在现代应用程序中,缓存被广泛应用以提高性能和减轻后端数据库的压力。本文将探讨面对 Redis 集群缓存分配不均问题时的解决方法。
我们的 Redis 集群部署包括 3 主 3 从,每个节点分配内存 4G(服务器总内存32G),内存淘汰策略相同:volatile-lru
。
在性能测试过程中,通过 pinpoint 监控发现报错:OOM
Error in execution; nested exception is io.lettuce.core.RedisCommandExecutionException: OOM command not allowed when used memory> 'maxmemory'
清空缓存后,再次执行性能测试一段时间,使用 info memory
查看各节点内存使用情况,发现 2、6 节点内存快速上升,其它节点内存使用率很低。
例如,2 节点内存达到 3.81G,6 节点内存达到 3.8G,而其他 4 个节点内存仅为 80M 左右。
[root@iZ2ze3e0bvnd1hf5xbkxn3Z redis01]# ./redis-cli -h 192.168.0.213 -p 6692 -c -a Tiye@54L2!
Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe.
192.168.0.213:6692> info memory
# Memory
used_memory:4095914816
used_memory_human:3.81G
used_memory_rss:4473937920
used_memory_rss_human:4.17G
used_memory_peak:4105896552
used_memory_peak_human:3.82G
used_memory_peak_perc:99.76%
used_memory_overhead:5159996
used_memory_startup:1483832
used_memory_dataset:4090754820
used_memory_dataset_perc:99.91%
allocator_allocated:4095978056
allocator_active:4728315904
allocator_resident:4765335552
total_system_memory:33019609088
total_system_memory_human:30.75G
used_memory_lua:30720
used_memory_lua_human:30.00K
used_memory_scripts:0
used_memory_scripts_human:0B
number_of_cached_scripts:0
maxmemory:4096000000
maxmemory_human:3.81G
maxmemory_policy:volatile-lru
allocator_frag_ratio:1.15
allocator_frag_bytes:632337848
allocator_rss_ratio:1.01
allocator_rss_bytes:37019648
rss_overhead_ratio:0.94
rss_overhead_bytes:-291397632
mem_fragmentation_ratio:1.09
mem_fragmentation_bytes:378064128
mem_not_counted_for_evict:0
mem_replication_backlog:1048576
mem_clients_slaves:20512
mem_clients_normal:2603516
mem_aof_buffer:0
mem_allocator:jemalloc-5.1.0
active_defrag_running:0
lazyfree_pending_objects:0
lazyfreed_objects:0
[root@iZ2ze3e0bvnd1hf5xbkxn3Z redis02]# ./redis-cli -h 192.168.0.213 -p 6691 -c -a Tiye@54L2!
Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe.
192.168.0.213:6691> info memory
# Memory
used_memory:84125536
used_memory_human:80.23M
used_memory_rss:107753472
used_memory_rss_human:102.76M
used_memory_peak:109050608
used_memory_peak_human:104.00M
used_memory_peak_perc:77.14%
used_memory_overhead:9292232
used_memory_startup:1483832
used_memory_dataset:74833304
used_memory_dataset_perc:90.55%
allocator_allocated:84208120
allocator_active:102572032
allocator_resident:108343296
total_system_memory:33019609088
total_system_memory_human:30.75G
used_memory_lua:30720
used_memory_lua_human:30.00K
used_memory_scripts:0
used_memory_scripts_human:0B
number_of_cached_scripts:0
maxmemory:4096000000
maxmemory_human:3.81G
maxmemory_policy:volatile-lru
allocator_frag_ratio:1.22
allocator_frag_bytes:18363912
allocator_rss_ratio:1.06
allocator_rss_bytes:5771264
rss_overhead_ratio:0.99
rss_overhead_bytes:-589824
mem_fragmentation_ratio:1.28
mem_fragmentation_bytes:23669520
mem_not_counted_for_evict:0
mem_replication_backlog:1048576
mem_clients_slaves:20512
mem_clients_normal:2603360
mem_aof_buffer:0
mem_allocator:jemalloc-5.1.0
active_defrag_running:0
lazyfree_pending_objects:0
lazyfreed_objects:0
cluster nodes
命令获取 Redis 集群中所有节点的信息,并判断各节点主从关系。
[root@iZ2ze3e0bvnd1hf5xbkxn3Z redis06]# ./redis-cli -h 192.168.0.213 -p 6696 -c -a Tiye@54L2!
Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe.
192.168.0.213:6696> cluster nodes
e73a5ec3e26ed23e9b4bf56811527c8820a7bd79 192.168.0.213:6696@16696 myself,slave e2a678a004bc99e76180a16a6a41e2cad1c96052 0 1691992895000 2 connected
25317f0f8f7b2eebdbdc0914c659ab96ed3dab18 192.168.0.213:6693@16693 master - 0 1691992897074 3 connected 10923-16383
27aba75f54cccbb42125edb20f2f9d7c2f777d6c 192.168.0.213:6695@16695 slave 5e08015f75cdb05b1c7ed78dead1d85cdb0e838f 0 1691992895070 1 connected
5e08015f75cdb05b1c7ed78dead1d85cdb0e838f 192.168.0.213:6691@16691 master - 0 1691992894068 1 connected 0-5460
053916b96426f790244d984cad3f69f9151e4ece 192.168.0.213:6694@16694 slave 25317f0f8f7b2eebdbdc0914c659ab96ed3dab18 0 1691992896072 3 connected
e2a678a004bc99e76180a16a6a41e2cad1c96052 192.168.0.213:6692@16692 master - 0 1691992894000 2 connected 5461-10922
根据节点信息中的master字段判断该节点是主节点还是从节点。如果节点的master字段为空,则表示它是主节点;如果master字段不为空,则表示它是从节点,并且该字段的值是对应的主节点ID。
e2a678a004bc99e76180a16a6a41e2cad1c96052 节点是主节点,e73a5ec3e26ed23e9b4bf56811527c8820a7bd79 节点是从节点,并且它的主节点是 e2a678a004bc99e76180a16a6a41e2cad1c96052
综上判断:redis集群6个节点,其中2节点的从节点是6节点。
在 Redis 集群中的任意一个节点上执行 cluster slots
命令,分析哈希槽数量分配是否均匀
192.168.0.213:6692> cluster slots
1) 1) (integer) 0
2) (integer) 5460
3) 1) "192.168.0.213"
2) (integer) 6691
3) "5e08015f75cdb05b1c7ed78dead1d85cdb0e838f"
4) 1) "192.168.0.213"
2) (integer) 6695
3) "27aba75f54cccbb42125edb20f2f9d7c2f777d6c"
2) 1) (integer) 5461
2) (integer) 10922
3) 1) "192.168.0.213"
2) (integer) 6692
3) "e2a678a004bc99e76180a16a6a41e2cad1c96052"
4) 1) "192.168.0.213"
2) (integer) 6696
3) "e73a5ec3e26ed23e9b4bf56811527c8820a7bd79"
3) 1) (integer) 10923
2) (integer) 16383
3) 1) "192.168.0.213"
2) (integer) 6693
3) "25317f0f8f7b2eebdbdc0914c659ab96ed3dab18"
4) 1) "192.168.0.213"
2) (integer) 6694
3) "053916b96426f790244d984cad3f69f9151e4ece"
根据提供的哈希槽信息,我们可以计算出以下三个哈希槽的数量:
综上判断:这三个哈希槽的数量分别是5461、5462和5461。即,哈希槽数量分配均匀。
跟产品开发人员沟通,没有做predixy适配,但可以尝试配置。最后发现产品启动服务报错,放弃该方案。尝试过程记录如下:
wget https://github.com/joyieldInc/predixy/releases/download/1.0.5/predixy-1.0.5-bin-amd64-linux.tar.gz
tar zxvf predixy-1.0.5-bin-amd64-linux.tar.gz
cd predixy-1.0.5
① 编辑conf/predixy.conf文件
Include cluster.conf #Include cluster.conf解开注释
# Include sentinel.conf
# Include try.conf #注释掉Include try.conf
## Worker threads
WorkerThreads 3 #从1改为3,表示开启的进程数
② 编辑conf/cluster.conf文件
lusterServerPool {
MasterReadPriority 0 #设置为0代表开启读写分离
Password Tiye@54L2! #redis集群密码
StaticSlaveReadPriority 50
DynamicSlaveReadPriority 50
RefreshInterval 1
ServerTimeout 1
ServerFailureLimit 10
ServerRetryTimeout 1
KeepAlive 120
Servers {
+ 192.168.0.213:6691 #redis集群节点
+ 192.168.0.213:6692
+ 192.168.0.213:6693
+ 192.168.0.213:6694
+ 192.168.0.213:6695
+ 192.168.0.213:6696
}
}
配置文件解析文档:https://github.com/joyieldInc/predixy/blob/master/doc/config_CN.md
① 创建启动和停止脚本,放到predixy-1.0.5目录下
mkdir -p /opt/predixy-1.0.5/logs/
cd /opt/predixy-1.0.5/bin
② 创建启动脚本up.sh
#!/bin/bash
path=`pwd`
nohup $path/bin/predixy conf/predixy.conf > $path/logs/predixy.log 2>&1 &
③ 查看日志
tail -f logs/predixy.log
④ 创建停止脚本down.sh
#!/bin/bash
path=`pwd`
pid=`ps -ef | grep $path/bin/predixy | grep -v grep | awk '{print $2}'`
kill -9 $pid
./redis-cli -p 7617
127.0.0.1:7617> mset b1 b2
OK
127.0.0.1:7617> get b1
"b2"
将应用配置文件中的集群地址修改为predixy地址,应用服务启动报错:
...Autowired annotation is not supported on static fields...
尝试缩减为 4 个节点,以提升单个节点的内存配置。但创建集群时提示至少需要 3 个主节点,因此该方案也不可行。
*** ERROR: Invalid configuration for cluster creation.
*** Redis Cluster requires at least 3 master nodes.
*** This is not possible with 4 nodes and 1 replicas per node.
*** At least 6 nodes are required
最终解决方案是修改集群节点内存:1、3、4、5 节点分配内存 3G,2、6 节点分配内存 8G。
面对 Redis 集群缓存分配不均的问题,我们可以通过逐步的优化方法来解决。首先,了解节点信息、主从关系和缓存占用情况,然后分析哈希槽分布情况,尝试不同的优化方案,最后通过调整内存分配来解决问题。