Package org.apache.hadoop.hbase.ipc

Class BlockingRpcConnection.CallSender

java.lang.Object
java.lang.Thread
org.apache.hadoop.hbase.ipc.BlockingRpcConnection.CallSender
All Implemented Interfaces:
Runnable
Enclosing class:
BlockingRpcConnection
private class BlockingRpcConnection.CallSender extends Thread
If the client wants to interrupt its calls easily (i.e. call Thread#interrupt), it gets into a java issue: an interruption during a write closes the socket/channel. A way to avoid this is to use a different thread for writing. This way, on interruptions, we either cancel the writes or ignore the answer if the write is already done, but we don't stop the write in the middle. This adds a thread per region server in the client, so it's kept as an option.

The implementation is simple: the client threads adds their call to the queue, and then wait for an answer. The CallSender blocks on the queue, and writes the calls one after the other. On interruption, the client cancels its call. The CallSender checks that the call has not been canceled before writing it.

When the connection closes, all the calls not yet sent are dismissed. The client thread is notified with an appropriate exception, as if the call was already sent but the answer not yet received.