← Knowledge Atlas · Source
The History of Pets vs Cattle
Pets vs Cattle 云计算类比：服务器可处置性的规范定义，基础设施设计的核心区分
源
Pets vs Cattle — Summary
A cloud-era term — imported into agent architecture in 2026
Bias (2011-2012) reframed Bill Baker’s scale-up / scale-out analogy around disposability —
“Is this server an irreplaceable individual, or a replaceable member of a herd?”
Key correction: k8s calling Cassandra/Kafka (cattle architecture) “Pet Sets” is a misuse — pets ≠ stateful, cattle = designed-for-failure
single-container session→Brain / Hands / Session decoupled
container failure = lost session→session externalized · wake() recovers
credentials in container→credentials never enter the sandbox
SeeAnthropic Managed Agents carries this pattern from infra into agents
→ bias-pets-vs-cattle · managed-agents · agent-sandboxingwiki summary
The History of Pets vs Cattle
- 来源:
sources/bias-pets-vs-cattle.md - URL: https://cloudscaling.com/blog/cloud-computing/the-history-of-pets-vs-cattle/
- 作者: Randy Bias
- 发布: 2016-09-19
摘要
Pets vs Cattle 类比的权威历史和规范定义。Bias 在 2011-2012 年将 Bill Baker 关于 SQL Server scale-up vs scale-out 的类比重新解读，强调核心区分是可处置性而非扩展方向：服务器是不可失去的唯一个体（宠物），还是可随时替换的群体成员（牲畜）。
文章纠正了一些常见误用（如 Kubernetes Pet Sets 将 Cassandra/Kafka 等 cattle 架构的数据存储称为”pet”），维护类比的精确性。
与其他 source 的关联
- Managed Agents 将此模式从基础设施层引入 agent 架构——初始的单容器设计是”宠物”，brain-hands 解耦后每个组件变为”牲畜”
- 与 Agent Sandboxing 互补——cattle 化是安全隔离的前提
References
sources/bias-pets-vs-cattle.md