Michaela Buergle wrote:
Hi,
I'm currently struggling with the PerformanceEvaluation MR job given as
an example on http://wiki.apache.org/hadoop/Hbase/MapReduce. I'm getting
the problems already described a few months ago on hadoop-core-user
("HBase PerformanceEvaluation failing", 15. Nov 07: job exits with
java.io.Exception - maybe a memory issue in my case - afterwards hadoop
fsck says CORRUPT)
Check that the CORRUPT HDFS is not a misreporting. See comment at end
of https://issues.apache.org/jira/browse/HBASE-509 for how to check.
The first time this happened, I couldn't start hbase afterwards; there
seemed to be a problem with the -ROOT- table. In the end the only way I
could get things going again was deleting all of my data and starting
from scratch. Unfortunately I've chucked away the logfiles in order to
have a clean reproduction of the problem on my newly reformatted hadoop
system - as you might have guessed, while PerformanceEvaluation keeps
exiting, HBase hasn't quit service since.
Logs would help us figure whats going on. Please enable DEBUG. See
FAQ, http://wiki.apache.org/hadoop/Hbase/FAQ#5, for how.
And now there is a new phenomenon:
-> I did "drop table TestTable;" in the HQL-Shell (TestTable is the
table created by PerformanceEvaluation)
-> The shell said "1 table(s) dropped successfully"
-> But now I have 2 orphaned TestTable regions on one of my
RegionServers that won't go away. When I try to run
PerformanceEvaluation it gives me hbase.TableExistsException, when I try
to get rid of TestTable in the shell it tells me that the table doesn't
exist.
Any hints on how to deal with that?
Was your table large? Offlining does not currently seem to work
reliably (HBASE-478), a prerequisite for delete.
If you do not mind, please scratch-all, enable DEBUG and rerun. Send
over the logs if it starts acting up. What size is your cluster? How
many clients are participating in the MR job.
Thanks Michaela,
St.Ack
