I have a Hbase table on my cluster (about 5M rows). I wrote a Mapreduce job to delete a mass rows of my table. After delete, there 's only about 30K rows in my table. But operations to my table become very slowly, for example: count, scan, disable, ... I looked at Cloudera Manager Dashboard and see the region size of my table still ~6G. And I think it was my problem. Anyone have any idea? Please help me!
The disk space won't be reclaimed from region files until there's been a major compaction. You can manually trigger that from the hbase web interface. This is a heavy operation if you do it on all regions at once, so be aware it will impact performance while it's running.