On Sun, May 15, 2011 at 10:03 PM, Lars George <[email protected]> wrote: > 2011-05-16 06:59:05,389 DEBUG > org.apache.hadoop.hbase.master.SplitLogManager: task not yet acquired > /hbase/splitlog/RESCAN0000148501 ver = 0
You are running on tip of the append branch or on cdh3 release? If not, you could see the above because you are missing the recover lease short-circuit that was added to branch-0.20-append. Otherwise it could be a peculiarity of pseudo-distributed mode w/ replica 1 (I presume). > This keeps on going up and up. What is the right way to recover from > this? Delete something from ZK? Delete something from HDFS? What shell > commands would help? > For sure all processes are down. Oh, in pseudo-distributed mode, the master and regionserver are probably sharing the same hdfs user identity; so the user holding the lease still 'exists' though the regionserver 'crashed'. St.Ack
