Without HDFS-265 there is the likelyhood of data loss in an extreme crash scenario (eg: kill -9, power failure, etc). HDFS-0.21 fixes these things, but its not out yet. In the mean time, try a regular kill if you want to take a regionserver down w/o data loss.
-ryan On Thu, Nov 26, 2009 at 3:13 AM, Rekha Joshi <rekha...@yahoo-inc.com> wrote: > Instinctively think second to be the cause. > Did you try flushing it to disk and then killing and found strange behavior? > - just one more experiment to remove the role of instinct :-) Thanks! > > On 11/26/09 4:23 PM, "Tux Racer" <tuxrace...@gmail.com> wrote: > > Hello Hbase Users! > > I installed the Hbase distribution on a one node cluster (my test linux box) > I wrote a small client that creates a table, and inset some data in it. > > Then if I do a > > pkill -9 java > > and restart Hbase, everything is gone (no table and thus no data in that > table any more). > > I would have expected that hbase recovers well from such a 'kill' (which > could simulate a node crash). > Am I missing something? > Is that because I only use one node? > is that because the data is in memory only and needs to be flushed to disk? > > Thanks > TR > > > >
