Charles Connell created HBASE-29363:
---------------------------------------
Summary: CompactSplit should not attempt to split non-default
regions
Key: HBASE-29363
URL: https://issues.apache.org/jira/browse/HBASE-29363
Project: HBase
Issue Type: Bug
Reporter: Charles Connell
Assignee: Charles Connell
I discovered a problem while running a large cluster with region replicas
enabled. From time to time, WAL flushes trigger splits, and when the region
being split is a secondary replica, this split causes problems. The
{{SplitTableRegionProcedure}} does not permit splitting secondary replicas. The
RegionServer will send a {{ReportRegionStateTransitionRequest}} to the HMaster,
which will send back an error response, logging as it does so:
{code:java}
2025-05-29T16:40:27,585
[RpcServer.priority.RWQ.Fifo.write.handler=12,queue=0,port=60000] ERROR
org.apache.hadoop.hbase.ipc.RpcServer: Unexpected throwable object
java.lang.IllegalArgumentException: Can't invoke split on non-default regions
directly at
org.apache.hadoop.hbase.master.assignment.SplitTableRegionProcedure.checkSplittable(SplitTableRegionProcedure.java:202)
~[hbase-server-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
org.apache.hadoop.hbase.master.assignment.SplitTableRegionProcedure.<init>(SplitTableRegionProcedure.java:137)
~[hbase-server-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
org.apache.hadoop.hbase.master.assignment.AssignmentManager.createSplitProcedure(AssignmentManager.java:1159)
~[hbase-server-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
org.apache.hadoop.hbase.master.assignment.AssignmentManager.updateRegionSplitTransition(AssignmentManager.java:1359)
~[hbase-server-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
org.apache.hadoop.hbase.master.assignment.AssignmentManager.reportRegionStateTransition(AssignmentManager.java:1207)
~[hbase-server-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
org.apache.hadoop.hbase.master.assignment.AssignmentManager.reportRegionStateTransition(AssignmentManager.java:1247)
~[hbase-server-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
org.apache.hadoop.hbase.master.MasterRpcServices.reportRegionStateTransition(MasterRpcServices.java:1807)
~[hbase-server-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos$RegionServerStatusService$2.callBlockingMethod(RegionServerStatusProtos.java:16722)
~[hbase-protocol-shaded-2.6-hubspot-SNAPSHOT.jar:2.6-hubspot-SNAPSHOT] at
...{code}
As a result, the RegionServer will retry the
{{ReportRegionStateTransitionRequest}} infinitely. The RegionServer is doing
this in a thread pool task that nothing is waiting on, so it's not really
bothered by this. In a large cluster, the HMaster request handling threads its
the logging system will eventually become saturated by the volume of these
requests coming from many RegionServers.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
