By default we had the default scanner caching set to 30 in 0.19, but it was set back to 1 in 0.20. See https://issues.apache.org/jira/browse/HBASE-1331
Now, in the shell it is set to 10 during the count. You can manually change it or, better, simply run the RowCounter MR job. J-D 2010/4/25 WANG Shicai <evan...@yahoo.cn>: > Hi, > > There was no such problem in HBase-0.19. It began in HBase-0.20. > > 2010-04-26 > WANG Shicai > > The shell uses a scanner prefetching of 10 rows in the shell, maybe > you set it higher in your own code? > But 3000 secs does seem unusually high. > J-D > 2010/4/23 WANG Shicai <evan...@yahoo.cn>: >> Hi, >> >> Count operation in shell over a 1,000,000 records table wasted 3000 seconds; >> while a scan operation using java api cost only 32 seconds. Why? >> >> WANG Shicai , >> >> >> > > __________________________________________________ > 赶快注册雅虎超大容量免费邮箱? > http://cn.mail.yahoo.com > >
