The company has recently been busy the whole link voltage measurement platform research and technical planning document output work, the full link reference pressure measurement solutions can be found to the whole network of industry manufacturers, to be summarized here, and would personally think that landing program make a key point finishing.
Technology Links
Full pressure drops measured solution road link
Ali Baba full link pressure measurement
Ali how to do a full 11-link double pressure measured?
US group pressure test automation practice full link
Link-pressure measurement platform in the practice of US corporations
Hungry exploration and practice it full link pressure measured
Hungry to achieve and what the whole principle of pressure measurement platform link
Like the measured pressure has full link Detailed design and implementation
Jingdong full link pressure measurement system (ForceBot) architecture decrypting
Luo Ji thinking practice and work notes in full respect of the measured pressure link
Manufacturers finishing point program
1, logical thinking
Positioning: protect the core infrastructure business stability;
Importance: high-profile business & technical team to withstand pressure;
Core objectives: service availability, stability, scalability;
2, Ali Baba
Process Management: Effective adequate preparation program + + fly stable platform;
Traffic identification: measured pressure drop shadow passthrough traffic flag library, multiple pressure measurements of the same API, to prevent interception, the IP same, whitelist mechanism;
System Transformation: ① business transformation: traffic identification, unity problem, limiting interception, excluding reports, dynamic verification;.
Data Preparation: Active scheme determination → → technical architecture review the business model pressure measurement range & & & data format of the order data;
. ① business data model: API & flow & magnitude ratio & → Business abstract model conversion funnel (funnel same proportion);
. ② basic data structure: Data middleweight & authenticity (buyers & sellers, merchants & commodities, prices);
Preheat systems: buffer preparation, system load preparation;
Login preparation: Log scene simulation long link (step by step user login), the protection of user services;
Pressure measurement mode: 0:00 pulse Mogao system, limiting degraded, destructive verification (disaster recovery exercise);
3, Jingdong
Scene: buyers and sellers;
Data: double eleven history as the base peak traffic flow, the dynamic pressure increases concurrency;
Traffic: daily traffic flow big promotion (the main library writing pressure);
Engine test pressure: jmeter / Ngrinder;
Test scripts / data: git / local?
Start Mode: incremental gradient pulse, stable level verification;
Implementation: immediate execution, execution timings;
Test scenario: a pressure source, the number of virtual users, test scripts, execution, start-up mode;
Pressure Source: docker cluster, a plurality of sets, no state (status check), the shared resource;
Pressure sensing data: unified storage (ES), aggregated (jmeter), grafana display (to be optimized);
Traffic identification, controlled release air;
4, there is praise
Traffic model: Flow from the buyer side, the normal water level - thrusting - down;
Cost of the machine: the core link according to the amount of expansion, the seller-side service on demand, peak load shifting;
Core Link: more people link & complex & comb filter core aggregated link → → → reject confirmation;
Pressure measurement strategy: single single single link metric mixing → → full link capacity link test pressure (water) → special exercise plans;
Traffic estimates: Monitoring statistics -QPS, the number of connections, IOPS, RT, cache hit rate, consumer group, topic;
datapool: basic data desensitization, script / test data, to provide a unified storage / issued / division function;
Shadow storage: DB route: ① instance with different schema (high risk); ② instance with different schema (high security, high costs);.
Redis route: ①.key value added unified prefix; ②.Redis-client do routing;
ES route: ①.index unified prefix, ES provides unified client for data access, routing done by the client;
Application changes: micro-services, unified isolation, flow-standard pass-through storage;
Send the traffic: data files: by scene discrimination (considering funnel Model - conversion);
Pressure test script: ① different scenes flow ratio; ② each scene to do the conversion from top to bottom by URL (gatling);..
Water level detection: Pressure measurement process, ① real-time acquisition of each service application resource usage + RT + TPS + success rate;. ② flow to intervene to protect production services will not be affected;.
Pressure test implementation: ① basic middleware development, routing policy framework upgrade, pressure test development and debugging engine selection - infrastructure;.
. ② business transformation and upgrading of the line + verification (functional verification, manually click, drop shadow data library) - functional testing;
③ business upgrading + production verification (functional verification, manually click, drop shadow data library) - functional testing.;
④.datapool preparation: data generation, cutting the script file issued - Business Development & Testing;
. ⑤ hair verification under low flow - field test students;
. ⑥ simulate real-world scenarios pressure measurement and verification - team collaboration;
Pressure measuring method: Flow increment / climbing (gradient increases, optimization expansion);
链路梳理：非核心链路-去依赖解耦；
长期规划：轮询化：线时链路测试机器人，实时检测；
常规化：减少人力成本投入；
日常化：尽可能少熬夜，白天完成；
图形化：链路压测规划图形化展示，与业务结合，一键完成数据准备工作；
5、美团
全链路压测思路
系统总体设计
重要程度：系统稳定性建设中的核心重要位置，也是最有效的方案；
技术背景：验证峰值流量下系统服务的伸缩性和稳定性；
验证新上线功能的可用性、稳定性；
限流、降级、熔断、告警灯故障演练；
线上服务容量评估
技术方案：获取线上真实流量-流量录制&流量回放；
快速创建压测环境-环境/服务隔离、流量标透传、灵活伸缩容；
支持多协议类型-http、tcp、webscoket、rpc、dubbo......
实时监控&过载保护；
必备功能：数据构造、压测隔离、场景管理、动态调整、实时监控、压测报告、分布式......
整体架构：web管理端：数据构造、环境准备、任务管理、场景管理、压测动态调整、报表展示；
调度中心：压测资源调度、任务分发、机器资源管理；
压测引擎：流量构造、模拟；
监控组件：实施监控、压测数据统计、聚合分析、展示；
链路梳理：工具化，提供自动构建压测入口链路完整的依赖信息，辅助提效手段；
挡板服务：配置化手段，完成外部依赖等相关接口的Mock配置，无需在业务代码中嵌入压测判断逻辑；
数据构造：流量复制、存储、清洗、解析、组合展示、偏移脱敏处理；
链路追踪：链路匹配分析定位
服务隔离：大促（业务低谷）&常规（机器隔离）
数据隔离：同库不同表（影子表）-成本低，风险较高
机器管理：动态扩容、灰度升级、异常摘除
压测引擎：jmeter&nGrinder&gatling
内存优化：内存管理&JVM参数
监控：秒级监控、实时展示、告警、服务保护
日志：压测日志采样、展示
服务治理：限流熔断降级保护
注意事项：小步快跑，及时响应、项目推广、开放生态、基础资源&赋能；
6、饿了么
用例管理：建立用例、文件上传、分类管理；
压测执行：一键启动，可指定线程数&预热时间&测试周期和负载机，文件切割分发，分布式执行；
监控数据：TPS、ART、Error%实时展示；
实时数据持久化-influxdb，设置过期时间；
冷数据持久化至MongoDB；
测试报告：用例执行完毕，自动生成测试报告；
集群监控：压测机使用状态监控，作为共享资源，提示用户可用的测试机；
安全保障：权限管理&异常操作限制；
分布式压测实现：二次开发；
异常干预：水位超限&服务保护&阈值告警&压测触发失败（形成闭环）；
influxdb轮询：http请求轮询频次较高，需优化；
预配置：提前配置，时段保留，节省时间，提高效率；
服务保护
权限分级：高峰期禁止直行；
压测干预，随时手动触发禁止动作，kill所有压测机上运行的压测进程；
限流熔断：根据错误率和告警阈值判断，达到或超过自动熔断；
兜底服务：系统不可用需要停止测试时，外部强制停止，过载保护；
以上内容，来自各大厂方案的汇总整理，仅供参考。。。