On Mon, Jul 13, 2009 at 11:01 PM, Joel Nothman <jnoth...@student.usyd.edu.au > wrote:
> > On Tue, 14 Jul 2009 03:42:28 +1000, Erik Holstad <erikhols...@gmail.com> > wrote: > >> Try to have a look at question number 6 on : >> http://wiki.apache.org/hadoop/Hbase/FAQ >> might be the cause of your problems. >> > > Yes, it was the cause of my original problems, and I have applied its fix. > > It is not the cause of the log-replay not happening. > > > On Tue, 14 Jul 2009 04:00:07 +1000, stack <st...@duboce.net> wrote: > >> Sounds like you were running in standalone mode. In standalone mode, both >> master and regionserver run in same JVM. I'm not sure how log replay >> works >> in this case, if at all. Maybe post more of the log in pastebin? >> > > Thanks St.Ack, > > The rest of the log didn't seem to be enlightening. A log from the crash on > is attached. > > Are you on TRUNK? If you are up to a recent TRUNK, the logs should be >> replayed even on restart of the replay conductor, the hbase master >> (HBASE-698). Previous, if master was restarted, when a log replay was >> needed, logs would not be recovered. >> > > That's a little frustrating: I wanted to test the reliability on my local > machine before establishing a cluster. > > After getting trunk and starting the daemons, it doesn't seem to have > restored the data. > > When I start a shell and ask for a rowcount, I get: > NativeException: org.apache.hadoop.hbase.TableNotFoundException: .META. > from org/apache/hadoop/hbase/client/HConnectionManager.java:574:in > `locateRegionInMeta' > from org/apache/hadoop/hbase/client/HConnectionManager.java:524:in > `locateRegion' > from org/apache/hadoop/hbase/client/HConnectionManager.java:491:in > `locateRegion' > from org/apache/hadoop/hbase/client/HConnectionManager.java:565:in > `locateRegionInMeta' > from org/apache/hadoop/hbase/client/HConnectionManager.java:528:in > `locateRegion' > from org/apache/hadoop/hbase/client/HConnectionManager.java:491:in > `locateRegion' > from org/apache/hadoop/hbase/client/HTable.java:124:in `<init>' > ... > > Does this mean the data storage format has changed since 0.19.3, and it > didn't automatically upgrade the DB? Let alone replay its logs? > The data format from 0.19 to 0.20 is totally different, that was the biggest patch/issue that went in and has caused some instability but also boosts major speed improvements for gets and scans plus a different API. We are working on a migration tool to make the transition as smooth as possible. > > I'll probably landing up loading the data from scratch, but had hoped that > there would be a reliable crash recovery solution. > > - Joel > Sorry for the inconvinience. Regards Erik
