Unwinding Uber’s Most Efficient Service
Buck Heroux

Uber talks about having to dynamically poll various data stores to keep main memory up to date, with the more complex search tree methods wouldn’t you experience downtime as the trees are getting up dated?

So if your main criteria is to as in sync as possible with the other data stores wouldn’t that time spent recreating those trees represent time that the machine is unable to process queries? If the time spent recreating the trees is greater than the time in between changes to the data store then the system would never actually catch up and be in sync