I remember trying Dynamodb around 2015/2016: You had to specify your expected read and write throughout and you would be billed for that. At that time we had a pretty spikey traffic use case which made using dynamodb efficiently impossible
I had a similar experience, but ultimately wrote a service to monitor our workloads and request increased provisioning during spikes. You could reduce your provisioning like 10 times a day, but after that you could only increase it and would be stuck with the higher rate for a time.
And then on-demand provisioning was released and it was cheap enough to be worth simplifying our workflows.
