部署硬件:
3台服务器6节点,每个节点部署在7.6TB的nvme盘,每台机器内存512GB,CPU: 128核
使用的是txnKv模式
现象:
时延:
二分之一的tikv region-size暴涨
紧急暂停 balance-region
PD_down_peer_region_nums大于2
top size的region信息
top3 reigon size
{"count":3,"regions":[{"id":8056302,"start_key":"3031393635623534FF2D333433342D3736FF63342D383436322DFF6261663834373038FF37363533FD41F6FEFFA404000000004300FF0000590000000000FA","end_key":"3031393635623534FF2D333433342D3736FF63342D383436322DFF6261663834373038FF37363533FD41F74DFFF200000000004900FE","epoch":{"conf_ver":77441,"version":6230},"peers":[{"role_name":"Voter","id":8056303,"store_id":1},{"role_name":"Voter","id":8056304,"store_id":3},{"role_name":"Voter","id":8056305,"store_id":2}],"leader":{"role_name":"Voter","id":8056303,"store_id":1},"cpu_usage":0,"written_bytes":54731793,"read_bytes":3532,"written_keys":385,"read_keys":10,"approximate_size":33846,"approximate_keys":371361},{"id":8064784,"start_key":"3031393635623534FF2D333433342D3736FF63342D383436322DFF6261663834373038FF37363533FD41F6FEFFA404000000004300FF00001E0000000000FA","end_key":"3031393635623534FF2D333433342D3736FF63342D383436322DFF6261663834373038FF37363533FD41F6FEFFA404000000004300FF00001F0000000000FA","epoch":{"conf_ver":77441,"version":6234},"peers":[{"role_name":"Voter","id":8064785,"store_id":1},{"role_name":"Voter","id":8064786,"store_id":3},{"role_name":"Voter","id":8064787,"store_id":2}],"leader":{"role_name":"Voter","id":8064785,"store_id":1},"cpu_usage":0,"written_bytes":131681829,"read_bytes":990,"written_keys":613,"read_keys":0,"approximate_size":33832,"approximate_keys":40985},{"id":8064640,"start_key":"3031393635623534FF2D333433342D3736FF63342D383436322DFF6261663834373038FF37363533FD41F6FEFFA404000000004300FF0000540000000000FA","end_key":"3031393635623534FF2D333433342D3736FF63342D383436322DFF6261663834373038FF37363533FD41F6FEFFA404000000004300FF0000550000000000FA","epoch":{"conf_ver":77441,"version":6231},"peers":[{"role_name":"Voter","id":8064641,"store_id":1},{"role_name":"Voter","id":8064642,"store_id":3},{"role_name":"Voter","id":8064643,"store_id":2}],"leader":{"role_name":"Voter","id":8064641,"store_id":1},"cpu_usage":0,"written_bytes":131681277,"read_bytes":1000,"written_keys":613,"read_keys":0,"approximate_size":33810,"approximate_keys":122880}]}
compact pending bytes
备注:我们在8月29日,因为balance hot region 导致时延到10秒级别,关闭了 scheduler-hot-balance-region,后续集群恢复正常
目前我们正在执行top 1 region size 的region 的compact,从15:48分开始到目前17:06分还没有结束
./tikv-ctl --host xxxx.xxxx.xxx.xxx:25200 compact -d kv --region 8056302
这个命令执行是卡住的:
./tikv-ctl --host xxxx.xxxx.xxx.xxx:25200 region-properties -r 8056302
补充后续进展:
9月24号 18:20 进行同一个物理机上的所有tikv的重启
重新执行这条命令,能正常返回结果
./tikv-ctl --host xxxx.xxxx.xxx.xxx:25200 region-properties -r 8056302
重启后最大的region size的sst文件有变化,rocksdb的日志显示有在执行compaction的












