Basically, every time our DB instance has a CPU spike is read, and very few write-induced CPUs are high. This means that read, random read, sort, will consume CPU. And the write is mainly IO behavior, especially sequential write, do not need to account for CPU.
This time, RDS is very high within three hours, always maintaining 50+, up to 98. Of course, our business availability is not dependent on RDS.
Observe for some time, and there is no obvious big slow query. But there will be some small queries sometimes have no, in fact, these can be seen by the naked eye may be the problem .
The second field, which is eventually positioned as a composite primary key, is queried too much, resulting in a full table scan. Fix after you add the index individually.
Not too slow, the amount of query data is not too much, but very often, the query volume is very large, often this kind of small SQL will be very covert to the CPU up.
The processing of database CPU high is recorded once