SPY-107: Ensure state of timedout is known across IO and user threads.

After seeing this in a log, I re-thought through the series of events here.

This flag on the op is known by user threads and the IO thread.  Thus it
is modified in these two different ways:

* operation is created and enqueued
* OperationFuture is returned
* .get() is called on OF
 **timed out**

* operation is removed from input queue
* elapsed time is checked
  **timed out**

Change-Id: I879f9b6459aa2fce1ddd97c5803da8c5392453a5
Reviewed-on: http://review.couchbase.org/23798
Reviewed-by: Michael Nitschinger <[email protected]>
Reviewed-by: Matt Ingenthron <[email protected]>
Tested-by: Matt Ingenthron <[email protected]>

Author: ingenthr

src/main/java/net/spy/memcached/protocol/BaseOperationImpl.java

@@ -57,7 +57,7 @@ public abstract class BaseOperationImpl extends SpyObject implements Operation {
   private OperationException exception = null;
   protected OperationCallback callback = null;
   private volatile MemcachedNode handlingNode = null;
-  private boolean timedout;
+  private volatile boolean timedout;
   private long creationTime;
   private boolean timedOutUnsent = false;
   protected Collection<MemcachedNode> notMyVbucketNodes =
@@ -208,14 +208,10 @@ public synchronized boolean isTimedOut(long ttlMillis) {
       timedout = true;
       callback.receivedStatus(TIMED_OUT);
       callback.complete();
-    } else {
+    } // else
       // timedout would be false, but we cannot allow you to untimeout an
-      // operation
-      if (timedout) {
-        throw new IllegalArgumentException("Operation has already timed out;"
-            + " ttl specified would allow it to be valid.");
-      }
-    }
+      // operation.  This can happen when the latch timeout is shorter than the
+      // default operation timeout.
     return timedout;
   }