Aerospike (database)

Aerospike Database is a real-time, distributed key-value store engineered for high throughput and low latency on flash storage, keeping indexes in memory while serving data directly from solid-state drives.Aerospike Database 是为在高吞吐、低延迟下使用闪存的实时分布式键值存储，索引常驻内存、数据直接由固态盘提供服务。

It targets latency-sensitive workloads such as programmatic advertising, fraud detection and digital payments, where large datasets must be served at sub-millisecond speeds.它面向程序化广告、欺诈检测与数字支付等延迟敏感负载，这类场景要求以亚毫秒速度服务大数据集。

Features

• A hybrid-memory architecture places record indexes in DRAM or persistent memory while storing data on flash, balancing performance against the cost of large datasets.混合内存架构把记录索引放在内存或持久内存中，数据则存于闪存，在大数据集上权衡性能与成本。
• Clusters are homogeneous and masterless, with automatic partitioning that spreads data evenly across nodes and rebalances without operator commands.集群同构且无主，自动分区让数据均匀分布到各节点，再平衡无需运维指令。
• Every record is located by a 160-bit digest derived from its key, giving constant-time routing without central coordination.每条记录由键派生出的 160 位摘要定位，实现无需中心协调的恒定时间路由。
• An availability-first mode provides eventual consistency, while a strong-consistency mode enables transactional guarantees when applications demand them.可用性优先模式提供最终一致，强一致模式则在应用要求时提供事务性保证。
• Secondary indexes and predicate filtering narrow scans to matching records, supporting real-time filtering over very large namespaces.二级索引与谓词过滤把扫描收窄到匹配记录，支持在超大命名空间上进行实时过滤。
• Server-side Lua user-defined functions execute logic close to the data, avoiding round trips for simple aggregation and transformation tasks.服务端 Lua 用户定义函数贴近数据执行逻辑，为简单的聚合与转换任务省去往返开销。
• Records carry a configurable time-to-live and a generation counter used for optimistic concurrency, protecting updates from lost-write races.记录携带可配置生存时间与用于乐观并发的代际计数器，保护更新免于丢失写入的竞态。
• Cross-datacenter replication streams changes between clusters over TCP or TLS, supporting disaster recovery and geographically distributed deployments.跨数据中心复制通过 TCP 或 TLS 在集群间流式同步变更，支撑灾难恢复与地理分布式部署。
• Intelligent clients are cluster-aware, talking directly to the node holding each record, pooling connections and retrying transparently during reconfiguration.智能客户端感知集群，直接与持有记录的节点通信，池化连接并在重配置期间透明重试。
• Flash-optimised writes use sequential logging and write coalescing, so solid-state drives receive few large writes instead of many small ones.闪存优化写入采用顺序日志与写合并，让固态盘接收少量大写入而非大量小写入。
• The data model organises records into namespaces and sets, with per-record bins holding values of mixed types without a required schema.数据模型把记录组织进命名空间与集合，每条记录的 Bin 可保存混合类型的值而无强制结构。
• Connectors to Kafka, Spark and S3 plus a full monitoring surface integrate the store with existing data pipelines and operations.面向 Kafka、Spark 与 S3 的连接器以及完整监控面，让存储融入既有数据管道与运维。
• Security features such as TLS encryption, role-based access control and audit logging satisfy the compliance demands of financial deployments.TLS 加密、基于角色的访问控制与审计日志等安全特性满足金融部署的合规要求。
• A command-line tools suite and endpoint monitoring integrate with existing observability stacks, supporting operational workflows around the store.命令行工具套件与端点监控融入既有可观测性栈，支撑存储周边的运维工作流。

History

• The company behind the database, now named Aerospike, was founded in 2009 by engineers with backgrounds in database systems, originally operating under the name Citrusleaf.数据库背后的公司现名 Aerospike，由具有数据库系统背景的工程师于 2009 年创立，最初以 Citrusleaf 之名运营。
• The product shipped under the Citrusleaf name in 2010, and both the company and the database adopted the Aerospike name in August 2012.产品于 2010 年以 Citrusleaf 之名交付，公司与数据库于 2012 年 8 月双双改用 Aerospike 名称。
• Early adoption concentrated in real-time advertising, where programmatic bidding demanded very low latency over enormous user-profile datasets.早期采用集中在实时广告领域，程序化竞价要求在海量用户画像数据集上实现极低延迟。
• A strong-consistency mode arrived in 2017, giving enterprises an option between availability-oriented semantics and transactional guarantees.强一致模式于 2017 年到来，为企业提供在可用性导向语义与事务保证之间的选项。
• Enterprise releases added persistent-memory support in the late 2010s, extending the memory tier beyond DRAM for even larger index footprints.企业版在 2010 年代末加入持久内存支持，把内存层扩展到内存之外，容纳更大的索引体量。
• Container and cloud deployments matured in the early 2020s with official Docker images and Kubernetes tooling alongside a managed cloud offering.2020 年代初，官方 Docker 镜像与 Kubernetes 工具及托管云产品一同让容器与云部署走向成熟。
• The 8.x server line shipped through 2025, sustaining the hybrid-memory model while refining operations for very large flash-backed clusters.8.x 服务器系列在 2025 年间持续交付，在维持混合内存模型的同时，为超大型闪存集群打磨运维。
• Throughout its history the product kept a narrow, focused scope as a key-value engine, deliberately avoiding SQL and document-model features.纵观历史，产品始终把范围收窄在键值引擎上，刻意不提供 SQL 与文档模型功能。
• Managed cloud operation arrived in the early 2020s through a dedicated database-as-a-service product alongside the self-managed editions.2020 年代初，专用数据库即服务产品与自管版本一同带来托管云运营。
• Distinct community and enterprise editions share one server codebase, with the enterprise tier adding replication, security and management capabilities.社区版与企业版共享同一服务器代码库，企业版额外提供复制、安全与管理能力。

Reception

• The database became a recognised fixture in advertising technology, with reported users including major ad platforms and identity-resolution providers.该数据库在广告技术领域成为公认的常客，据报用户包括大型广告平台与身份解析服务商。
• Financial institutions and fraud teams adopted it for real-time decisioning, where sub-millisecond lookups over large historical datasets matter.金融机构与反欺诈团队为实时决策采用它，这类场景看重对大型历史数据的亚毫秒查询。
• Compared with in-memory databases, Aerospike offers larger working datasets at lower cost by keeping only indexes in memory and data on flash.与全内存数据库相比，Aerospike 仅把索引放内存、数据放闪存，从而以更低成本支撑更大的工作数据集。
• Vendor benchmarks advertise millions of operations per second with predictable latency, a claim reviewers generally confirm on properly sized clusters.厂商基准宣传每秒数百万操作且延迟可预测，评论者普遍确认在合理规模的集群上确实如此。
• Critics point to a steeper operational learning curve and a smaller community than broader ecosystems such as Cassandra or Redis enjoy.批评者指出其运维学习曲线更陡，社区规模也小于 Cassandra 或 Redis 这类更广的生态。
• The choice between availability-first and strong-consistency modes is frequently cited as a distinctive, production-relevant feature.在可用性优先与强一致模式之间的选择，常被引为该产品独特且贴近生产的功能。
• Academic and industry evaluations place Aerospike favourably for read-heavy real-time workloads, while noting that write-heavy patterns stress the flash path.学术与行业评测对读密集实时负载给出好评，同时指出写密集模式会给闪存路径带来压力。

Community

• Open-source client libraries cover Java, C, Python, Go, Node.js and other languages, with the server community edition available for self-hosted evaluation.开源客户端库覆盖 Java、C、Python、Go、Node.js 等语言，服务器社区版可供自托管评估。
• Documentation, release notes and the Aerospike Academy training platform provide structured learning for operators and developers.文档、发布说明与 Aerospike Academy 培训平台为运维者与开发者提供结构化学习。
• A community forum hosts deployment questions, while GitHub issues track bugs and feature requests for the open client code.社区论坛承载部署问题，GitHub 议题跟踪开源客户端代码的缺陷与功能请求。
• Official Docker images for community and enterprise editions, plus Helm charts, make container-based evaluation straightforward.社区版与企业版的官方 Docker 镜像以及 Helm 图表让容器化评估变得简单。
• Connector projects for Kafka, Spark and S3 tie the database into the surrounding big-data ecosystem with maintained integrations.面向 Kafka、Spark 与 S3 的连接器项目以受维护的集成，把数据库接入周边大数据生态。
• User events and developer meetups reinforce a smaller but specialised community centred on extreme-scale real-time workloads.用户活动与开发者聚会滋养着一个规模较小却专注极端规模实时负载的专门社区。
• A maintained set of example applications and benchmark suites demonstrate recommended patterns for throughput testing and capacity planning.受维护的示例应用与基准套件演示吞吐测试与容量规划的推荐模式。

Related software

• Redis holds data entirely in memory, trading dataset size and cost for simplicity, and it often sits beside Aerospike as a caching tier.Redis 把数据全部放在内存，以数据集规模与成本换取简单，常作为缓存层与 Aerospike 并置。
• Apache Cassandra favours wide-column flexibility and tunable consistency over raw single-record latency, suiting different web-scale workloads.Apache Cassandra 更看重宽列灵活性与可调一致性而非单记录原始延迟，适应不同的网络规模负载。
• Couchbase Server provides a richer multi-model surface with SQL-style queries, appealing when query language matters more than raw key-value speed.Couchbase Server 提供带 SQL 风格查询的更丰富多模型面，当查询语言比原始键值速度更重要时更具吸引力。
• Amazon DynamoDB offers managed key-value storage in the cloud, trading deployment control for operational simplicity in AWS-native environments.Amazon DynamoDB 在云端提供托管键值存储，以部署控制权换取 AWS 原生环境下的运维简单。
• FoundationDB supplies strongly consistent distributed transactions over a key-value model, suiting applications that need serializability above all.FoundationDB 在键值模型上提供强一致分布式事务，适合把可串行化放在首位的应用。
• Memcached remains a pure cache with no persistence, marking the simplest end of the key-value spectrum that Aerospike extends.Memcached 仍是无需持久化的纯缓存，处于 Aerospike 所延伸的键值谱系最简一端。
• TiKV, the storage layer of TiDB, offers another Raft-based distributed key-value engine, sharing the per-node efficiency goals with a different ecosystem.作为 TiDB 存储层的 TiKV 提供另一款基于 Raft 的分布式键值引擎，与 Aerospike 追求相近的单节点效率，但生态不同。
• Apache HBase suits very large row-oriented datasets on top of HDFS, occupying a heavier, batch-oriented corner of the distributed storage space.Apache HBase 适合 HDFS 之上超大的行式数据集，占据分布式存储空间里更重、更面向批处理的角落。
• Key-value caches such as Redis Cluster and Memcached remain popular for hot data, while Aerospike occupies the persistent, flash-optimised middle ground.Redis Cluster 与 Memcached 等键值缓存仍是热数据的流行选择，Aerospike 则占据持久化、闪存优化的中间地带。