To make sure your dfs.namenode.name.dir is by default. then, how did you find /user exists? hdfs dfs -ls ? or you checked dfs.datanode.data.dir? if the latter, then don't worry.
On Wed, Oct 8, 2014 at 11:56 AM, Tianyin Xu <t...@cs.ucsd.edu> wrote: > Hi, > > I wanna run some experiments on Hadoop which requires a clean, initial > system state of HDFS for every job execution, i.e., the HDFS should be > formatted and contain nothing. > > I keep *dfs.datanode.data.dir* and *dfs.namenode.name.dir* the default, > which are located in /tmp > > Every time before running a job, > > 1. I first delete dfs.datanode.data.dir and dfs.namenode.name.dir > #rm -Rf /tmp/hadoop-tianyin* > > 2. Then I format the nameNode > #bin/hdfs namenode -format > > 3. Start HDFS > sbin/start-dfs.sh > > 4. However, I still find the previous metadata (e.g., the directory I > previously created) in HDFS, for example, > #bin/hdfs dfs -mkdir /user > mkdir: `/user': File exists > > Could anyone tell me what I missed or misunderstood? Why I can still see > the old data after both physically delete the directories and reformat the > HDFS nameNode? > > Thanks a lot for your help! > Tianyin >
