[HBASE-10949] Reversed scan could hang

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.98.0
• Fix Version/s: 0.99.0, 0.98.2
• Hadoop Flags: Reviewed
• Assignee: Jimmy Xiang
• Reporter: Jimmy Xiang
• Created: 09/Apr/14 18:55
• Updated: 21/Feb/15 23:33
• Resolved: 13/Apr/14 18:38

Description

When I play with ITBLL (from trunk tip), sometimes, meta scan hangs when the cluster is rolling restarted. When this happens, the master takes about 1000% of CPU. It looks like there is an infinite loop somewhere. The logs show nothing interesting except some meta scanner RPC calls timed out. Jstask shows the 10 high QoS RPC handlers are busy with meta scanning.

However, if I run it again without HBASE-10018, things are fine. I suspect there is something to do with the small/reverse scan.

By the way, I see this problem even with log replay off and hfile version = 2.

Issue Links

• relates to: HBASE-10018 Remove region location prefetching