{"type":"doc","content":[{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"设计理念","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"空间换时间","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"多级缓存，静态化","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 客户端页面缓存（http header中包含Expires/Cache of Control，last modified(304，server不返回body，客户端可以继续用cache，减少流量)，ETag）","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"反向代理缓存","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"应用端的缓存(memcache)","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"内存数据库","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"Buffer、cache机制（数据库，中间件等）","attrs":{}}]}]}],"attrs":{}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"索引机制","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":" 哈希、B（B+）树、倒排、bitmap","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"哈希索引：适合综合数组的寻址和链表的插入特性，可以实现数据的快速存取。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"B（B+）树索引：适合于查询为主导的场景，避免多次的IO，提高查询的效率。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"倒排索引：实现单词到文档映射关系的最佳实现方式和最有效的索引结构，广泛用在搜索领域。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"Bitmap：是一种非常简洁快速的","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}}],"text":"数据结构","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"，他能同时使存储空间和速度最优化（而不必空间换时间），适合于海量数据的的计算场景。","attrs":{}}]}]}],"attrs":{}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"并行与分布式计算","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"任务切分、分而治之(MR)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"大规模的数据中，数据存在一定局部性的特征，利用局部性的原理将海量数据计算的问题分而治之","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":"MR模型是无共享的架构，数据集分布至各个节点。处理时，每个节点就近读取本地存储的数据处理(map)，将处理后的数据进行合并(","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"combine","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":")、排序(","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"shuffle and sort","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":")后再分发(至reduce节点)，避免了大量数据的传输，提高了处理效率。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"多进程、多线程并行执行(MPP)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"并行计算（Parallel Computing）是指同时使用多种计算资源解决计算问题的过程，是提高计算机系统计算速度和处理能力的一种有效手段。","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":"它的基本思想是用多个处理器/进程/线程来协同求解同一问题，即将被求解的问题分解成若干个部分，各部分均由一个独立的处理机来并行计算。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":"和MR的区别在于，它是基于问题分解的，而不是基于数据分解。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"多维度的可用","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"负载均衡、容灾、备份","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"随着平台并发量的增大，需要扩容节点进行集群，利用负载均衡设备进行请求的分发；","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"负载均衡","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"设备通常在提供","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"负载均衡","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"的同时，也提供","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"失效检测","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"功能；同时为了提高可用性，需要有容灾备份，以防止节点宕机失效带来的不可用问题；备份有在线的和离线备份，可以根据失效性要求的不同，进行选择不同备份策略。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"读写分离","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"读写分离是对数据库来讲的，随着系统并发量的增大，提高数据访问可用性的一个重要手段就是写数据和读数据进行分离；","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"当然在读写分离的同时，需要关注数据的一致性问题；对于一致性的问题，在分布式的系统CAP定量中，更多的关注于可用性。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"依赖关系","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"平台中各个模块之间的关系尽量是低耦合的，可以通过相关的消息组件进行交互，能异步则异步，分清楚数据流转的主流程和副流程，主副是异步的，比如记录日志可以是异步操作的，增加整个系统的可用性。当然在异步处理中，为了确保数据得到接收或者处理，往往需要确认机制(confirm、ack)。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 但是有些场景中，虽然请求已经得到处理，但是因其他原因(比如网络不稳定)，确认消息没有返回，那么这种情况下需要进行请求的重发，对请求的处理设计因重发因素需要考虑幂等性。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"监控","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":" 监控也是提高整个平台可用性的一个重要手段，多平台进行多个维度的监控；模块在运行时候是透明的，以达到运行期白盒化。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"伸缩","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"拆分","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":" 拆分包括对业务的拆分和对数据库的拆分。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"系统的资源总是有限的，一段比较长的业务执行如果是一竿子执行的方式，在大量并发的操作下，这种阻塞的方式，无法有效的及时释放资源给其他进程执行，这样系统的吞吐量不高。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 需要把业务进行逻辑的分段，采用异步非阻塞的方式，提高系统的吞吐量。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 随着数据量和并发量的增加，读写分离不能满足系统并发性能的要求，需要对数据进行切分，包括对数据进行分库和分表。这种分库分表的方式，需要增加对数据的路由逻辑支持。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"无状态","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 对于系统的伸缩性而言，模块最好是无状态的，通过增加节点就可以提高整个的吞吐量。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"优化资源利用","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"系统容量有限","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 系统的容量是有限的，承受的并发量也是有限的，在架构设计时，一定需要考虑流量的控制，防止因意外攻击或者瞬时并发量的冲击导致系统崩溃。在设计时增加流控的措施，可考虑对请求进行排队，超出预期的范围，可以进行告警或者丢弃。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"原子操作与并发控制","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"对于共享资源的访问，为了防止冲突，需要进行并发的控制，同时有些交易需要有事务性来保证交易的一致性，所以在交易系统的设计时，需考虑原子操作和并发控制。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 保证并发控制一些常用高性能手段有，乐观锁、Latch、mutex、写时复制、CAS等；多版本的并发控制MVCC通常是保证一致性的重要手段，这个在数据库的设计中经常会用到。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"逻辑的不同，采取不一样策略","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":" 平台中业务逻辑存在不同的类型，有计算复杂型的，有消耗IO型的，同时就同一种类型而言，不同的业务逻辑消耗的资源数量也是不一样的，这就需要针对不同的逻辑采取不同的策略。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":" 针对IO型的，可以采取基于事件驱动的异步非阻塞的方式，","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"单线程方式可以减少线程的切换引起的开销，或者在多线程的情况下采取自旋spin的方式","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}}],"text":"，","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"减少对线程的切换","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 对于计算型的，充分利用多线程进行操作。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 同一类型的调用方式，不同的业务进行合适的资源分配，设置不同的计算节点数量或者线程数量，对业务进行分流，优先执行优先级别高的业务。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"容错隔离","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 系统的有些业务模块在出现错误时，为了减少并发下对正常请求的处理的影响，有时候需要考虑对这些异常状态的请求进行单独渠道的处理，","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"甚至暂时自动禁止这些异常的业务模块","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 有些请求的失败可能是偶然的暂时的失败(比如网络不稳定)，需要进行请求重试的考虑。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"资源释放","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 系统的资源是有限的，在使用资源时，一定要在最后释放资源，无论是请求走的是正常路径还是异常的路径，以便于资源的及时回收，供其他请求使用。在设计通信的架构时，往往需要考虑超时的控制。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}
🌏【架构师指南】带你全面认识实现"三高"架构设计方案
{"type":"doc","content":[{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"设计理念","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"空间换时间","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"多级缓存，静态化","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 客户端页面缓存（http header中包含Expires/Cache of Control，last modified(304，server不返回body，客户端可以继续用cache，减少流量)，ETag）","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"反向代理缓存","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"应用端的缓存(memcache)","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"内存数据库","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"strong","attrs":{}}],"text":"Buffer、cache机制（数据库，中间件等）","attrs":{}}]}]}],"attrs":{}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"索引机制","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":" 哈希、B（B+）树、倒排、bitmap","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"哈希索引：适合综合数组的寻址和链表的插入特性，可以实现数据的快速存取。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"B（B+）树索引：适合于查询为主导的场景，避免多次的IO，提高查询的效率。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"倒排索引：实现单词到文档映射关系的最佳实现方式和最有效的索引结构，广泛用在搜索领域。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"Bitmap：是一种非常简洁快速的","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}}],"text":"数据结构","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":9}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"，他能同时使存储空间和速度最优化（而不必空间换时间），适合于海量数据的的计算场景。","attrs":{}}]}]}],"attrs":{}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"并行与分布式计算","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"任务切分、分而治之(MR)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"大规模的数据中，数据存在一定局部性的特征，利用局部性的原理将海量数据计算的问题分而治之","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":"MR模型是无共享的架构，数据集分布至各个节点。处理时，每个节点就近读取本地存储的数据处理(map)，将处理后的数据进行合并(","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"combine","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":")、排序(","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"shuffle and sort","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":")后再分发(至reduce节点)，避免了大量数据的传输，提高了处理效率。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"多进程、多线程并行执行(MPP)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"并行计算（Parallel Computing）是指同时使用多种计算资源解决计算问题的过程，是提高计算机系统计算速度和处理能力的一种有效手段。","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":"它的基本思想是用多个处理器/进程/线程来协同求解同一问题，即将被求解的问题分解成若干个部分，各部分均由一个独立的处理机来并行计算。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":"和MR的区别在于，它是基于问题分解的，而不是基于数据分解。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"多维度的可用","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"负载均衡、容灾、备份","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"随着平台并发量的增大，需要扩容节点进行集群，利用负载均衡设备进行请求的分发；","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"负载均衡","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"设备通常在提供","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"负载均衡","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"的同时，也提供","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"失效检测","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"功能；同时为了提高可用性，需要有容灾备份，以防止节点宕机失效带来的不可用问题；备份有在线的和离线备份，可以根据失效性要求的不同，进行选择不同备份策略。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"读写分离","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"读写分离是对数据库来讲的，随着系统并发量的增大，提高数据访问可用性的一个重要手段就是写数据和读数据进行分离；","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"当然在读写分离的同时，需要关注数据的一致性问题；对于一致性的问题，在分布式的系统CAP定量中，更多的关注于可用性。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"依赖关系","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"平台中各个模块之间的关系尽量是低耦合的，可以通过相关的消息组件进行交互，能异步则异步，分清楚数据流转的主流程和副流程，主副是异步的，比如记录日志可以是异步操作的，增加整个系统的可用性。当然在异步处理中，为了确保数据得到接收或者处理，往往需要确认机制(confirm、ack)。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 但是有些场景中，虽然请求已经得到处理，但是因其他原因(比如网络不稳定)，确认消息没有返回，那么这种情况下需要进行请求的重发，对请求的处理设计因重发因素需要考虑幂等性。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"监控","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":" 监控也是提高整个平台可用性的一个重要手段，多平台进行多个维度的监控；模块在运行时候是透明的，以达到运行期白盒化。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"伸缩","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"拆分","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":" 拆分包括对业务的拆分和对数据库的拆分。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"系统的资源总是有限的，一段比较长的业务执行如果是一竿子执行的方式，在大量并发的操作下，这种阻塞的方式，无法有效的及时释放资源给其他进程执行，这样系统的吞吐量不高。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 需要把业务进行逻辑的分段，采用异步非阻塞的方式，提高系统的吞吐量。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 随着数据量和并发量的增加，读写分离不能满足系统并发性能的要求，需要对数据进行切分，包括对数据进行分库和分表。这种分库分表的方式，需要增加对数据的路由逻辑支持。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"无状态","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 对于系统的伸缩性而言，模块最好是无状态的，通过增加节点就可以提高整个的吞吐量。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"优化资源利用","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"系统容量有限","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 系统的容量是有限的，承受的并发量也是有限的，在架构设计时，一定需要考虑流量的控制，防止因意外攻击或者瞬时并发量的冲击导致系统崩溃。在设计时增加流控的措施，可考虑对请求进行排队，超出预期的范围，可以进行告警或者丢弃。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"原子操作与并发控制","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"对于共享资源的访问，为了防止冲突，需要进行并发的控制，同时有些交易需要有事务性来保证交易的一致性，所以在交易系统的设计时，需考虑原子操作和并发控制。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 保证并发控制一些常用高性能手段有，乐观锁、Latch、mutex、写时复制、CAS等；多版本的并发控制MVCC通常是保证一致性的重要手段，这个在数据库的设计中经常会用到。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"逻辑的不同，采取不一样策略","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":" 平台中业务逻辑存在不同的类型，有计算复杂型的，有消耗IO型的，同时就同一种类型而言，不同的业务逻辑消耗的资源数量也是不一样的，这就需要针对不同的逻辑采取不同的策略。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"strong","attrs":{}}],"text":" 针对IO型的，可以采取基于事件驱动的异步非阻塞的方式，","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"单线程方式可以减少线程的切换引起的开销，或者在多线程的情况下采取自旋spin的方式","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}}],"text":"，","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"减少对线程的切换","attrs":{}}]},{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 对于计算型的，充分利用多线程进行操作。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 同一类型的调用方式，不同的业务进行合适的资源分配，设置不同的计算节点数量或者线程数量，对业务进行分流，优先执行优先级别高的业务。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"容错隔离","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 系统的有些业务模块在出现错误时，为了减少并发下对正常请求的处理的影响，有时候需要考虑对这些异常状态的请求进行单独渠道的处理，","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}},{"type":"color","attrs":{"color":"#F5222D","name":"red"}},{"type":"strong","attrs":{}}],"text":"甚至暂时自动禁止这些异常的业务模块","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 有些请求的失败可能是偶然的暂时的失败(比如网络不稳定)，需要进行请求重试的考虑。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"资源释放","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" ","attrs":{}},{"type":"text","marks":[{"type":"size","attrs":{"size":10}}],"text":" 系统的资源是有限的，在使用资源时，一定要在最后释放资源，无论是请求走的是正常路径还是异常的路径，以便于资源的及时回收，供其他请求使用。在设计通信的架构时，往往需要考虑超时的控制。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}
發表評論
所有評論
還沒有人評論，想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章