[HBASE-18625] Splitting of region with replica, doesn't update region list in serverHolding. A server crash leads to overlap.

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.2.6
• Fix Version/s: 1.3.2, 1.4.1, 1.2.7
• Component/s: read replicas
• Assignee: Hua Xiang
• Reporter: Igloo
• Created: 18/Aug/17 02:44
• Updated: 01/Feb/19 19:55
• Resolved: 03/Jan/18 20:36

Description

The situation can appear in following steps in release hbase1.2.6

1. create 'testtable', 'info',

{REGION_REPLICATION=>2}

2. write somerecords into 'testtable'
3. split the table 'testtable'
4. after the spliting, the serverHoldings in RegionStates still holds the regioninfo for the replica of parent region
5. restart the regionserver where the parent replica-region located
6. the offlined replica of parent region will be assigned in ServerCrashProcedure.

hbase hbck 'testtable‘

ERROR: Region

{ meta => null, hdfs => null, deployed => qabb-qa-hdp-hbase1,16020,1503022958093;testtable,,1503022907686_0001.42d11cfe195b3cc4d08b2c078a687f6d ., replicaId => 1 }

not in META, but deployed on qabb-qa-hdp-hbase1,16020,1503022958093
18 ERROR: No regioninfo in Meta or HDFS.

{ meta => null, hdfs => null, deployed => qabb-qa-hdp-hbase1,16020,1503022958093;testtable,,1503022907686_0001.42d11cfe 195b3cc4d08b2c078a687f6d., replicaId => 1 }

Issue Links

• causes: HBASE-20016 TestCatalogJanitorInMemoryStates#testInMemoryForReplicaParentCleanup is flaky
• Is contained by: HBASE-18223 Track the effort to improve/bug fix read replica feature