[Mono-bugs] [Bug 62395][Nor] Changed - Kill() won't kill

bugzilla-daemon@bugzilla.ximian.com bugzilla-daemon@bugzilla.ximian.com
Wed, 4 Aug 2004 17:12:01 -0400 (EDT)


Please do not reply to this email- if you want to comment on the bug, go to the
URL shown below and enter your comments there.

Changed by michi@zeroc.com.

http://bugzilla.ximian.com/show_bug.cgi?id=62395

--- shadow/62395	2004-08-04 13:44:15.000000000 -0400
+++ shadow/62395.tmp.32471	2004-08-04 17:12:01.000000000 -0400
@@ -1,13 +1,13 @@
 Bug#: 62395
 Product: Mono: Runtime
 Version: unspecified
 OS: unknown
 OS Details: RedHat 8, Dual CPU machine
-Status: RESOLVED   
-Resolution: NOTABUG
+Status: REOPENED   
+Resolution: 
 Severity: Unknown
 Priority: Normal
 Component: misc
 AssignedTo: mono-bugs@ximian.com                            
 ReportedBy: michi@zeroc.com               
 QAContact: mono-bugs@ximian.com
@@ -39,6 +39,28 @@
 
 On RH9 with NPTL the second WriteLine doesn't appear, even without the
 sleep.  I guess this is just linuxthreads taking some time to clean up.
 
 Resolving as NOTABUG; reopen if you disagree.
 
+
+------- Additional Comments From michi@zeroc.com  2004-08-04 17:12 -------
+Ah, that causes me a bit of grief. The sleep doesn't really help in 
+my case. Scenario: we have a set of test suites for Ice. One of those 
+tests fault tolerance, that is, fail-over of client invocations to 
+replica servers. As part of the tests, we simulate a crashed server. 
+This works fine in C++ (where we call exit(1)) and Java (where we 
+call Runtime.getRuntime().halt(1).
+
+The problem in C# is that not all threads are killed immediately, 
+meaning that the server manages to actually return a reply to the 
+client before it dies. As a result, the tests fail because the client 
+actually gets a reply instead of getting a socket error (which would 
+lead the client to conclude that the server has died).
+
+Of course, part of the problem is the test setup here. But I think it 
+would be nice if Kill() would be a little bit more radical, along the 
+lines of _exit() or abort(). Any chance of doing that?
+
+Cheers,
+
+Michi.