I know how annoying that can be. Our focus for 0.7.1 is to remove this timing bug & provide the ability to restart a specific role as an workaround.
-- Andrei Savu On Mon, Jan 30, 2012 at 5:29 PM, Tim Robertson <[email protected]>wrote: > Indeed - seems like intermittent timing bug as on a few relaunches as > I can't reproduce now, and have terminated the cluster which failed so > logs gone. If it resurfaces I'll capture master logs and create > issue. > > Thanks, > Tim > > > > On Mon, Jan 30, 2012 at 4:08 PM, Andrei Savu <[email protected]> > wrote: > > I think you are affected by timing issues: I suspect the HBase master is > > trying to start before HDFS is available. > > > > On Mon, Jan 30, 2012 at 3:09 PM, Tim Robertson < > [email protected]> > > wrote: > >> > >> What is the correct way to start the master please? Should I ssh and > >> use a particular user to start-hbase.sh? It doesn't work as my user, > >> with various permission issues but I suspect this not to be the > >> correct way anyway. > >> [NameNode, JobTracker, Zookeeper all start, but HBase master does not] > > > > > > Check configure_hbase.sh: > > > https://github.com/apache/whirr/blob/trunk/services/hbase/src/main/resources/functions/configure_hbase.sh > > > > > > > > > > function start_hbase_daemon() { > > > > > > > > if which dpkg &> /dev/null; then > > > > > > > > AS_HADOOP="su -s /bin/bash - hadoop -c" > > > > > > > > elif which rpm &> /dev/null; then > > > > > > > > AS_HADOOP="/sbin/runuser -s /bin/bash - hadoop -c" > > > > > > > > fi > > > > $AS_HADOOP "$HBASE_HOME/bin/hbase-daemon.sh start $1" > > > > > > > > } > > > > > > > > > > We are working on splitting that into multiple files: > > > > > > > > https://issues.apache.org/jira/browse/WHIRR-295 > > > > > > I hope this is useful for you. > > > > > > Anything relevant in the HBase Master log file? Something like: "Unable > to > > connect to Namenode / HDFS"? > > > > > > > > >
