[HBASE-5201] Utilize TThreadedSelectorServer and remove redundant code in ThriftServer and HRegionThriftServer

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.94.0
• Component/s: regionserver
• Hadoop Flags: Reviewed

Release Note

New config parameters:

"hbase.thrift.selector.threads" Number of selector threads for reading and writing socket
"hbase.thrift.worker.threads" Number of threads for processing the thrift calls
"hbase.thrift.stop.timeout.seconds" Time to wait for server to stop gracefully
"hbase.thrift.accept.queue.size.per.selector" Maximum number of accepted elements per selector
"hbase.thrift.accept.policy" The strategy for handling new accepted connections

• Assignee: Scott Chen
• Reporter: Scott Chen
• Created: 14/Jan/12 22:48
• Updated: 17/Jul/13 11:57
• Resolved: 17/Jan/12 17:58

Description

TThreadedSelectorServer is good for RPC-heavy situation because IO are not limited to one CPU. See
https://issues.apache.org/jira/browse/Thrift-1167
I am porting the related classes form thrift trunk (it is not there in thrift-0.7.0).

There are lots of repeat codes in ThriftServer and HRegionThriftServer.
These codes are now moved to a Runnable called ThriftServerRunner.

Issue Links

• relates to: HBASE-8851 Ensure HBASE-5201 (Add TThreadedSelectorServer) is covered in Thrift 2