is received.
65%VerifiedRAGFlow depends on multiple services including Elasticsearch (full-text search), Redis (caching), and MySQL (metadata storage)
65%VerifiedSalvatore Sanfilippo created the Redis project in 2009, originally to solve performance problems he encountered while developing the real-time web analytics system LLOOGG.
65%VerifiedAll Redis commands are executed serially in a single main thread, using a single-threaded event-driven architecture.
65%VerifiedRAGFlow is deployed based on Docker containerization and runs five containers upon startup: Redis, MinIO, MySQL, Elasticsearch, and RAGFlow Server.
65%VerifiedCelery's core architecture consists of three components—Producer, Broker, and Worker—where the Broker typically uses Redis or RabbitMQ.
65%VerifiedLangGraph's long-term storage supports multiple backends, including PostgreSQL for structured data, vector databases such as Pinecone and Chroma for semantic retrieval, and Redis for high-speed caching.
65%Unverifiedantirez stepped down as Redis maintainer in 2020
90%Unverified防止重复扣费的幂等性实现方案包括数据库唯一约束、状态机检查、Redis分布式锁和乐观锁版本号
90%UnverifiedRedis 6.0后引入了I/O多线程，但命令执行仍然是单线程的
90%UnverifiedRedis的SCAN命令使用反向二进制迭代（reverse binary iteration）的游标推进算法
90%UnverifiedRedis 7.0后将紧凑内存编码格式统一为listpack
90%UnverifiedRedis在4.0版本（2017年）引入了模块系统（Modules API），允许第三方通过动态加载方式扩展自定义数据类型
90%UnverifiedRedis此前在KEYS和SCAN命令中使用的是简单的glob风格模式匹配，仅支持*、?、[]等通配符
90%