There is a data race in s2's incrementQueryCount() method, which makes it possible for the query_count_ variable to be updated at the same time by multiple threads. To fix this, make query_count_ an atomic (should be faster than a mutex)