[HBASE-28481] Prompting table already exists after failing to create table with many region replications

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.4.13
• Fix Version/s: 2.6.0, 2.4.18, 3.0.0-beta-2, 2.5.9
• Labels: pull-request-available

Environment

Centos

• Hadoop Flags: Reviewed
• Assignee: guluo
• Reporter: guluo
• Created: 01/Apr/24 15:38
• Updated: 08/Apr/24 21:32
• Resolved: 07/Apr/24 10:37

Description

Reproduction steps:

# Create table with 65537 region replications 
# we would get errors as follow,  this step is no problem 
hbase:005:0> create 't01', 'info', {REGION_REPLICATION => 65537} 
ERROR: java.lang.IllegalArgumentException: ReplicaId cannot be greater than65535 
For usage try 'help "create"' 
Took 0.7590 seconds

# list, and found the table does not exist, as follow 
hbase:006:0> list TABLE 
0 row(s) Took 0.0100 seconds 
=> []

# we create this tale agin by the correct way 
# we would get message that this table already exists 
hbase:007:0> create 't01', 'info' 
ERROR: Table already exists: t01! 
For usage try 'help "create"' 
Took 0.1210 seconds 

 
Reason:
In the CreateTableProcedure, we update this table descriptor into HBase cluster at stage  CREATE_TABLE_WRITE_FS_LAYOUT
 

env.getMasterServices().getTableDescriptors().update(tableDescriptor, true); 

 
and then, we check if the Region Replication Count is legal at stage CREATE_TABLE_ADD_TO_META.
 
 

newRegions = addTableToMeta(env, tableDescriptor, newRegions);

// MutableRegionInfo.checkReplicaId 
private static int checkReplicaId(int regionId) {     
  if (regionId > MAX_REPLICA_ID) {         
    throw new IllegalArgumentException("ReplicaId cannot be greater than" + 
     MAX_REPLICA_ID);    
   }     
return regionId;
}

 
 
So, we can not create the same name table by correct way after faling to create table with many region replications (exceed 65536), because the table descriptor has been updated into cluster and there is no rollback.

So i think we can check if the region replication count at stage CREATE_TABLE_PRE_OPERATION to avoid this problem

Issue Links

• links to: GitHub Pull Request #5789