[HBASE-12542] Delete a family of table online will crash regionserver

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Cannot Reproduce
• Fix Version/s: 2.0.0
• Component/s: regionserver
• Hadoop Flags: Reviewed
• Assignee: Peter Somogyi
• Reporter: Shaohui Liu
• Created: 20/Nov/14 04:31
• Updated: 17/Jun/22 18:37
• Resolved: 06/Jul/17 14:26

Description

Using alter command to delete a family of table online will make the regionsevers that serve the regions of the table crash.

alter 't', NAME => 'f', METHOD => 'delete'

The reason is that TableDeleteFamilyHandler in HMaster delete the family dir firstly and then reopen all the regions of table.
When the regionserver reopen the region, it will crash for the exception in flushing memstore to hfile of the deleted family during closing the region, because the parent dir of the hfile has been deleted in TableDeleteFamilyHandler.
See: TableDeleteFamilyHandler.java #57

A simple solution is change the order of operations in TableDeleteFamilyHandler.

• update table descriptor first,
• reopen all the regions,
• delete the the family dir at last.

Suggestions are welcomed.