Hey Otis: Yeah, we're a bit crass when it comes to dealing with exceptions that come up out of HDFS. We'll just abort the server rather than try fancy footwork to get around the outage. HBASE-2183 is about doing a better job of riding over HDFS outage.
St.Ack On Sat, Dec 25, 2010 at 11:11 AM, Otis Gospodnetic <[email protected]> wrote: > Hello, > > Is this normal: > > 2010-12-25 18:59:48,689 ERROR org.apache.hadoop.hdfs.DFSClient: Exception > closing file /hbase/.logs/example.com,60020,1293204828665/10.208.42.97%3A > 60020.1293302073168 : java.io.IOException: All datanodes 127.0.0.1:50010 are > bad. Aborting... > java.io.IOException: All datanodes 127.0.0.1:50010 are bad. Aborting... > > I understand the dependency on DN(s), but why completely self-abort? > > In this particular case I restarted the DNs (really 1 of 1 of them total), > which > automatically "killed" the HBase RS, which I didn't expect. > > Why not just refuse any new Puts until DNs come back? > > Thanks, > Otis > ---- > Sematext :: http://sematext.com/ :: Solr - Lucene - Nutch - Hadoop - HBase > Hadoop ecosystem search :: http://search-hadoop.com/ > >
