We have an r6g.large write instance in our cluster which is perfectly enough for us during off-peak hours which is about 16 hours a day on workdays.
But during the 8 peak hours we would need a larger instance let's say an r6g.2xlarge minimum. Between the peak and off-peak hours it's not allowed to have any downtime!
Our solution would be to have the r6g.large write instance and use the r6g.2xlarge reader instance during the peak time. But in the off-peak hours the reader should stop completely.
But to our knowledge this on demand 6g.2xlarge reader instance can't be stopped this is a limitation of the stop function (right?).
So what's the best practice to do this? Delete the reader instance every day and create a new one? Is it a good way to do it? Seems a bit odd because we lose all the metrics and reports but running it 0-24 costs a lot.
(We tried serverless V2 but it's buggy and expensive, it has random spikes and doesn't scale down to 0.5-1 ACU we tried everything.)
AWS OFFICIALUpdated 2 years ago
Yes, we tried this solution before but our problem is that we would only need the extra capacity during the peak hours which is predictable (from 7am to 5pm).
During the off-peak hours we run heavy background tasks on the writer instance but these are not time sensitive or important so for us it's ok if it runs for 8-10 hours at maximum capacity. Starting extra read replicas would make the usage costs much higher.
(If I set the min to 0 then I will have the writer instance and zero read instance, right?)