使用key查询不稳定

[wroldchinalee]

使用三台虚拟机，配置8核64G内存，大概70亿数据，表模型为unique，按照key查询时间不稳定，请问有什么优化措施吗

[morningman]

The unqiue key model use merge-on-read logic to read the data.
Which means the latency of query depends on the compaction progress of data file.

Are there actual metrics showing how unstable it is?