Hi St.Ack,
removing the zookeeper data dir did the trick. The master stopped again
after complaining that it couldn't resolve the old node name (which I
was only able to grep under the dfs data - any idea why the master host
name is persisted there and read again?) but defining that as a
localhost alias in /etc/hosts also fixed that and as far as I can tell
all data is accessible again.
Thanks a lot!!!
Will put that into our troubleshooting guide right away.
Is there further implications of deleting the zookeeper data dir? And
once we are at it, is there some general information on what to do if
you need to rename nodes (in particular master nodes)?
Thanks again!
Henning
On 01/17/2012 09:02 PM, Stack wrote:
On Tue, Jan 17, 2012 at 7:19 AM, Henning Blohm<[email protected]> wrote:
Hi,
After an upgrade of hadoop and hbase (to 0.90.4-cdh3u2) from 0.90 hbase and
0.20-append hadoop on a single node test installation everything worked fine
initially.
Would suggest you go to 0.90.5 while you are going about upgrades Henning.
Then there was some DNS changes and host name changes which resulted in a
lot "hostname<oldname> cannot be resolved" problems in the logs and the
master web interface would only show a stack trace from a bad lookup
("hostname can't be null").
Did you scrub the zk data for this cluster? It was picking up old
root location from zk? That might have helped.
Master Startup, Assigning ROOT region, 10050s
i.e. it never continues.
Paste a bit of log Henning up in pastebin.
St.Ack
