It's a problem I've been aware of for some time, many tests don't
specify a port, they just want three or four or however many registrar's
started and they have to specify it using a LookupLocator, unfortunately
you can't specify an ephemeral LookupLocator, it's not allowed, so if
you don't specify a particular port, you get 4160 and the portInUse
method is supposed to work around that.
When a test finds a port in use, it updates the registrar wanted to an
available port. If it thinks 4160 is available, when it isn't, boom,
test failed.
Regards,
Peter.
Greg Trasuk wrote:
I hope you're not adding this to try to help my configuration issue,
because I'm pretty sure that's not the problem, and I hate to screw
around with the regression tests unneccessarily.
Evidence-wise, the "lookup started != initial lookups wanted" issue
looks like this:
When I enable logging at the FINE level, I see three registrars started,
with unicast discovery on random ports - this is Reggie's fallback
position when it attempts to open its unicast port and it's already in
use. However, the test that's checking for "lookup started == initial
lookups wanted" is testing for three separate registrars on the default
port. Obviously that's an impossible proposition. My working theory is
that the test is supposed to be configured to start Reggie on three
separate unicast ports (e.e. 7700, 7701, 7702), and the assertion is
checking to see if that happened correctly. This theory is supported by
the existence of
"qa/src/com/sun/jini/test/share/reggie3_2Ports.properties", which
includes properties called
"net.jini.core.lookup.ServiceRegistrar.port.0=7700",
"net.jini.core.lookup.ServiceRegistrar.port.1=7701",
"net.jini.core.lookup.ServiceRegistrar.port.2=7702", etc.
Hence my suspicion that I don't understand the configuration mechanism
well enough yet. My theory is that the test is supposed to start up
Reggie's on three different ports, but there is something about my test
environment setup that causes the configuration to be missed. Anpother
possibility is that the actual test configuration got messed up at some
point (I did see a comment in one of the properties files about "these
entries are commented out because they are now included in the test
descriptions".
Testing that theory is going to take me a little while to understand the
multi-layer test configuration mechanism. Unfortunately I've got a full
morning of other work ahead of me.
Anyhow, I don't really think changing the tests is the correct approach
at this point (at least in my troubleshooting process).
Cheers,
Greg.
On Tue, 2013-04-09 at 07:05, peter_firmst...@apache.org wrote:
Author: peter_firmstone
Date: Tue Apr 9 11:05:32 2013
New Revision: 1465969
URL: http://svn.apache.org/r1465969
Log:
Be more thorough when checking for port availability on localhost during
testing, don't assume port free if current test isn't using it.
Modified:
river/jtsk/skunk/qa_refactor/trunk/qa/src/com/sun/jini/test/share/LookupServices.java
Modified:
river/jtsk/skunk/qa_refactor/trunk/qa/src/com/sun/jini/test/share/LookupServices.java
URL:
http://svn.apache.org/viewvc/river/jtsk/skunk/qa_refactor/trunk/qa/src/com/sun/jini/test/share/LookupServices.java?rev=1465969&r1=1465968&r2=1465969&view=diff
==============================================================================
---
river/jtsk/skunk/qa_refactor/trunk/qa/src/com/sun/jini/test/share/LookupServices.java
(original)
+++
river/jtsk/skunk/qa_refactor/trunk/qa/src/com/sun/jini/test/share/LookupServices.java
Tue Apr 9 11:05:32 2013
@@ -27,7 +27,15 @@ import com.sun.jini.test.spec.discoverys
import com.sun.jini.test.spec.discoveryservice.AbstractBaseTest;
import
com.sun.jini.test.spec.discoveryservice.AbstractBaseTest.DiscoveryStruct;
import com.sun.jini.test.spec.discoveryservice.AbstractBaseTest.RegGroupsPair;
+import java.io.IOException;
import java.net.InetAddress;
+import java.net.InetSocketAddress;
+import java.net.ServerSocket;
+import java.net.Socket;
+import java.net.SocketAddress;
+import java.net.SocketException;
+import java.net.SocketOptions;
+import java.net.SocketTimeoutException;
import java.net.UnknownHostException;
import java.rmi.RemoteException;
import java.util.ArrayList;
@@ -660,12 +668,47 @@ public class LookupServices {
* @return true if port in use.
*/
private boolean portInUse(int port) {
+ if (port == 0) return false; // Ephemeral
for(int i=0;i<lookupsStarted.size();i++) {
LocatorGroupsPair pair = lookupsStarted.get(i);
int curPort = (pair.getLocator()).getPort();
if(port == curPort) return true;
}//end loop
- return false;
+ // Open a client ephemeral socket and attempt to connect to
+ // port on localhost to see if someone's listening.
+ Socket sock = null;
+ try {
+ sock = new Socket();
+ if (sock instanceof SocketOptions){
+ // Socket terminates with a RST rather than a FIN, so there's
no TIME_WAIT
+ try {
+ ((SocketOptions) sock).setOption(SocketOptions.SO_LINGER,
Integer.valueOf(0));
+ } catch (SocketException se) {
+ // Ignore, not supported.
+ logger.log( Level.FINEST, "SocketOptions set SO_LINGER threw an
Exception", se);
+ }
+ }
+ SocketAddress add = new InetSocketAddress(port);
+ sock.connect(add, 3000); // Try to connect for up to three seconds
+ // We were able to connect to a socket listening on localhost
+ return true;
+ } catch (SocketTimeoutException e){
+ // There might be a stale process assume in use.
+ logger.log( Level.FINEST, "Socket timed out while trying to
connect", e);
+ return true;
+ } catch (IOException e){
+ // There was nothing listening on the socket so it's probably free.
+ // or it timed out.
+ return false;
+ } finally {
+ if (sock != null){
+ try {
+ sock.close();
+ } catch (IOException ex){
+ logger.log( Level.FINEST, "Socket threw exception while
attempting to close", ex);
+ }// Ignore
+ }
+ }
}//end portInUse
private void refreshLookupLocatorListsAt(int index){
