CQRS 物化视图架构

Architecture

CQRS 物化视图架构图，多个投影器构建特定用途的读模型：订单摘要、客户仪表板、分析立方体和搜索索引，全部由单一事件流提供。该模板展示如何从相同的写入事件创建多个优化的查询视图，每个视图针对特定用例定制，读取延迟低于毫秒级。适合需要从单一数据源进行多样化查询模式的系统。

下载 .fz 文件

完整 FlowZap 代码

WriteModel { # Write Model
n1: circle label:"Incoming Command"
n2: rectangle label:"Command Handler"
n3: rectangle label:"Validate Against Aggregate"
n4: rectangle label:"Persist to Write Store"
n5: rectangle label:"Emit Domain Event"
n1.handle(right) -> n2.handle(left)
n2.handle(right) -> n3.handle(left)
n3.handle(right) -> n4.handle(left) [label="Normalized"]
n4.handle(right) -> n5.handle(left) [label="After Commit"]
n5.handle(bottom) -> EventBus.n6.handle(top) [label="Publish"]
}
EventBus { # Event Bus
n6: rectangle label:"Route Event"
n7: rectangle label:"Guarantee Ordering"
n8: rectangle label:"Deliver to Projectors"
n6.handle(right) -> n7.handle(left) [label="Partition Key"]
n7.handle(right) -> n8.handle(left)
n8.handle(bottom) -> Projectors.n9.handle(top) [label="Fan-Out"]
}
Projectors { # View Projectors
n9: rectangle label:"Order Summary Projector"
n10: rectangle label:"Customer Dashboard Projector"
n11: rectangle label:"Analytics Projector"
n12: rectangle label:"Search Index Projector"
n9.handle(bottom) -> Views.n13.handle(top) [label="Build View"]
n10.handle(bottom) -> Views.n14.handle(top) [label="Build View"]
n11.handle(bottom) -> Views.n15.handle(top) [label="Build View"]
n12.handle(bottom) -> Views.n16.handle(top) [label="Build View"]
}
Views { # Materialized Views
n13: rectangle label:"Order Summary Table"
n14: rectangle label:"Customer Dashboard Cache"
n15: rectangle label:"Analytics Cube"
n16: rectangle label:"Elasticsearch Index"
n17: rectangle label:"Query API"
n18: circle label:"Fast Read Response"
n13.handle(right) -> n17.handle(left)
n14.handle(right) -> n17.handle(left)
n15.handle(right) -> n17.handle(left)
n16.handle(right) -> n17.handle(left)
n17.handle(right) -> n18.handle(left) [label="Sub-ms Latency"]
}

为什么需要这个工作流？

Different consumers need different views of the same data: dashboards need aggregations, search needs full-text indexes, and APIs need denormalized joins. Materialized views built from a single event stream allow each consumer to have a purpose-built read model without complex cross-service queries.

工作原理

Step 1: The write model persists changes and emits domain events to the event bus.
Step 2: The event bus guarantees ordered delivery to all registered projectors.
Step 3: Each projector builds a specific materialized view: order summaries, dashboards, analytics cubes, or search indexes.
Step 4: Views are stored in technology-appropriate stores (Redis, Elasticsearch, OLAP cubes).
Step 5: The query API routes requests to the appropriate materialized view.
Step 6: Views can be rebuilt from scratch by replaying the event stream.

替代方案

Database views and materialized views work for single-database scenarios. GraphQL federation can compose data without materialized views. This template shows how to build multiple optimized read models from a single event stream.

Key Facts

Template Name	CQRS 物化视图架构
Category	Architecture
Steps	6 workflow steps
Format	FlowZap Code (.fz file)