Inserting a batch of measurements that compress well that should fit into one bucket may lead to creation of multiple buckets because the size estimation cannot make too many assumptions on compressibility of data. This is a problem during bulk load, and not during steady state since in steady state a partially full bucket is compressed, and adding one uncompressed measurement leads to less error in the size estimation.

Consider a server parameter that allows the customer to influence the size estimation in relation to timeseriesBucketMaxSize.