Redis为什么变慢了？一文讲透如何排查Redis性能问题 | 万字长文

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 作为优秀的内存数据库，其拥有非常高的性能，单个实例的 OPS 能够达到 10W 左右。但也正因此如此，当我们在使用 Redis 时，如果发现操作延迟变大的情况，就会与我们的预期不符。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你也许或多或少地，也遇到过以下这些场景：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在 Redis 上执行同样的命令，为什么有时响应很快，有时却很慢？","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为什么 Redis 执行 SET、DEL 命令耗时也很久？","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为什么我的 Redis 突然慢了一波，之后又恢复正常了？","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为什么我的 Redis 稳定运行了很久，突然从某个时间点开始变慢了？","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"...","attrs":{}}]}],"attrs":{}}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你并不清楚 Redis 内部的实现原理，那么在排查这种延迟问题时就会一头雾水。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你也遇到了以上情况，那么，这篇文章将会给你一个「全面」的问题排查思路，并且针对这些导致变慢的场景，我还会给你一个高效的解决方案。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"在正文开始之前，我需要提醒你的是，这篇文章很长，涵盖的 Redis 知识点也非常广，全篇文章接近 2W 字，如果此时你的阅读环境不适合专注阅读，我建议你先收藏此文章，然后在合适的时间专注阅读这篇文章。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你能耐心且认真地读完这篇文章，我可以保证，你对 Redis 的性能调优将会有非常大的收获。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你准备好了，那就跟着我的思路开始吧！","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/ea/ea3e484f08ebe9dde9b117a994d218e8.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"Redis真的变慢了吗？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，在开始之前，你需要弄清楚 Redis 是否真的变慢了？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你发现你的业务服务 API 响应延迟变长，首先你需要先排查服务内部，究竟是哪个环节拖慢了整个服务。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"比较高效的做法是，在服务内部集成","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"链路追踪","attrs":{}},{"type":"text","text":"，也就是在服务访问外部依赖的出入口，记录下每次请求外部依赖的响应延时。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/a3/a33803fc385125d42f4b2b65d341fa31.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你发现确实是操作 Redis 的这条链路耗时变长了，那么此刻你需要把焦点关注在业务服务到 Redis 这条链路上。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从你的业务服务到 Redis 这条链路变慢的原因可能也有 2 个：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"业务服务器到 Redis 服务器之间的网络存在问题，例如网络线路质量不佳，网络数据包在传输时存在延迟、丢包等情况","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 本身存在问题，需要进一步排查是什么原因导致 Redis 变慢","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"通常来说，第一种情况发生的概率比较小，如果是服务器之间网络存在问题，那部署在这台业务服务器上的所有服务都会发生网络延迟的情况，此时你需要联系网络运维同事，让其协助解决网络问题。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们这篇文章，重点关注的是第二种情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"也就是从 Redis 角度来排查，是否存在导致变慢的场景，以及都有哪些因素会导致 Redis 的延迟增加，然后针对性地进行优化。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"排除网络原因，如何确认你的 Redis 是否真的变慢了？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，你需要对 Redis 进行基准性能测试，了解你的 Redis 在生产环境服务器上的基准性能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"什么是基准性能？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"简单来讲，基准性能就是指 Redis 在一台负载正常的机器上，其最大的响应延迟和平均响应延迟分别是怎样的？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为什么要测试基准性能？我参考别人提供的响应延迟，判断自己的 Redis 是否变慢不行吗？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"答案是否定的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因为 Redis 在不同的软硬件环境下，它的性能是各不相同的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"例如，我的机器配置比较低，当延迟为 2ms 时，我就认为 Redis 变慢了，但是如果你的硬件配置比较高，那么在你的运行环境下，可能延迟是 0.5ms 时就可以认为 Redis 变慢了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，你只有了解了你的 Redis 在生产环境服务器上的基准性能，才能进一步评估，当其延迟达到什么程度时，才认为 Redis 确实变慢了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"具体如何做？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为了避免业务服务器到 Redis 服务器之间的网络延迟，你需要直接在 Redis 服务器上测试实例的响应延迟情况。执行以下命令，就可以测试出这个实例 60 秒内的最大响应延迟：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"$ redis-cli -h 127.0.0.1 -p 6379 --intrinsic-latency 60\nMax latency so far: 1 microseconds.\nMax latency so far: 15 microseconds.\nMax latency so far: 17 microseconds.\nMax latency so far: 18 microseconds.\nMax latency so far: 31 microseconds.\nMax latency so far: 32 microseconds.\nMax latency so far: 59 microseconds.\nMax latency so far: 72 microseconds.\n\n1428669267 total runs (avg latency: 0.0420 microseconds / 42.00 nanoseconds per run).\nWorst run took 1429x longer than the average latency.","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从输出结果可以看到，这 60 秒内的最大响应延迟为 72 微秒（0.072毫秒）。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你还可以使用以下命令，查看一段时间内 Redis 的最小、最大、平均访问延迟：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"$ redis-cli -h 127.0.0.1 -p 6379 --latency-history -i 1\nmin: 0, max: 1, avg: 0.13 (100 samples) -- 1.01 seconds range\nmin: 0, max: 1, avg: 0.12 (99 samples) -- 1.01 seconds range\nmin: 0, max: 1, avg: 0.13 (99 samples) -- 1.01 seconds range\nmin: 0, max: 1, avg: 0.10 (99 samples) -- 1.01 seconds range\nmin: 0, max: 1, avg: 0.13 (98 samples) -- 1.00 seconds range\nmin: 0, max: 1, avg: 0.08 (99 samples) -- 1.01 seconds range\n...","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以上输出结果是，每间隔 1 秒，采样 Redis 的平均操作耗时，其结果分布在 0.08 ~ 0.13 毫秒之间。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"了解了基准性能测试方法，那么你就可以按照以下几步，来判断你的 Redis 是否真的变慢了：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"在相同配置的服务器上，测试一个正常 Redis 实例的基准性能","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"找到你认为可能变慢的 Redis 实例，测试这个实例的基准性能","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"如果你观察到，这个实例的运行延迟是正常 Redis 基准性能的 2 倍以上，即可认为这个 Redis 实例确实变慢了 ","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"确认是 Redis 变慢了，那如何排查是哪里发生了问题呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下面跟着我的思路，我们从易到难，一步步来分析可能导致 Redis 变慢的因素。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"使用复杂度过高的命令","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，第一步，你需要去查看一下 Redis 的慢日志（slowlog）。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 提供了慢日志命令的统计功能，它记录了有哪些命令在执行时耗时比较久。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"查看 Redis 慢日志之前，你需要设置慢日志的阈值。例如，设置慢日志的阈值为 5 毫秒，并且保留最近 500 条慢日志记录：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":""},"content":[{"type":"text","text":"# 命令执行耗时超过 5 毫秒，记录慢日志\nCONFIG SET slowlog-log-slower-than 5000\n# 只保留最近 500 条慢日志\nCONFIG SET slowlog-max-len 500","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"设置完成之后，所有执行的命令如果操作耗时超过了 5 毫秒，都会被 Redis 记录下来。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时，你可以执行以下命令，就可以查询到最近记录的慢日志：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":""},"content":[{"type":"text","text":"127.0.0.1:6379> SLOWLOG get 5\n1) 1) (integer) 32693 # 慢日志ID\n 2) (integer) 1593763337 # 执行时间戳\n 3) (integer) 5299 # 执行耗时(微秒)\n 4) 1) \"LRANGE\" # 具体执行的命令和参数\n 2) \"user_list:2000\"\n 3) \"0\"\n 4) \"-1\"\n2) 1) (integer) 32692\n 2) (integer) 1593763337\n 3) (integer) 5044\n 4) 1) \"GET\"\n 2) \"user_info:1000\"\n...","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"通过查看慢日志，我们就可以知道在什么时间点，执行了哪些命令比较耗时。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你的应用程序执行的 Redis 命令有以下特点，那么有可能会导致操作延迟变大：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"经常使用 O(N) 以上覆杂度的命令，例如 SORT、SUNION、ZUNIONSTORE 聚合类命令","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"使用 O(N) 复杂度的命令，但 N 的值非常大","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第一种情况导致变慢的原因在于，Redis 在操作内存数据时，时间复杂度过高，要花费更多的 CPU 资源。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第二种情况导致变慢的原因在于，Redis 一次需要返回给客户端的数据过多，更多时间花费在数据协议的组装和网络传输过程中。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，我们还可以从资源使用率层面来分析，如果你的应用程序操作 Redis 的 OPS 不是很大，但 Redis 实例的 ","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"CPU 使用率却很高","attrs":{}},{"type":"text","text":"，那么很有可能是使用了复杂度过高的命令导致的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除此之外，我们都知道，Redis 是单线程处理客户端请求的，如果你经常使用以上命令，那么当 Redis 处理客户端请求时，一旦前面某个命令发生耗时，就会导致后面的请求发生排队，对于客户端来说，响应延迟也会变长。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/0c/0ce7e8c0f53edbaf17a2c13d91c66882.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"针对这种情况如何解决呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"答案很简单，你可以使用以下方法优化你的业务：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"尽量不使用 O(N) 以上覆杂度过高的命令，对于数据的聚合操作，放在客户端做","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"执行 O(N) 命令，保证 N 尽量的小（推荐 N <= 300），每次获取尽量少的数据，让 Redis 可以及时处理返回","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"操作bigkey","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你查询慢日志发现，并不是复杂度过高的命令导致的，而都是 SET / DEL 这种简单命令出现在慢日志中，那么你就要怀疑你的实例否写入了 bigkey。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 在写入数据时，需要为新的数据分配内存，相对应的，当从 Redis 中删除数据时，它会释放对应的内存空间。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果一个 key 写入的 value 非常大，那么 Redis 在","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"分配内存时就会比较耗时","attrs":{}},{"type":"text","text":"。同样的，当删除这个 key 时，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"释放内存也会比较耗时","attrs":{}},{"type":"text","text":"，这种类型的 key 我们一般称之为 bigkey。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时，你需要检查你的业务代码，是否存在写入 bigkey 的情况。你需要评估写入一个 key 的数据大小，尽量避免一个 key 存入过大的数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果已经写入了 bigkey，那有没有什么办法可以扫描出实例中 bigkey 的分布情况呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"答案是可以的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 提供了扫描 bigkey 的命令，执行以下命令就可以扫描出，一个实例中 bigkey 的分布情况，输出结果是以类型维度展示的：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":""},"content":[{"type":"text","text":"$ redis-cli -h 127.0.0.1 -p 6379 --bigkeys -i 0.01\n\n...\n-------- summary -------\n\nSampled 829675 keys in the keyspace!\nTotal key length in bytes is 10059825 (avg len 12.13)\n\nBiggest string found 'key:291880' has 10 bytes\nBiggest list found 'mylist:004' has 40 items\nBiggest set found 'myset:2386' has 38 members\nBiggest hash found 'myhash:3574' has 37 fields\nBiggest zset found 'myzset:2704' has 42 members\n\n36313 strings with 363130 bytes (04.38% of keys, avg size 10.00)\n787393 lists with 896540 items (94.90% of keys, avg size 1.14)\n1994 sets with 40052 members (00.24% of keys, avg size 20.09)\n1990 hashs with 39632 fields (00.24% of keys, avg size 19.92)\n1985 zsets with 39750 members (00.24% of keys, avg size 20.03)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从输出结果我们可以很清晰地看到，每种数据类型所占用的最大内存 / 拥有最多元素的 key 是哪一个，以及每种数据类型在整个实例中的占比和平均大小 / 元素数量。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其实，使用这个命令的原理，就是 Redis 在内部执行了 SCAN 命令，遍历整个实例中所有的 key，然后针对 key 的类型，分别执行 STRLEN、LLEN、HLEN、SCARD、ZCARD 命令，来获取 String 类型的长度、容器类型（List、Hash、Set、ZSet）的元素个数。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这里我需要提醒你的是，当执行这个命令时，要注意 2 个问题：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"对线上实例进行 bigkey 扫描时，Redis 的 OPS 会突增，为了降低扫描过程中对 Redis 的影响，最好控制一下扫描的频率，指定 -i 参数即可，它表示扫描过程中每次扫描后休息的时间间隔，单位是秒","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"扫描结果中，对于容器类型（List、Hash、Set、ZSet）的 key，只能扫描出元素最多的 key。但一个 key 的元素多，不一定表示占用内存也多，你还需要根据业务情况，进一步评估内存占用情况","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那针对 bigkey 导致延迟的问题，有什么好的解决方案呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这里有两点可以优化：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"业务应用尽量避免写入 bigkey","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"如果你使用的 Redis 是 4.0 以上版本，用 UNLINK 命令替代 DEL，此命令可以把释放 key 内存的操作，放到后台线程中去执行，从而降低对 Redis 的影响","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"如果你使用的 Redis 是 6.0 以上版本，可以开启 lazy-free 机制（lazyfree-lazy-user-del = yes），在执行 DEL 命令时，释放内存也会放到后台线程中执行","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但即便可以使用方案 2，我也不建议你在实例中存入 bigkey。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这是因为 bigkey 在很多场景下，依旧会产生性能问题。例如，bigkey 在分片集群模式下，对于数据的迁移也会有性能影响，以及我后面即将讲到的数据过期、数据淘汰、透明大页，都会受到 bigkey 的影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"集中过期","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你发现，平时在操作 Redis 时，并没有延迟很大的情况发生，但在某个时间点突然出现一波延时，其现象表现为：","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"变慢的时间点很有规律，例如某个整点，或者每间隔多久就会发生一波延迟。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果是出现这种情况，那么你需要排查一下，业务代码中是否存在设置大量 key 集中过期的情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果有大量的 key 在某个固定时间点集中过期，在这个时间点访问 Redis 时，就有可能导致延时变大。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为什么集中过期会导致 Redis 延迟变大？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这就需要我们了解 Redis 的过期策略是怎样的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 的过期数据采用被动过期 + 主动过期两种策略：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"被动过期：只有当访问某个 key 时，才判断这个 key 是否已过期，如果已过期，则从实例中删除","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"主动过期：Redis 内部维护了一个定时任务，默认每隔 100 毫秒（1秒10次）就会从全局的过期哈希表中随机取出 20 个 key，然后删除其中过期的 key，如果过期 key 的比例超过了 25%，则继续重复此过程，直到过期 key 的比例下降到 25% 以下，或者这次任务的执行耗时超过了 25 毫秒，才会退出循环","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"注意，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"这个主动过期 key 的定时任务，是在 Redis 主线程中执行的","attrs":{}},{"type":"text","text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"也就是说如果在执行主动过期的过程中，出现了需要大量删除过期 key 的情况，那么此时应用程序在访问 Redis 时，必须要等待这个过期任务执行结束，Redis 才可以服务这个客户端请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时就会出现，应用访问 Redis 延时变大。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果此时需要过期删除的是一个 bigkey，那么这个耗时会更久。而且，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"这个操作延迟的命令并不会记录在慢日志中","attrs":{}},{"type":"text","text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因为慢日志中","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"只记录一个命令真正操作内存数据的耗时","attrs":{}},{"type":"text","text":"，而 Redis 主动删除过期 key 的逻辑，是在命令真正执行之前执行的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，此时你会看到，慢日志中没有操作耗时的命令，但我们的应用程序却感知到了延迟变大，其实时间都花费在了删除过期 key 上，这种情况我们需要尤为注意。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/25/25dad196bad2e2bb79edc2059acde8f1.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那遇到这种情况，如何分析和排查？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时，你需要检查你的业务代码，是否存在集中过期 key 的逻辑。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一般集中过期使用的是 expireat / pexpireat 命令，你需要在代码中搜索这个关键字。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"排查代码后，如果确实存在集中过期 key 的逻辑存在，但这种逻辑又是业务所必须的，那此时如何优化，同时又不对 Redis 有性能影响呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一般有两种方案来规避这个问题：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"集中过期 key 增加一个随机过期时间，把集中过期的时间打散，降低 Redis 清理过期 key 的压力","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"如果你使用的 Redis 是 4.0 以上版本，可以开启 lazy-free 机制，当删除过期 key 时，把释放内存的操作放到后台线程中执行，避免阻塞主线程","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第一种方案，在设置 key 的过期时间时，增加一个随机时间，伪代码可以这么写：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":""},"content":[{"type":"text","text":"# 在过期时间点之后的 5 分钟内随机过期掉\nredis.expireat(key, expire_time + random(300))","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这样一来，Redis 在处理过期时，不会因为集中删除过多的 key 导致压力过大，从而避免阻塞主线程。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第二种方案，Redis 4.0 以上版本，开启 lazy-free 机制：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"# 释放过期 key 的内存，放到后台线程执行\nlazyfree-lazy-expire yes","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，除了业务层面的优化和修改配置之外，你还可以通过运维手段及时发现这种情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"运维层面，你需要把 Redis 的各项运行状态数据监控起来，在 Redis 上执行 INFO 命令就可以拿到这个实例所有的运行状态数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在这里我们需要重点关注 expired_keys 这一项，它代表整个实例到目前为止，累计删除过期 key 的数量。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你需要把这个指标监控起来，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"当这个指标在很短时间内出现了突增","attrs":{}},{"type":"text","text":"，需要及时报警出来，然后与业务应用报慢的时间点进行对比分析，确认时间是否一致，如果一致，则可以确认确实是因为集中过期 key 导致的延迟变大。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"实例内存达到上限","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你的 Redis 实例设置了内存上限 maxmemory，那么也有可能导致 Redis 变慢。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当我们把 Redis 当做纯缓存使用时，通常会给这个实例设置一个内存上限 maxmemory，然后设置一个数据淘汰策略。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而当实例的内存达到了 maxmemory 后，你可能会发现，在此之后每次写入新数据，操作延迟变大了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这是为什么？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"原因在于，当 Redis 内存达到 maxmemory 后，每次写入新的数据之前，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"Redis 必须先从实例中踢出一部分数据，让整个实例的内存维持在 maxmemory 之下","attrs":{}},{"type":"text","text":"，然后才能把新数据写进来。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个踢出旧数据的逻辑也是需要消耗时间的，而具体耗时的长短，要取决于你配置的淘汰策略：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"allkeys-lru：不管 key 是否设置了过期，淘汰最近最少访问的 key","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"volatile-lru：只淘汰最近最少访问、并设置了过期时间的 key","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"allkeys-random：不管 key 是否设置了过期，随机淘汰 key","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"volatile-random：只随机淘汰设置了过期时间的 key","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"allkeys-ttl：不管 key 是否设置了过期，淘汰即将过期的 key","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"noeviction：不淘汰任何 key，实例内存达到 maxmeory 后，再写入新数据直接返回错误","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"allkeys-lfu：不管 key 是否设置了过期，淘汰访问频率最低的 key（4.0+版本支持）","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"volatile-lfu：只淘汰访问频率最低、并设置了过期时间 key（4.0+版本支持）","attrs":{}}]}],"attrs":{}}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"具体使用哪种策略，我们需要根据具体的业务场景来配置。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一般最常使用的是 allkeys-lru / volatile-lru 淘汰策略，它们的处理逻辑是，每次从实例中随机取出一批 key（这个数量可配置），然后淘汰一个最少访问的 key，之后把剩下的 key 暂存到一个池子中，继续随机取一批 key，并与之前池子中的 key 比较，再淘汰一个最少访问的 key。以此往复，直到实例内存降到 maxmemory 之下。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"需要注意的是，Redis 的淘汰数据的逻辑与删除过期 key 的一样，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"也是在命令真正执行之前执行的","attrs":{}},{"type":"text","text":"，也就是说它也会增加我们操作 Redis 的延迟，而且，写 OPS 越高，延迟也会越明显。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，如果此时你的 Redis 实例中还存储了 bigkey，那么","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"在淘汰删除 bigkey 释放内存时，也会耗时比较久","attrs":{}},{"type":"text","text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/36/367a42fe57927cee24476747d2f42e43.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"看到了么？bigkey 的危害到处都是，这也是前面我提醒你尽量不存储 bigkey 的原因。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"针对这种情况，如何解决呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我给你 4 个方面的优化建议：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"避免存储 bigkey，降低释放内存的耗时","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"淘汰策略改为随机淘汰，随机淘汰比 LRU 要快很多（视业务情况调整）","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"拆分实例，把淘汰 key 的压力分摊到多个实例上","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"如果使用的是 Redis 4.0 以上版本，开启 layz-free 机制，把淘汰 key 释放内存的操作放到后台线程中执行（配置 lazyfree-lazy-eviction = yes）","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"fork耗时严重","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为了保证 Redis 数据的安全性，我们可能会开启后台定时 RDB 和 AOF rewrite 功能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但如果你发现，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"操作 Redis 延迟变大，都发生在 Redis 后台 RDB 和 AOF rewrite 期间","attrs":{}},{"type":"text","text":"，那你就需要排查，在这期间有可能导致变慢的情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当 Redis 开启了后台 RDB 和 AOF rewrite 后，在执行时，它们都需要主进程创建出一个子进程进行数据的持久化。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"主进程创建子进程，会调用操作系统提供的 fork 函数。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而 fork 在执行过程中，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"主进程需要拷贝自己的内存页表给子进程","attrs":{}},{"type":"text","text":"，如果这个实例很大，那么这个拷贝的过程也会比较耗时。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而且这个 fork 过程会消耗大量的 CPU 资源，在完成 fork 之前，整个 Redis 实例会被阻塞住，无法处理任何客户端请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果此时你的 CPU 资源本来就很紧张，那么 fork 的耗时会更长，甚至达到秒级，这会严重影响 Redis 的性能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那如何确认确实是因为 fork 耗时导致的 Redis 延迟变大呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你可以在 Redis 上执行 INFO 命令，查看 latest","attrs":{}},{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"fork","attrs":{}},{"type":"text","text":"usec 项，单位微秒。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"# 上一次 fork 耗时，单位微秒\nlatest_fork_usec:59477","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个时间就是主进程在 fork 子进程期间，整个实例阻塞无法处理客户端请求的时间。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你发现这个耗时很久，就要警惕起来了，这意味在这期间，你的整个 Redis 实例都处于不可用的状态。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除了数据持久化会生成 RDB 之外，当主从节点第一次建立数据同步时，主节点也创建子进程生成 RDB，然后发给从节点进行一次全量同步，所以，这个过程也会对 Redis 产生性能影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/53/53e020351d0f7483480735a9e103a9fe.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"要想避免这种情况，你可以采取以下方案进行优化：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"控制 Redis 实例的内存：尽量在 10G 以下，执行 fork 的耗时与实例大小有关，实例越大，耗时越久","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"合理配置数据持久化策略：在 slave 节点执行 RDB 备份，推荐在低峰期执行，而对于丢失数据不敏感的业务（例如把 Redis 当做纯缓存使用），可以关闭 AOF 和 AOF rewrite","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 实例不要部署在虚拟机上：fork 的耗时也与系统也有关，虚拟机比物理机耗时更久","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"降低主从库全量同步的概率：适当调大 repl-backlog-size 参数，避免主从全量同步","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"开启内存大页","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除了上面讲到的子进程 RDB 和 AOF rewrite 期间，fork 耗时导致的延时变大之外，这里还有一个方面也会导致性能问题，这就是操作系统是否开启了","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"内存大页机制","attrs":{}},{"type":"text","text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"什么是内存大页？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们都知道，应用程序向操作系统申请内存时，是按","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"内存页","attrs":{}},{"type":"text","text":"进行申请的，而常规的内存页大小是 4KB。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Linux 内核从 2.6.38 开始，支持了","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"内存大页机制","attrs":{}},{"type":"text","text":"，该机制允许应用程序以 2MB 大小为单位，向操作系统申请内存。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"应用程序每次向操作系统申请的内存单位变大了，但这也意味着申请内存的耗时变长。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这对 Redis 会有什么影响呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当 Redis 在执行后台 RDB 和 AOF rewrite 时，采用 fork 子进程的方式来处理。但主进程 fork 子进程后，此时的","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"主进程依旧是可以接收写请求的","attrs":{}},{"type":"text","text":"，而进来的写请求，会采用 Copy On Write（写时复制）的方式操作内存数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"也就是说，主进程一旦有数据需要修改，Redis 并不会直接修改现有内存中的数据，而是","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"先将这块内存数据拷贝出来，再修改这块新内存的数据","attrs":{}},{"type":"text","text":"，这就是所谓的「写时复制」。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"写时复制你也可以理解成，谁需要发生写操作，谁就需要先拷贝，再修改。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这样做的好处是，父进程有任何写操作，并不会影响子进程的数据持久化（子进程只持久化 fork 这一瞬间整个实例中的所有数据即可，不关心新的数据变更，因为子进程只需要一份内存快照，然后持久化到磁盘上）。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但是请注意，主进程在拷贝内存数据时，这个阶段就涉及到新内存的申请，如果此时操作系统开启了内存大页，那么在此期间，客户端即便只修改 10B 的数据，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"Redis 在申请内存时也会以 2MB 为单位向操作系统申请，申请内存的耗时变长，进而导致每个写请求的延迟增加，影响到 Redis 性能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"同样地，如果这个写请求操作的是一个 bigkey，那主进程在拷贝这个 bigkey 内存块时，一次申请的内存会更大，时间也会更久。可见，bigkey 在这里又一次影响到了性能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/40/4020506b7da20ae3bd4c2e76b4c867ba.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那如何解决这个问题？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"很简单，你只需要关闭内存大页机制就可以了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，你需要查看 Redis 机器是否开启了内存大页：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"$ cat /sys/kernel/mm/transparent_hugepage/enabled\n[always] madvise never","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果输出选项是 always，就表示目前开启了内存大页机制，我们需要关掉它：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"$ echo never > /sys/kernel/mm/transparent_hugepage/enabled","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其实，操作系统提供的内存大页机制，其优势是，可以在一定程序上降低应用程序申请内存的次数。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但是对于 Redis 这种对性能和延迟极其敏感的数据库来说，我们希望 Redis 在每次申请内存时，耗时尽量短，所以我不建议你在 Redis 机器上开启这个机制。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"开启AOF","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"前面我们分析了 RDB 和 AOF rewrite 对 Redis 性能的影响，主要关注点在 fork 上。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其实，关于数据持久化方面，还有影响 Redis 性能的因素，这次我们重点来看 AOF 数据持久化。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你的 AOF 配置不合理，还是有可能会导致性能问题。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当 Redis 开启 AOF 后，其工作原理如下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 执行写命令后，把这个命令写入到 AOF 文件内存中（write 系统调用）","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 根据配置的 AOF 刷盘策略，把 AOF 内存数据刷到磁盘上（fsync 系统调用）","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为了保证 AOF 文件数据的安全性，Redis 提供了 3 种刷盘机制：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"appendfsync always：主线程每次执行写操作后立即刷盘，此方案会占用比较大的磁盘 IO 资源，但数据安全性最高","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"appendfsync no：主线程每次写操作只写内存就返回，内存数据什么时候刷到磁盘，交由操作系统决定，此方案对性能影响最小，但数据安全性也最低，Redis 宕机时丢失的数据取决于操作系统刷盘时机","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"appendfsync everysec：主线程每次写操作只写内存就返回，然后由后台线程每隔 1 秒执行一次刷盘操作（触发fsync系统调用），此方案对性能影响相对较小，但当 Redis 宕机时会丢失 1 秒的数据","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下面我们依次来分析，这几个机制对性能的影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你的 AOF 配置为 appendfsync always，那么 Redis 每处理一次写操作，都会把这个命令写入到磁盘中才返回，整个过程都是在主线程执行的，这个过程必然会加重 Redis 写负担。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"原因也很简单，操作磁盘要比操作内存慢几百倍，采用这个配置会严重拖慢 Redis 的性能，因此我不建议你把 AOF 刷盘方式配置为 always。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们接着来看 appendfsync no 配置项。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在这种配置下，Redis 每次写操作只写内存，什么时候把内存中的数据刷到磁盘，交给操作系统决定，此方案对 Redis 的性能影响最小，但当 Redis 宕机时，会丢失一部分数据，为了数据的安全性，一般我们也不采取这种配置。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你的 Redis 只用作纯缓存，对于数据丢失不敏感，采用配置 appendfsync no 也是可以的。","attrs":{}}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"看到这里，我猜你肯定和大多数人的想法一样，选比较折中的方案 appendfsync everysec 就没问题了吧？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个方案优势在于，Redis 主线程写完内存后就返回，具体的刷盘操作是放到后台线程中执行的，后台线程每隔 1 秒把内存中的数据刷到磁盘中。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这种方案既兼顾了性能，又尽可能地保证了数据安全，是不是觉得很完美？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"但是，这里我要给你泼一盆冷水了，采用这种方案你也要警惕一下，因为这种方案还是存在导致 Redis 延迟变大的情况发生，甚至会阻塞整个 Redis。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这是为什么？我把 AOF 最耗时的刷盘操作，放到后台线程中也会影响到 Redis 主线程？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你试想这样一种情况：当 Redis 后台线程在执行 AOF 文件刷盘时，如果此时磁盘的 IO 负载很高，那这个后台线程在执行刷盘操作（fsync系统调用）时就会被阻塞住。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时的主线程依旧会接收写请求，紧接着，主线程又需要把数据写到文件内存中（write 系统调用），","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"但此时的后台子线程由于磁盘负载过高，导致 fsync 发生阻塞，迟迟不能返回，那主线程在执行 write 系统调用时，也会被阻塞住","attrs":{}},{"type":"text","text":"，直到后台线程 fsync 执行完成后，主线程执行 write 才能成功返回。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"看到了么？在这个过程中，主线程依旧有阻塞的风险。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，尽管你的 AOF 配置为 appendfsync everysec，也不能掉以轻心，要警惕磁盘压力过大导致的 Redis 有性能问题。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/54/54d128ba902795a6000bc66cb7905729.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那什么情况下会导致磁盘 IO 负载过大？以及如何解决这个问题呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我总结了以下几种情况，你可以参考进行问题排查：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"子进程正在执行 AOF rewrite，这个过程会占用大量的磁盘 IO 资源","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"有其他应用程序在执行大量的写文件操作，也会占用磁盘 IO 资源","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"对于情况1，说白了就是，Redis 的 AOF 后台子线程刷盘操作，撞上了子进程 AOF rewrite！","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这怎么办？难道要关闭 AOF rewrite 才行？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"幸运的是，Redis 提供了一个配置项，当子进程在 AOF rewrite 期间，可以让后台子线程不执行刷盘（不触发 fsync 系统调用）操作。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这相当于在 AOF rewrite 期间，临时把 appendfsync 设置为了 none，配置如下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"# AOF rewrite 期间，AOF 后台子线程不进行刷盘操作\n# 相当于在这期间，临时把 appendfsync 设置为了 none\nno-appendfsync-on-rewrite yes","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，开启这个配置项，在 AOF rewrite 期间，如果实例发生宕机，那么此时会丢失更多的数据，性能和数据安全性，你需要权衡后进行选择。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果占用磁盘资源的是其他应用程序，那就比较简单了，你需要定位到是哪个应用程序在大量写磁盘，然后把这个应用程序迁移到其他机器上执行就好了，避免对 Redis 产生影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，如果你对 Redis 的性能和数据安全都有很高的要求，那么我建议从","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"硬件层面","attrs":{}},{"type":"text","text":"来优化，更换为 SSD 磁盘，提高磁盘的 IO 能力，保证 AOF 期间有充足的磁盘资源可以使用。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"绑定CPU","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"很多时候，我们在部署服务时，为了提高服务性能，降低应用程序在多个 CPU 核心之间的上下文切换带来的性能损耗，通常采用的方案是进程绑定 CPU 的方式提高性能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但在部署 Redis 时，如果你需要绑定 CPU 来提高其性能，我建议你仔细斟酌后再做操作。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为什么？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因为 Redis 在绑定 CPU 时，是有很多考究的，如果你不了解 Redis 的运行原理，随意绑定 CPU 不仅不会提高性能，甚至有可能会带来相反的效果。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们都知道，一般现代的服务器会有多个 CPU，而每个 CPU 又包含多个物理核心，每个物理核心又分为多个逻辑核心，每个物理核下的逻辑核共用 L1/L2 Cache。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而 Redis Server 除了主线程服务客户端请求之外，还会创建子进程、子线程。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其中子进程用于数据持久化，而子线程用于执行一些比较耗时操作，例如异步释放 fd、异步 AOF 刷盘、异步 lazy-free 等等。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你把 Redis 进程只绑定了一个 CPU 逻辑核心上，那么当 Redis 在进行数据持久化时，fork 出的子进程会继承父进程的 CPU 使用偏好。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"而此时的子进程会消耗大量的 CPU 资源进行数据持久化（把实例数据全部扫描出来需要耗费CPU），这就会导致子进程会与主进程发生 CPU 争抢，进而影响到主进程服务客户端请求，访问延迟变大。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这就是 Redis 绑定 CPU 带来的性能问题。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那如何解决这个问题呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你确实想要绑定 CPU，可以优化的方案是，不要让 Redis 进程只绑定在一个 CPU 逻辑核上，而是绑定在多个逻辑核心上，而且，绑定的多个逻辑核心最好是同一个物理核心，这样它们还可以共用 L1/L2 Cache。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，即便我们把 Redis 绑定在多个逻辑核心上，也只能在一定程度上缓解主线程、子进程、后台线程在 CPU 资源上的竞争。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因为这些子进程、子线程还是会在这多个逻辑核心上进行切换，存在性能损耗。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如何再进一步优化？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可能你已经想到了，我们是否可以让主线程、子进程、后台线程，分别绑定在固定的 CPU 核心上，不让它们来回切换，这样一来，他们各自使用的 CPU 资源互不影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其实，这个方案 Redis 官方已经想到了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 在 6.0 版本已经推出了这个功能，我们可以通过以下配置，对主线程、后台线程、后台 RDB 进程、AOF rewrite 进程，绑定固定的 CPU 逻辑核心：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"# Redis Server 和 IO 线程绑定到 CPU核心 0,2,4,6\nserver_cpulist 0-7:2\n\n# 后台子线程绑定到 CPU核心 1,3\nbio_cpulist 1,3\n\n# 后台 AOF rewrite 进程绑定到 CPU 核心 8,9,10,11\naof_rewrite_cpulist 8-11\n\n# 后台 RDB 进程绑定到 CPU 核心 1,10,11\n# bgsave_cpulist 1,10-1","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你使用的正好是 Redis 6.0 版本，就可以通过以上配置，来进一步提高 Redis 性能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这里我需要提醒你的是，一般来说，Redis 的性能已经足够优秀，除非你对 Redis 的性能有更加严苛的要求，否则不建议你绑定 CPU。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从上面的分析你也能看出，绑定 CPU 需要你对计算机体系结构有非常清晰的了解，否则谨慎操作。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们继续分析还有什么场景会导致 Redis 变慢。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"使用Swap","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你发现 Redis 突然变得非常慢，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"每次的操作耗时都达到了几百毫秒甚至秒级","attrs":{}},{"type":"text","text":"，那此时你就需要检查 Redis 是否使用到了 Swap，在这种情况下 Redis 基本上已经无法提供高性能的服务了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"什么是 Swap？为什么使用 Swap 会导致 Redis 的性能下降？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你对操作系统有些了解，就会知道操作系统为了缓解内存不足对应用程序的影响，允许把一部分内存中的数据换到磁盘上，以达到应用程序对内存使用的缓冲，这些内存数据被换到磁盘上的区域，就是 Swap。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"问题就在于，当内存中的数据被换到磁盘上后，Redis 再访问这些数据时，就需要从磁盘上读取，访问磁盘的速度要比访问内存慢几百倍！","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"尤其是针对 Redis 这种对性能要求极高、性能极其敏感的数据库来说，这个操作延时是无法接受的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时，你需要检查 Redis 机器的内存使用情况，确认是否存在使用了 Swap。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你可以通过以下方式来查看 Redis 进程是否使用到了 Swap：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"# 先找到 Redis 的进程 ID\n$ ps -aux | grep redis-server\n\n# 查看 Redis Swap 使用情况\n$ cat /proc/$pid/smaps | egrep '^(Swap|Size)'","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"输出结果如下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"Size: 1256 kB\nSwap: 0 kB\nSize: 4 kB\nSwap: 0 kB\nSize: 132 kB\nSwap: 0 kB\nSize: 63488 kB\nSwap: 0 kB\nSize: 132 kB\nSwap: 0 kB\nSize: 65404 kB\nSwap: 0 kB\nSize: 1921024 kB\nSwap: 0 kB\n...","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个结果会列出 Redis 进程的内存使用情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"每一行 Size 表示 Redis 所用的一块内存大小，Size 下面的 Swap 就表示这块 Size 大小的内存，有多少数据已经被换到磁盘上了，如果这两个值相等，说明这块内存的数据都已经完全被换到磁盘上了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果只是少量数据被换到磁盘上，例如每一块 Swap 占对应 Size 的比例很小，那影响并不是很大。","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"如果是几百兆甚至上 GB 的内存被换到了磁盘上","attrs":{}},{"type":"text","text":"，那么你就需要警惕了，这种情况 Redis 的性能肯定会急剧下降。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时的解决方案是：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"增加机器的内存，让 Redis 有足够的内存可以使用","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"整理内存空间，释放出足够的内存供 Redis 使用，然后释放 Redis 的 Swap，让 Redis 重新使用内存","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"释放 Redis 的 Swap 过程通常要重启实例，为了避免重启实例对业务的影响，一般会先进行主从切换，然后释放旧主节点的 Swap，重启旧主节点实例，待从库数据同步完成后，再进行主从切换即可。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可见，当 Redis 使用到 Swap 后，此时的 Redis 性能基本已达不到高性能的要求（你可以理解为武功被废），所以你也需要提前预防这种情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"预防的办法就是，你需要对 Redis 机器的内存和 Swap 使用情况进行监控，在内存不足或使用到 Swap 时报警出来，及时处理。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"碎片整理","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 的数据都存储在内存中，当我们的应用程序频繁修改 Redis 中的数据时，就有可能会导致 Redis 产生内存碎片。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"内存碎片会降低 Redis 的内存使用率，我们可以通过执行 INFO 命令，得到这个实例的内存碎片率：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"# Memory\nused_memory:5709194824\nused_memory_human:5.32G\nused_memory_rss:8264855552\nused_memory_rss_human:7.70G\n...\nmem_fragmentation_ratio:1.45","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个内存碎片率是怎么计算的？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"很简单，mem","attrs":{}},{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"fragmentation","attrs":{}},{"type":"text","text":"ratio = used","attrs":{}},{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"memory","attrs":{}},{"type":"text","text":"rss / used_memory。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其中 used","attrs":{}},{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"memory 表示 Redis 存储数据的内存大小，而 used","attrs":{}},{"type":"text","text":"memory_rss 表示操作系统实际分配给 Redis 进程的大小。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果 mem","attrs":{}},{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"fragmentation","attrs":{}},{"type":"text","text":"ratio > 1.5，说明内存碎片率已经超过了 50%，这时我们就需要采取一些措施来降低内存碎片了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"解决的方案一般如下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":"1","normalizeStart":1},"content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"如果你使用的是 Redis 4.0 以下版本，只能通过重启实例来解决","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"如果你使用的是 Redis 4.0 版本，它正好提供了自动碎片整理的功能，可以通过配置开启碎片自动整理","attrs":{}}]}],"attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"但是，开启内存碎片整理，它也有可能会导致 Redis 性能下降。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"原因在于，Redis 的碎片整理工作是也在","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"主线程","attrs":{}},{"type":"text","text":"中执行的，当其进行碎片整理时，必然会消耗 CPU 资源，产生更多的耗时，从而影响到客户端的请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，当你需要开启这个功能时，最好提前测试评估它对 Redis 的影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 碎片整理的参数配置如下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"shell"},"content":[{"type":"text","text":"# 开启自动内存碎片整理（总开关）\nactivedefrag yes\n\n# 内存使用 100MB 以下，不进行碎片整理\nactive-defrag-ignore-bytes 100mb\n\n# 内存碎片率超过 10%，开始碎片整理\nactive-defrag-threshold-lower 10\n# 内存碎片率超过 100%，尽最大努力碎片整理\nactive-defrag-threshold-upper 100\n\n# 内存碎片整理占用 CPU 资源最小百分比\nactive-defrag-cycle-min 1\n# 内存碎片整理占用 CPU 资源最大百分比\nactive-defrag-cycle-max 25\n\n# 碎片整理期间，对于 List/Set/Hash/ZSet 类型元素一次 Scan 的数量\nactive-defrag-max-scan-fields 1000","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你需要结合 Redis 机器的负载情况，以及应用程序可接受的延迟范围进行评估，合理调整碎片整理的参数，尽可能降低碎片整理期间对 Redis 的影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"网络带宽过载","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果以上产生性能问题的场景，你都规避掉了，而且 Redis 也稳定运行了很长时间，但在某个时间点之后开始，操作 Redis 突然开始变慢了，而且一直持续下去，这种情况又是什么原因导致？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此时你需要排查一下 Redis 机器的网络带宽是否过载，是否存在某个实例把整个机器的网路带宽占满的情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"网络带宽过载的情况下，服务器在 TCP 层和网络层就会出现数据包发送延迟、丢包等情况。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Redis 的高性能，除了操作内存之外，就在于网络 IO 了，如果网络 IO 存在瓶颈，那么也会严重影响 Redis 的性能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果确实出现这种情况，你需要及时确认占满网络带宽 Redis 实例，如果属于正常的业务访问，那就需要及时扩容或迁移实例了，避免因为这个实例流量过大，影响这个机器的其他实例。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"运维层面，你需要对 Redis 机器的各项指标增加监控，包括网络流量，在网络流量达到一定阈值时提前报警，及时确认和扩容。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"其他原因","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"好了，以上这些方面就是如何排查 Redis 延迟问题的思路和路径。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除了以上这些，还有一些比较小的点，你也需要注意一下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"1) 频繁短连接","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你的业务应用，应该使用长连接操作 Redis，避免频繁的短连接。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"频繁的短连接会导致 Redis 大量时间耗费在连接的建立和释放上，TCP 的三次握手和四次挥手同样也会增加访问延迟。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2) 运维监控","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"前面我也提到了，要想提前预知 Redis 变慢的情况发生，必不可少的就是做好完善的监控。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"监控其实就是对采集 Redis 的各项运行时指标，通常的做法是监控程序定时采集 Redis 的 INFO 信息，然后根据 INFO 信息中的状态数据做数据展示和报警。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这里我需要提醒你的是，在写一些监控脚本，或使用开源的监控组件时，也不能掉以轻心。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在写监控脚本访问 Redis 时，尽量采用长连接的方式采集状态信息，避免频繁短连接。同时，你还要注意控制访问 Redis 的频率，避免影响到业务请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在使用一些开源的监控组件时，最好了解一下这些组件的实现原理，以及正确配置这些组件，防止出现监控组件发生 Bug，导致短时大量操作 Redis，影响 Redis 性能的情况发生。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们当时就发生过，DBA 在使用一些开源组件时，因为配置和使用问题，导致监控程序频繁地与 Redis 建立和断开连接，导致 Redis 响应变慢。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"3）其它程序争抢资源","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后需要提醒你的是，你的 Redis 机器最好专项专用，只用来部署 Redis 实例，不要部署其他应用程序，尽量给 Redis 提供一个相对「安静」的环境，避免其它程序占用 CPU、内存、磁盘资源，导致分配给 Redis 的资源不足而受到影响。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"总结","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"好了，以上就是我总结的在使用 Redis 过程中，常见的可能导致延迟、甚至阻塞的问题场景，以及如何快速定位和分析这些问题，并且针对性地提供了解决方案。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这里我也汇总成了思维导图，方便你在排查 Redis 性能问题时，快速地去分析和定位。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/f3/f3ce2d49fd3f146a4ea12511a9fbff48.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这里再简单总结一下，Redis 的性能问题，既涉及到了业务开发人员的使用方面，也涉及到了 DBA 的运维方面。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"作为业务开发人员，我们需要了解 Redis 的基本原理，例如各个命令执行的时间复杂度、数据过期策略、数据淘汰策略等，从而更合理地使用 Redis 命令，并且结合业务场景进行优化。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"作为 DBA 和运维人员，需要了解 Redis 运行机制，例如数据持久化、内存碎片整理、进程绑核配置。除此之外，还需要了解操作系统相关知识，例如写时复制、内存大页、Swap 机制等等。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"同时，DBA 在部署 Redis 时，需要提前对进行容量规划，预留足够的机器资源，还要对 Redis 机器和实例做好完善的监控，这样才能尽可能地保证 Redis 的稳定运行。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"后记","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你能耐心地看到这里，想必你肯定已经对 Redis 的性能调优有了很大的收获。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你应该也发现了，Redis 的性能问题，涉及到的知识点非常广，几乎涵盖了 CPU、内存、网络、甚至磁盘的方方面面，同时，你还需要了解计算机的体系结构，以及操作系统的各种机制。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从资源使用角度来看，包含的知识点如下：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU 相关：使用复杂度过高命令、数据的持久化，都与耗费过多的 CPU 资源有关","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"内存相关：bigkey 内存的申请和释放、数据过期、数据淘汰、碎片整理、内存大页、内存写时复制都与内存息息相关","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"磁盘相关：数据持久化、AOF 刷盘策略，也会受到磁盘的影响","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"网络相关：短连接、实例流量过载、网络流量过载，也会降低 Redis 性能","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"计算机系统：CPU 结构、内存分配，都属于最基础的计算机系统知识","attrs":{}}]}],"attrs":{}},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"操作系统：写时复制、内存大页、Swap、CPU 绑定，都属于操作系统层面的知识","attrs":{}}]}],"attrs":{}}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"没想到吧？Redis 为了把性能做到极致，涉及到了这么多项优化。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果这篇文章内容，你能吸收 90% 以上，说明你对 Redis 原理、计算机基础、操作系统都已经有了较为深刻的理解。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你能吸收 50% 左右，那你可以好好梳理一下，哪些方面是自己的知识盲区，这样可以针对性地去学习。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你吸收的只在 30% 以下，那么你可以先从 Redis 的基本原理出发，先了解 Redis 的各种机制，进而思考 Redis 为了提高性能，为什么使用这些机制？这些机制又是利用了计算机和操作系统的哪些特性去做的？进而一步步地去扩充你的知识体系，这是一个非常高效的学习路径。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"由于篇幅限制，关于 Redis 的很多细节无法全部展开，其实，这篇文章提到的每一个导致 Redis 性能问题的场景，如果展开来讲，都可以写出一篇文章出来。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"例如，关于 Redis 进程绑定 CPU，以及操作系统使用 Swap，其实这些还涉及到了非一致性内存访问 NUMA 架构的影响，其中也有很多细节没有展开来讲。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"这篇文章主要站在一个宏观层面进行性能分析，如果你想了解更多细节，欢迎关注我的公众号「水滴与银弹」，我会在后续的文章中，带你深度剖析 Redis 的运行原理，以及产生性能问题的更多细节。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"关注「水滴与银弹」公众号，第一时间获取优质技术干货。7年资深后端研发，用简单的方式把技术讲清楚。","attrs":{}}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}