Have you taken jstack for the slow scans ? If so, can you pastebin the stack trace ?
1.0.0 is quite old. Any chance of upgrading to 1.2 release ? Cheers > On Oct 10, 2016, at 2:04 AM, 陆巍 <luwei...@163.com> wrote: > > Hi All, > > I met with a problem where the scan perfoamance decreases over time. > Hbase connections are kept in a data access service (in tomcat), and there > are table scan operations. The scan performance for each scan batch(~10 > parallel scan) increases as below: > day avg. cost(ms) > 1 56.213115 > 2 43.697054 > 3 36.925063 > 4 50.683257 > 5 62.749022 > 6 84.943314 > 7 92.237783 > 8 94.452549 > 9 103.853937 > 10 114.725657 > 11 128.601287 > > The time cost for each remote scan batch is now over 100ms. > In order to make sure Hbase cluster is fine, I started another same data > access service, but find the time cost is around 30ms. So, I think hbase > cluster is fine and the problem is in the data access service, which is the > client of hbase. > > I do believe there is some resource missed to release, but really have no > idea where it is. > I am using hbase 1.0.0-chd5.5.1. > > Here is the code: > // the connection is created as a static instance: > connection=ConnectionFactory.createConnection(getConfiguration()) > // scan logic for each remote call > try { > Table table = connection.getTable(tableName); > rs = table.getScanner(scan); > } finally { > rs.close(); > table.close(); > } > > > > Thanks, > Wei