[RIVER-230] (mux) SelectionManager catch Error block assumes getMessage() returns non-null

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: jtsk_2.0
• Fix Version/s: River_2.1.2
• Component/s: net_jini_jeri
• Bugtraq ID: 6332547
• Assignee: Peter Jones
• Reporter: Fred Oliver
• Created: 10/Aug/07 22:34
• Updated: 06/Jan/10 06:04
• Resolved: 24/Apr/08 04:25

Description

Bugtraq ID 6332547

com.sun.jini.jeri.internal.runtime.SelectionManager's waitForReadyKey method has the following catch block in order to work around JDK bug 4458268:

		} catch (Error e) {
		    if (e.getMessage().startsWith("POLLNVAL")) {
			Thread.yield();
			continue;		// work around 4458268
		    } else {
			throw e;
		    }

The problem is that Error.getMessage() may return null (in cases other than those that this code was intended to handle), in which case a NullPointerException will occur. Such a NullPointerException has been observed in the output of tests that cause an out-of-memory condition due to thread creation (in particular, the regression test /vob/qa/jtreg/net/jini/jeri/tcp/outOfThreads/OutOfThreads.java on Solaris 9):

Oct 3, 2005 2:00:52 AM com.sun.jini.jeri.internal.runtime.SelectionManager$SelectLoop run
WARNING: select loop throws
java.lang.NullPointerException
	at com.sun.jini.jeri.internal.runtime.SelectionManager.waitForReadyKey(SelectionManager.java:369)
	at com.sun.jini.jeri.internal.runtime.SelectionManager.access$600(SelectionManager.java:80)
	at com.sun.jini.jeri.internal.runtime.SelectionManager$SelectLoop.run(SelectionManager.java:287)
	at com.sun.jini.thread.ThreadPool$Worker.run(ThreadPool.java:136)
	at java.lang.Thread.run(Thread.java:595)
(repeated)

With the fix for 6304782, the select loop will continue after such an exception, but the type and stack trace of the original exception does not get logged properly (only the NPE gets logged).

-------------

Two examples of this phenomenon in test failures: (Sun internal URLs removed).

In the second case, the VM terminated abruptly because of the out-of-memory condition anyway (not much we can do about that test failure mode?).

In the first case, the harness timed out the test. Was that because I/O could not make progress because of the recurring select loop failures (perhaps also slowed by the loop throttling after repeated failures)?

Issue Links

• is related to: RIVER-304 Reactivate River jtreg tests