Description
When using the rebalance
command, the tolerance
flag specifies a distance (%) from the mean storage utilization that is used as the target for both source and destination brokers in relocations.
For example, if the mean storage utilization were 2,000GB and we are using a tolerance of 10%, we want to schedule relocations so that the most loaded brokers are brought down to a range of 1,800-2,200GB in storage utilization, but not dropping below 1,800GB. Similarly, when relocating partitions to destinations, we want to bring them up into the range of 1,800-2,200GB, not exceeding 2,200GB.
After extensive use, I've found that the default tolerance of 10% is a good baseline, but I frequently find myself playing with values that ultimately result in the most optimal map: that is, the output with the lowest storage utilization range (the delta between the most and least utilized broker in terms fo storage utilization). Topicmappr should have an option to iteratively try tolerance values and automatically select the one that results in the most optimal map.