cluster may not recover if RS hosting ROOT and META together crashes
--------------------------------------------------------------------
Key: HBASE-2766
URL: https://issues.apache.org/jira/browse/HBASE-2766
Project: HBase
Issue Type: Bug
Reporter: Eugene Koontz
Assignee: Eugene Koontz
If both -ROOT- and .META. are located on the same regionserver, and that
regionserver crashes, the master is unable to reassign these tables to a new
regionserver.
Andrew writes:
"I ran a webtable scenario up on EC2 using the latest TM-2 and a cluster with 1
ZK, 1 master, 5 slaves, and 1 auxiliary node, using the HBase cluster scripts
at https://tm-files.s3.amazonaws.com/hbase-ec2.tar.bz2. On the aux node I
uploaded the Faulkner utility --
https://tm-files.s3.amazonaws.com/faulkner.tar.gz -- and ran the 'webtable.sh'
script in that tarball. On the master I waited about 30 minutes for a fair
amount of regions to proliferate and then ran:
# nice -10 hbase shell
hbase> count 'TestTable'
and walked away, leaving it to chew on the heavy write load and scan.
At some point during this test scenario a region server crashed, due to a JVM
segfault. The client (Faulkner) never recovered.
As far as I can see, in this test scenario the master never reassigns regions
away from a crashed RS."
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
