前言
redis是一款非常流行的kv数据库,以高性能著称,其高吞吐、低延迟等特性让广大开发者趋之若鹜,每每看到别人发出的redis故障报告都让我产生一种居安思危,以史为鉴的危机感,恰逢今年十一西安烟雨不断,抽时间学习了几个redis监控命令,和大家分享一波。
redis-cli --stat【连续统计】
redis-cli --stat 默认每秒输出一条新行,其中包含有用信息和每个采集点的请求次数差异。使用此命令可以轻松了解内存使用情况、客户端连接计数以及有关已连接 Redis 数据库的各种其他统计信息。
可以使用-i修改采样频率,默认值为1秒,如下面这个命令代表每2s采集一次数据:
redis-cli --stat -i 2 ------- data ------ --------------------- load -------------------- - child - keys mem clients blocked requests connections 8890 131.89M 47 0 1705992846 (+0) 2595 8890 131.93M 47 0 1705992897 (+51) 2595 8890 131.93M 47 0 1705992954 (+57) 2595 8890 131.97M 47 0 1705992991 (+37) 2595 8890 131.89M 47 0 1705993043 (+52) 2595 8890 131.97M 47 0 1705993088 (+45) 2595 8890 132.01M 47 0 1705993122 (+34) 2595 8890 132.01M 47 0 1705993168 (+46) 2595 8890 132.01M 47 0 1705993194 (+26) 2595 8890 131.93M 47 0 1705993267 (+73) 2595
redis-cli --bigkeys【统计大key】
这个命令用作键空间分析器,它扫描数据集中的大键,但也提供有关数据集所包含的数据类型的信息。
# redis-cli --bigkeys # Scanning the entire keyspace to find biggest keys as well as # average sizes per key type. You can use -i 0.1 to sleep 0.1 sec # per 100 SCAN commands (not usually needed). [00.00%] Biggest hash found so far '"hash_big"' with 6 fields [00.00%] Biggest set found so far '"set_big"' with 6 members [00.00%] Biggest string found so far '"string_big"' with 979 bytes [00.00%] Biggest string found so far '"string_big_2"' with 1365 bytes -------- summary ------- Sampled 5 keys in the keyspace! Total key length in bytes is 38 (avg len 7.60) Biggest hash found '"hash_big"' has 6 fields Biggest string found '"string_big_2"' has 1365 bytes Biggest set found '"set_big"' has 6 members 0 lists with 0 items (00.00% of keys, avg size 0.00) 1 hashs with 6 fields (20.00% of keys, avg size 6.00) 3 strings with 2420 bytes (60.00% of keys, avg size 806.67) 0 streams with 0 entries (00.00% of keys, avg size 0.00) 1 sets with 6 members (20.00% of keys, avg size 6.00) 0 zsets with 0 members (00.00% of keys, avg size 0.00)
在输出的第一部分中,将报告遇到的每个大于前一个较大key(相同类型)的新key。摘要部分提供有关 Redis 实例内数据的一般统计信息。
该程序使用 SCAN 命令,因此它可以在繁忙的服务器上执行而不会影响操作,当然也可以使用-i选项来限制每个 SCAN 命令的指定秒数部分的扫描过程。
例如,redis-cli --bigkeys -i 1 代表每次SCAN执行之后sleep 1s。
可以看到--bigkeys给出了每种数据结构的top 1 bigkey,同时给出了每种数据类型的键值个数以及平均大小。
redis-cli monitor【监控命令执行】
redis-cli monitor 1665128881.578949 [0 127.0.0.1:46046] "COMMAND" "DOCS" 1665128885.870333 [0 127.0.0.1:46046] "get" "a" 1665128891.200705 [0 127.0.0.1:46046] "set" "a" "asdfasdfasd" "asdfasdf" 1665128897.234390 [0 127.0.0.1:46046] "sadd" "test" "aaa" 1665128902.439247 [0 127.0.0.1:46046] "smembers" "test" 1665128906.257225 [0 127.0.0.1:46046] "smembers" "test" 1665128910.073980 [0 127.0.0.1:46046] "smembers" "test" 1665128914.688753 [0 127.0.0.1:46046] "hget" "all" "hello" 1665128918.006031 [0 127.0.0.1:46046] "hget" "all" "hello"
可以看到目前smembers和hget命令执行的比较频繁,可能是异常流量导致,需要引起我们的注意了。
更方便的是redis-cli monitor可以和管道配合使用,比如redis-cli monitor | grep goods_test_001
redis-cli monitor |grep goods_test_001 1665129150.063322 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129150.935202 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129151.486148 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129152.012097 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129152.550077 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129153.059130 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129153.595023 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129154.166608 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129154.687753 [0 127.0.0.1:46046] "get" "goods_test_001" 1665129155.204012 [0 127.0.0.1:46046] "get" "goods_test_001"
结合grep goods_test_001可以发现goods_test_001这个key当前有大量的读请求。
Pub/sub mode【发布订阅模式】
redis-cli可以用来发布/订阅消息,如果你的系统中使用了redis的发布订阅功能,可以使用redis-cli的这一特性来进行一些调试工作。
比如,使用redis-cli发布一条消息到mychannel
redis-cli publish mychannel helloworld
同样的,使用redis-cli订阅mychannel发来的消息
redis-cli subscribe mychannel Reading messages... (press Ctrl-C to quit) 1) "subscribe" 2) "mychannel" 3) (integer) 1 1) "message" 2) "mychannel" 3) "helloworld"
Monitoring the latency of Redis instances【监控延迟】
redis-cli提供了多种工具帮助我们发现延迟,涉及的指标有最小值、最大值、平均值、延迟分布情况等。
基本的延迟检查工具是redis-cli --latency。使用--latency,redis-cli 运行一个循环,以每秒100次的速度向redis发送PING命令,并测量收到回复的时间,统计信息在控制台中实时更新。
# redis-cli --latency min: 0, max: 3, avg: 0.28 (536 samples)
统计数据以毫秒为单位,上面的测试一共发了536个PING命令,最小响应时间为0毫秒(0不代表没有延迟,只是说毫秒统计不到),最大为3毫秒,平均值为0.28毫秒。
有时我们更希望看到redis延迟变化的趋势,这时--latency-history就可以派上用场,它的工作机制和--latency相同,只是每15秒(默认)重新开启一个测试会话。
redis-cli --latency-history min: 0, max: 7, avg: 0.25 (1432 samples) -- 15.00 seconds range min: 0, max: 1, avg: 0.24 (1435 samples) -- 15.00 seconds range min: 0, max: 15, avg: 0.27 (1429 samples) -- 15.01 seconds range min: 0, max: 5, avg: 0.28 (1431 samples) -- 15.01 seconds range min: 0, max: 5007, avg: 7.71 (839 samples) -- 15.01 seconds range min: 1, max: 18, avg: 3.58 (1092 samples) -- 15.01 seconds range min: 0, max: 13, avg: 3.56 (1093 samples) -- 15.01 seconds range min: 1, max: 15, avg: 3.61 (1090 samples) -- 15.00 seconds range min: 1, max: 17, avg: 3.60 (1091 samples) -- 15.01 seconds range min: 0, max: 26, avg: 2.57 (1178 samples) -- 15.00 seconds range
可以看到,上面每隔15秒输出一组数据,在第5个15秒开始时耗时明显增加。
内部还实现了另一个非比寻常的延迟检测工具,它不检查 Redis 实例的延迟,而是检查运行的计算机的延迟,此延迟是内核计划程序、虚拟机管理程序(如果是虚拟化实例)等所固有的。
redis称之为内在延迟,因为它对程序员来说基本上是不透明的,如果 redis 实例具有高延迟,检查其他因素之外,还值得检查内核本身的延迟。
通过测量内在延迟,我们就知道这是基准,redis 无法超越内核,使用redis-cli --intrinsic-latency <持续时间>开启测试,持续时间5秒。
redis-cli --intrinsic-latency 5 Max latency so far: 1 microseconds. Max latency so far: 16 microseconds. Max latency so far: 70 microseconds. Max latency so far: 109 microseconds. Max latency so far: 145 microseconds. Max latency so far: 205 microseconds. Max latency so far: 283 microseconds. Max latency so far: 363 microseconds. Max latency so far: 2507 microseconds. Max latency so far: 4541 microseconds. 100063828 total runs (avg latency: 0.0500 microseconds / 49.97 nanoseconds per run). Worst run took 90878x longer than the average latency. # redis-cli --intrinsic-latency 5 Max latency so far: 1 microseconds. Max latency so far: 39 microseconds. Max latency so far: 41 microseconds. Max latency so far: 45 microseconds. Max latency so far: 62 microseconds. Max latency so far: 8839 microseconds. Max latency so far: 9357 microseconds. Max latency so far: 10310 microseconds. Max latency so far: 10322 microseconds. Max latency so far: 10573 microseconds. Max latency so far: 10682 microseconds. Max latency so far: 11177 microseconds. Max latency so far: 11514 microseconds. 35539207 total runs (avg latency: 0.1407 microseconds / 140.69 nanoseconds per run). Worst run took 81840x longer than the average latency.
注意:--intrinsic-latency只能在redis实例所在机器运行。
从上面的输出可以看到内核的最大延迟达到了11514微秒(115毫秒左右),也从侧面说明执行redis命令的最大延迟起码在115毫秒之上。
Replica mode【副本模式】
redis-cli --replica sending REPLCONF capa eof sending REPLCONF rdb-filter-only SYNC with master, discarding bytes of bulk transfer until EOF marker... SYNC done after 211 bytes. Logging commands from master. sending REPLCONF ACK 0 "ping" "SELECT","0" "set","a","b" "hset","hash","name","jack"
可以看到主节点上执行了set,hset等指令,命令行实时输出。
如果你正在开发一个跨机房同步的redis同步工具,当你的从节点未按预期收到指令时,就可以使用这一命令做一些调试和诊断,为了方便理解,我放一张老东家自研的redis跨机房同步工具流程图。
Performing an LRU simulation【模拟LRU访问】
该工具使用80/20法则来执行 GET 、SET操作 ,意味着 20% 的key将在 80% 的次数内被请求,这符合一般缓存场景中的请求分布。
我们假设给redis分配的内存为10兆,内存驱逐策略为allkeys-lru,预期有100万个key,期望命中率是90%,测试一下看是否符合预期:
# 设置最大内存10兆 config set maxmemory 10MB # lru-test redis-cli --lru-test 1000000 119250 Gets/sec | Hits: 43654 (36.61%) | Misses: 75596 (63.39%) 125250 Gets/sec | Hits: 46002 (36.73%) | Misses: 79248 (63.27%) 127500 Gets/sec | Hits: 46860 (36.75%) | Misses: 80640 (63.25%) 122500 Gets/sec | Hits: 45228 (36.92%) | Misses: 77272 (63.08%) 126750 Gets/sec | Hits: 46623 (36.78%) | Misses: 80127 (63.22%) 125250 Gets/sec | Hits: 46150 (36.85%) | Misses: 79100 (63.15%) 120000 Gets/sec | Hits: 43962 (36.63%) | Misses: 76038 (63.37%) 121000 Gets/sec | Hits: 44630 (36.88%) | Misses: 76370 (63.12%) 123250 Gets/sec | Hits: 45616 (37.01%) | Misses: 77634 (62.99%)
命中率明显不符合预期,36%离90%相差甚远,我们将maxmemory扩大一倍接着测试
# 设置最大内存20兆 config set maxmemory 20MB # lru-test redis-cli --lru-test 1000000 134500 Gets/sec | Hits: 65181 (48.46%) | Misses: 69319 (51.54%) 133500 Gets/sec | Hits: 86515 (64.81%) | Misses: 46985 (35.19%) 133000 Gets/sec | Hits: 98930 (74.38%) | Misses: 34070 (25.62%) 123500 Gets/sec | Hits: 95223 (77.10%) | Misses: 28277 (22.90%) 122000 Gets/sec | Hits: 94237 (77.24%) | Misses: 27763 (22.76%) 122250 Gets/sec | Hits: 94430 (77.24%) | Misses: 27820 (22.76%) 122500 Gets/sec | Hits: 94564 (77.20%) | Misses: 27936 (22.80%) 124000 Gets/sec | Hits: 95517 (77.03%) | Misses: 28483 (22.97%) 125000 Gets/sec | Hits: 96723 (77.38%) | Misses: 28277 (22.62%) 129000 Gets/sec | Hits: 99839 (77.39%) | Misses: 29161 (22.61%)
内存增加一倍以后命中率达到了77%左右,继续调整maxmemory直到符合预期。
redis-cli --lru-test切记不要在生产环境使用,会给服务器带来较大压力;