rate per second=rps-为google云数据库标志考虑的建议

log_error=/mysql/logs/error.log  # from default of stderr will allow you to view your entries and take corrective action
max_connections=512  # from 4000 to conserve RAM and your max_used_connections was 110 in 6 days
thread_cache_size=100  # from 48 to reduce threads_created of 204 (for CPU busy reduction)
innodb_flush_neighbors=0  # from 2  you have SSD and should be using 0 per refman
innodb_lru_scan_depth=100  # from 2048 to reduce CPU busy every second
innodb_page_cleaners=4  # from 1 to expedite cleaning and reduce innodb_buffer_pool_pages_dirty of 791
open_files_limit=10000  # from 1,048,576 will conserve RAM and be adequate
query_cache_size=0  # from 1M to conserve RAM - you are NOT using QC
query_cache_limit=0  # from 1M to conserve RAM - you are NOT using QC

需要进行研究来探索为什么 a)6天内56133个中止的客户 b)com_stmt_prepare比com_stmt_close大213倍,意味着资源在完成后的6天内未释放213次 c)是否有必要在标志中显示innodb_log_compressed_页面? 这会导致每1/2小时记录1GB日志的一部分,这些日志文件在系统上每14分钟旋转一次。

使用名为findfragtables.sql的实用程序脚本将帮助您确认表已被压缩,因此使用CPU对每个活动进行解压缩/压缩。在innodb_compression_level=6时,您在压缩上花费了大量的cpu时间。 删除,插入,更新,以6538平均每秒6天似乎是非常繁忙。

你可以在我的个人资料中找到我的联系信息,网络个人资料。