[
https://issues.apache.org/jira/browse/HBASE-19757?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16325436#comment-16325436
]
Ted Yu edited comment on HBASE-19757 at 1/15/18 6:16 PM:
---------------------------------------------------------
In master, we have the following code in RSGroupInfoManagerImpl#refresh()
{code:java}
if(!masterServices.isInitialized()) {
specialTables = Arrays.asList(AccessControlLists.ACL_TABLE_NAME,
TableName.META_TABLE_NAME,
TableName.NAMESPACE_TABLE_NAME, RSGROUP_TABLE_NAME);
} else {
specialTables =
masterServices.listTableNamesByNamespace(NamespaceDescriptor.SYSTEM_NAMESPACE_NAME_STR);
}
{code}
If acl table is about to be created, the call in else branch may end up not
having hbase:acl as one of the special tables.
In RSGroupBasedLoadBalancer, due to lack of rs group, no server is provided for
hbase:acl table, leading to the deadlock.
was (Author: yuzhih...@gmail.com):
In master, we have the following code in RSGroupInfoManagerImpl#refresh()
{code}
if(!masterServices.isInitialized()) {
specialTables = Arrays.asList(AccessControlLists.ACL_TABLE_NAME,
TableName.META_TABLE_NAME,
TableName.NAMESPACE_TABLE_NAME, RSGROUP_TABLE_NAME);
} else {
specialTables =
masterServices.listTableNamesByNamespace(NamespaceDescriptor.SYSTEM_NAMESPACE_NAME_STR);
}
{code}
If acl table is about to be created, the call in else branch may end up not
having hbase:acl as one of the special tables.
By always using the assignment in if block, TestRSGroupsWithACL passes.
> System table gets stuck after enabling region server group feature in secure
> cluster
> ------------------------------------------------------------------------------------
>
> Key: HBASE-19757
> URL: https://issues.apache.org/jira/browse/HBASE-19757
> Project: HBase
> Issue Type: Bug
> Reporter: Ted Yu
> Assignee: Ted Yu
> Priority: Major
> Attachments: 19757.v1.txt, 19757.v2.txt, 19757.v3.txt
>
>
> I was testing on an hbase-2 secure cluster against hadoop 3 where some tables
> were created without region server group feature.
> After adding the RSGroupAdminEndpoint and RSGroupBasedLoadBalancer to
> hbase-site, I restarted the whole cluster.
> After the restart, hbase:meta region got stuck in transition (forever).
> {code}
> 2018-01-10 21:20:16,696 INFO
> [org.apache.hadoop.hbase.rsgroup.RSGroupInfoManagerImpl$RSGroupStartupWorker-ctr-e137-1514896590304-8706-01-000002.hwx.site,20000,1515619212617]
> zookeeper.MetaTableLocator: Failed verification of hbase:meta,,1 at
> address=ctr-e137-1514896590304-8706-01-000004.hwx.site,16020,1515618538016,
> exception=org.apache.hadoop. hbase.NotServingRegionException:
> hbase:meta,,1 is not online on
> ctr-e137-1514896590304-8706-01-000004.hwx.site,16020,1515619181453
> at
> org.apache.hadoop.hbase.regionserver.HRegionServer.getRegionByEncodedName(HRegionServer.java:3314)
> at
> org.apache.hadoop.hbase.regionserver.HRegionServer.getRegion(HRegionServer.java:3291)
> at
> org.apache.hadoop.hbase.regionserver.RSRpcServices.getRegion(RSRpcServices.java:1355)
> at
> org.apache.hadoop.hbase.regionserver.RSRpcServices.getRegionInfo(RSRpcServices.java:1667)
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
