Ted: I probably pointed you at this before, http://hbase.apache.org/apidocs/org/apache/hadoop/hbase/client/HConnectionManager.html? Its how our Connection stuff works. Each new HConnection will open a new connection to zk ensemble. If clients share the Configuration instance, then they'll all use same zk ensemble client.
Regards HBase, each RegionServer and each Master should be hosting one client of zk. Each of your HBase clients will have a zk client embedded. If HTablePool or multithreaded client and all HTable instances used in this clien all share the same Configuration, then this is still one zk client only. If you create a new Configuration instance for each new HTable instance, then you will be creating a new zk client. Any of the above zk clients embedded in hbase should be closed on exit. Does that help? St.Ack On Wed, Jan 26, 2011 at 10:44 AM, Ted Dunning <[email protected]> wrote: > There is a close method on the zk client data structure. > > Sounds like that isn't getting called > > On Wed, Jan 26, 2011 at 10:19 AM, Ted Yu <[email protected]> wrote: > >> In 0.90, I am looking for a programmatic way of releasing connections to >> zookeeper. >> When I run flow on 7-node cluster, there were 460 connections listed under >> Clients of zk.jsp >> >> Thanks >> >
