Thanks to Gopal. But why disable mapjoin has better performance when we don't use cast to string(user always lazy)?
Join key values comparison in in reduce stage is more quickly? Zhiwen Sun p;9456 On Wed, Dec 23, 2015 at 2:36 AM, Gopal Vijayaraghavan <gop...@apache.org> wrote: > > > We found that when we join on two different type keys , hive will > >convert all join key to Double. > > This is because of type coercions for BaseCompare, so that String:Integer > comparisons with "<=" will work similarly to "=". > > > b.id to double. When the conversion occurs, map join will become very > >slow. > ... > > Does anyone how to solve it more effectively? > > This is an issue that only affects mapreduce mode in Hive. The broadcast > joins in Tez switched to Murmur hash to avoid this issue (HIVE-6924 + > HIVE-7121). > > As a workaround, you can insert explicit casts to String to make this > faster. > > Cheers, > Gopal > > >
