Michael, But can you claim this is "yet another" HBase failure? Or are these DFS problems related to running with too small a cluster? It's been some time so my recollection is hazy, but didn't you mention you have a cluster of 4 nodes only?
I found that most of my DFS issues were caused by attempting to host too much load on too few physical resources, and that adding nodes to distribute the load solved my problems. Running with dfs.datanode.max.xcievers=2048 helped for a while, but that was an indication that per DataNode load was too high. Best regards, - Andy > From: Michael Dagaev <[email protected]> > Subject: Re: Yet another Hbase failure > To: [email protected] > Date: Friday, March 20, 2009, 7:04 AM > Hi, stack > > See the hadoop-site.xml in the attachment. > dfs.datanode.socket.write.timeout = 0, > dfs.datanode.max.xcievers=1023 > > The hbase-site.xml is not interesting. > It contains only "hbase.rootdir" and > "hbase.master" > > I checked the logs (I did not know hbase logged ulimit). > On all region server hosts ulimit is 32768. On the master > host ulimit is 1024. > > Thank you for your cooperation, > M.
