I am trying to understand the downsides of having a large number of hfiles by having a large hbase.hstore.compactionThreshold
This delays major compaction. However, the amount of data that needs to be read and re-written as a single hfile during major compaction will remain the same unless we have large number of deletes or expired rows
I understand the random reads will be affected since each hfile may be a candidate for the row, but is there any other downside I am missing?
Thanh Do 2013-06-12, 16:34