Keith, This is cause your dfs.data.dir and dfs.name.dir are, by default, on /tmp. When your /tmp is cleared by the OS (a regular thing people forget to think of), your HDFS is essentially wiped away.
Configure dfs.name.dir and dfs.data.dir to be on a proper directory that isn't cleaned up periodically and/or at boot, and you'll have a proper HDFS across 'sessions'. On Mon, Nov 7, 2011 at 2:45 AM, Keith Thompson <[email protected]> wrote: > Hi, > > I am running Hadoop in pseudo-distributed mode on Linux. For some reason, > I have to reformat the namenode every time I start up Hadoop because it > will fail whenever I try to connect to the HDFS. After I reformat, it runs > fine for that session; however, if I try to run it again later it will have > the same issue. There is probably some setting I forgot to set somewhere. > Can anyone help? > > -- > *Keith Thompson* > Graduate Research Associate > SUNY Research Foundation > Dept. of Systems Science and Industrial Engineering > Binghamton University > -- Harsh J
