Thanks for responding Stack.  In a 4 slave cluster (rf=3) cluster we
disabled an hdfs/hbase node which also seemed to make a second regionserver
crash.  I'm not sure exactly what went wrong.

There is data in the .META. directory

[r...@hadoop Master(hadoopmaster):/mnt/hadoop-0.20.2] bin/hadoop fs -lsr
/hbase/.META.
drwxr-xr-x   - root supergroup          0 2010-09-20 16:44
/hbase/.META./1028785192
-rw-r--r--   3 root supergroup        932 2010-09-20 14:25
/hbase/.META./1028785192/.regioninfo
drwxr-xr-x   - root supergroup          0 2010-09-21 15:41
/hbase/.META./1028785192/info
-rw-r--r--   3 root supergroup       5018 2010-09-21 15:41
/hbase/.META./1028785192/info/4694688683869779325
-rw-r--r--   3 root supergroup          0 2010-09-21 14:48
/hbase/.META./1028785192/info/6832908118284006198
-rw-r--r--   3 root supergroup     228130 2010-09-21 14:44
/hbase/.META./1028785192/info/7773537252698623585
drwxr-xr-x   - root supergroup          0 2010-09-21 14:44
/hbase/.META./compaction.dir


but it appears the -ROOT- region is empty.  If i scan '-ROOT-' in the hbase
shell i get:

hbase(main):001:0> scan '-ROOT-'
ROW                          COLUMN+CELL
0 row(s) in 0.0840 seconds


even though it appears -ROOT- exists in hdfs.  Any idea why ROOT would not
get loaded from hdfs after a master restart?

[r...@hadoop Master(hadoopmaster):/mnt/hadoop-0.20.2] bin/hadoop fs -lsr
/hbase/-ROOT-
drwxr-xr-x   - root supergroup          0 2010-09-21 14:50
/hbase/-ROOT-/70236052
-rw-r--r--   3 root supergroup        618 2010-09-21 14:50
/hbase/-ROOT-/70236052/.regioninfo
drwxr-xr-x   - root supergroup          0 2010-09-21 15:41
/hbase/-ROOT-/70236052/info
-rw-r--r--   3 root supergroup        489 2010-09-21 15:41
/hbase/-ROOT-/70236052/info/308400263036654323
-rw-r--r--   3 root supergroup        462 2010-09-21 13:43
/hbase/-ROOT-/70236052/info/8250773031047005400


I'll just follow your suggestion of generating a new .META. and copying
over.  Thanks,

Matt


On Tue, Sep 21, 2010 at 5:20 PM, Stack <[email protected]> wrote:

> On Tue, Sep 21, 2010 at 1:07 PM, Matt Corgan <[email protected]> wrote:
> > Is there a way to create a new .META. table if I lost it?  For example,
> I'm
> > getting this exception everywhere:
> >
> > org.apache.hadoop.hbase.TableNotFoundException: .META. (NativeException)
> >
>
> How'd you do that?  Thats pretty impressive.
>
> >
> > The add_table script doesn't appear to handle .META., giving this error:
> >
> > java.lang.IllegalArgumentException: Illegal first character <46> at 0.
> > User-space table names can only start with 'word characters': i.e.
> > [a-zA-Z_0-9]: .META.
> >
> >
> > And the hbase shell will only create user-space table names.  Restarting
> the
> > hbase master also does not appear to re-create it.
> >
> > I'm using 0.20.6.
> >
> Whats the master log look like when you start up hbase?  Whats it say
> about missing .META.?
>
> If you do a
>
> $ bin/hadoop fs -lsr /hbase/.META.
>
> ... is there anything in your filesystem?
>
> If nothing there, one thing you could do is run an hbase elsewhere.
> Let it create a new "/hbase" dir and then copy the .META. table from
> here up to the filesystem with data (Run add_table.rb after .META. is
> in place to repopulate its rows).
>
> St.Ack
>

Reply via email to