舱壁隔离架构

Architecture

舱壁隔离架构图，为关键、标准和批处理工作负载提供独立的线程池，每个都有独立的连接池、耗尽检测和背压策略，防止一个工作负载饿死其他工作负载。该模板模拟受船体舱壁启发的隔离模式，资源隔离确保一个组件的故障或过载不会级联到其他组件。对于需要为高优先级操作保证可用性的系统至关重要。

下载 .fz 文件

完整 FlowZap 代码

Gateway { # API Gateway
n1: circle label:"Incoming Request"
n2: rectangle label:"Classify Request Type"
n3: diamond label:"Request Category?"
n4: rectangle label:"Route to Critical Pool"
n5: rectangle label:"Route to Standard Pool"
n6: rectangle label:"Route to Batch Pool"
n1.handle(right) -> n2.handle(left)
n2.handle(right) -> n3.handle(left)
n3.handle(right) -> n4.handle(left) [label="Critical"]
n3.handle(bottom) -> n5.handle(top) [label="Standard"]
n3.handle(left) -> n6.handle(right) [label="Batch"]
n4.handle(bottom) -> CriticalPool.n7.handle(top) [label="Priority"]
n5.handle(bottom) -> StandardPool.n11.handle(top) [label="Normal"]
n6.handle(bottom) -> BatchPool.n15.handle(top) [label="Low Priority"]
}
CriticalPool { # Critical Thread Pool (10 threads)
n7: rectangle label:"Dedicated Connection Pool"
n8: rectangle label:"Process Payment Request"
n9: diamond label:"Pool Exhausted?"
n10: rectangle label:"Queue with Timeout"
n7.handle(right) -> n8.handle(left) [label="Thread Available"]
n7.handle(bottom) -> n9.handle(top) [label="Check"]
n9.handle(right) -> n10.handle(left) [label="Yes"]
n9.handle(top) -> n8.handle(bottom) [label="No"]
}
StandardPool { # Standard Thread Pool (20 threads)
n11: rectangle label:"Shared Connection Pool"
n12: rectangle label:"Process API Request"
n13: diamond label:"Pool Exhausted?"
n14: rectangle label:"Return 503 Service Unavailable"
n11.handle(right) -> n12.handle(left) [label="Thread Available"]
n11.handle(bottom) -> n13.handle(top) [label="Check"]
n13.handle(right) -> n14.handle(left) [label="Yes"]
n13.handle(top) -> n12.handle(bottom) [label="No"]
}
BatchPool { # Batch Thread Pool (5 threads)
n15: rectangle label:"Rate-Limited Pool"
n16: rectangle label:"Process Batch Job"
n17: diamond label:"Pool Exhausted?"
n18: rectangle label:"Backpressure: Delay Retry"
n15.handle(right) -> n16.handle(left) [label="Thread Available"]
n15.handle(bottom) -> n17.handle(top) [label="Check"]
n17.handle(right) -> n18.handle(left) [label="Yes"]
n17.handle(top) -> n16.handle(bottom) [label="No"]
}

为什么需要这个工作流？

A single slow or failing dependency can consume all available threads or connections, starving healthy operations. The bulkhead pattern isolates workloads into separate resource pools—like watertight compartments in a ship—so that a failure in one pool cannot affect the availability of others.

工作原理

Step 1: Incoming requests are classified by type: critical, standard, or batch.
Step 2: Each category is routed to a dedicated thread pool with fixed capacity.
Step 3: The critical pool (payments) has guaranteed resources that cannot be consumed by other workloads.
Step 4: The standard pool handles regular API requests with a larger thread allocation.
Step 5: The batch pool processes background jobs with rate limiting and lower priority.
Step 6: When a pool is exhausted, requests are either queued with timeout or rejected with 503.

替代方案

A single shared thread pool is simpler but vulnerable to noisy-neighbor problems. Kubernetes resource limits provide pod-level isolation. This template shows application-level bulkhead isolation for fine-grained workload protection.

Key Facts

Template Name	舱壁隔离架构
Category	Architecture
Steps	6 workflow steps
Format	FlowZap Code (.fz file)