Some requests are MinimizeLatency, some are FullConsistency. I should add some metrics to better understand the distribution of consistency levels at the time of the spikes. We are self-hosted in AWS and we are using Aurora postgres.