Prashant, You may be able to salvage some free temp-data space (if used up by earlier maps or reduces) from shared mapred.local.dir directories by restarting the TaskTracker. Also clear out unwanted items from /tmp (which is what may be blocking your JVMs). Or if you have another free disk, set HADOOP_OPTS in hadoop-env.sh as "-Djava.io.tmpdir=/path/to/dir/on/free/disk" and make sure the chosen path has permissions 777, and then once HDFS is up delete away some unwanted files.
On Tue, May 22, 2012 at 6:13 PM, shan s <mysub...@gmail.com> wrote: > My cluster went out of disk during a job execution and now shows occupied > disk space = 100%. When I attempt hadoop fs -ls input, I get below. > > Java HotSpot(TM) 64-Bit Server VM warning: Insufficient space for shared > memory file: /tmp/hsperfdata_user/32062 > Try using the -Djava.io.tmpdir= option to select an alternate temp location. > Error: tmp.h sperfdata Could not find or load main class ___._user.32074 > Is there an easy way to come out of this and be able to delete hdfs files, > to get it back to working. > Thanks, Prashant. -- Harsh J
