Avoid NPE potentially caused by race condition

[jdpgrailsdev]

Periodically, we are seeing a java.lang.NullPointerException with the following stack trace:

…t.spy.memcached.protocol.TCPMemcachedNodeImpl.toString (TCPMemcachedNodeImpl.java:488)
java.util.Formatter$FormatSpecifier.printString (Formatter.java:2886)
java.util.Formatter$FormatSpecifier.print (Formatter.java:2763)
java.util.Formatter.format (Formatter.java:2520)
java.util.Formatter.format (Formatter.java:2455)
java.lang.String.format (String.java:2940)
net.spy.memcached.compat.log.AbstractLogger.info (AbstractLogger.java:161)
net.spy.memcached.MemcachedConnection.handleIO (MemcachedConnection.java:716)
net.spy.memcached.MemcachedConnection.handleIO (MemcachedConnection.java:450)
net.spy.memcached.MemcachedConnection.run (MemcachedConnection.java:1457)

As far as I can tell, there must be some sort of race going on where the second call to getSk() returns null even after the first check passes, causing the de-reference of the isValid() method to throw the NPE. I am wondering if the real fix is to put some sort of lock around the SK or have the getSk() method return a copy to avoid this issue. It's unclear what the impact is when this happens, but I suspect it is causing an operation to fail/not be executed because the runtime exception.