I have a large phoenix table with ~1800 regions. After running a compaction, I noticed degraded query performance. After purging the system.stat table, query performance returned back to normal.

Surprisingly, using sqlline.py does not cause this slow down. Here are some numbers,

(before and after purging system.stat):

sqlline query took: 1 row selected (2.869 seconds)

after compaction, before purging system.stat table:

              mean = 47536.55 milliseconds
            stddev = 2657.65 milliseconds

after purging system.stat table:

              mean = 57.63 milliseconds
            stddev = 39.82 milliseconds

Purging the system.stat table removed about ~1755 records.

I have the following related configuration in my hbase-site.xml:


Any thoughts?