Try upping your xcievers to 2047 or thereabouts. I had to do that with a cluster of your size.
Was there any errors on the datanode side you could find? -ryan On Tue, Feb 3, 2009 at 1:58 AM, Michael Dagaev <[email protected]>wrote: > Hi, all > > We ran an HBase cluster of (1 master/name node + 3 region > server/data nodes). > We upped the number of open files per process, increased the heap size > of the region > servers and data nodes to 2G, and set dfs.datanode.socket.write.timeout=0, > and > dfs.datanode.max.xcievers=1023 > > The cluster seems to run ok but the Hbase logged exceptions at INFO/DEBUG > level. > For instance > > org.apache.hadoop.dfs.DFSClient: Could not obtain block <block name> > from any node: java.io.IOException: No live nodes contain current block > > org.apache.hadoop.dfs.DFSClient: Failed to connect to <host name>:50010: > java.io.IOException: Got error in response to OP_READ_BLOCK for > file <filer name> > > Does anybody know what these exceptions mean and how to fix them? > > Thank you for your cooperation, > M. >
