[
https://issues.apache.org/jira/browse/HBASE-12570?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14223968#comment-14223968
]
Enis Soztutar commented on HBASE-12570:
---------------------------------------
This is from HMaster.sanityCheckTableDescriptor() method introduced in
HBASE-10591
{code}
// check split policy class can be loaded
try {
RegionSplitPolicy.getSplitPolicyClass(htd, conf);
} catch (Exception ex) {
throw new DoNotRetryIOException(ex);
}
{code}
So it should have failed the alter table or create table. Which version of
HBase was the user running? Maybe it was disabled via configuration.
However, we load other classes (not just split policy), and they will have the
same affect as well. For example we have seen a similar problem for
co-processors ([~ndimiduk] was involved in a case where a typo in co-processor
name caused the cluster to go down). Similarly storage engine classes, wal
readers etc.
> Missing/invalid split policy class name brings down your HBase cluster
> ----------------------------------------------------------------------
>
> Key: HBASE-12570
> URL: https://issues.apache.org/jira/browse/HBASE-12570
> Project: HBase
> Issue Type: Bug
> Reporter: James Taylor
>
> See PHOENIX-1473. If a split policy class cannot be resolved, then your HBase
> cluster will be brought down as each region server that successively attempts
> to open the region will not find the class and will bring itself down.
> One idea to prevent this would be to fail the CREATE TABLE or ALTER TABLE
> admin call if the split policy class cannot be found.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
