Hi Frank,

thanks, the fix appears to work.


Christof

Frank Fock wrote:
Hi Christof,

Sorry, this seems to be indeed a regression.
Please try the fix below to fix broadcast
async request processing:

Index: Snmp.java
===================================================================
--- Snmp.java    (revision 402)
+++ Snmp.java    (working copy)
@@ -1117,17 +1117,7 @@
       synchronized (pendingRequests) {
         request = (PendingRequest) pendingRequests.get(handle);
         if (request != null) {
-          if (request.isResponseReceived()) {
-            pendingRequests.remove(handle);
-            request = null;
-            if (logger.isInfoEnabled()) {
- logger.info("Received second response for request with handle "+
-                          handle);
-            }
-          }
-          else {
-            request.responseReceived();
-          }
+          request.responseReceived();
         }
       }
       if (request == null) {


I will consider isolating the Counter64 check into
a protected method.

Best regards,
Frank



Christof Meerwald wrote:
Hi,

it appears that with 1.9.1 SNMP broadcasts don't work any more - to be
more precise, you now only get the first response (with code that
worked perfectly fine with previous snmp4j versions). I think this is
related to the following "fix" in 1.9.1:

"Fixed: With async request processing and retry>0 there could have been
a race condition where a response event had been generated twice for a
request."

which appears to now only allow a single response event for each
request (even if the request is not cancelled in the listener as
mentioned in the JavaDoc: "Not canceling a request immediately can be
useful when sending a request to a broadcast address.")


I also noticed that 1.9.1 now contains a factory interface for creating
the PDU class for incoming messages. While this allows one to provide a
PDUv1 class that doesn't throw an exception when encountering a
Counter64 value, it's still a bit of a hassle as it requires quite a
bit of code duplication. I think, snmp4j could quite easily simplify
that particular case further by moving the Counter64 check into a
separate method. That way, a derived "PDUv1WithCounter64" could easily
overwrite that method without having to duplicate any of the decodeBER
code. So my proposal would be to add a new protected method to PDUv1:

protected void checkVariableBindingSyntax(VariableBinding vb)
  throws IOException
{
  if (vb.getSyntax() == SMIConstants.SYNTAX_COUNTER64) {
    throw new MessageException("Counter64 encountered in SNMPv1 PDU
"+
                               "(RFC 2576 ยง4.1.2.1)");
  }
}

and call checkVariableBindingSyntax(vb) in decodeBER (line 169).

Any thoughts on this?


Christof




--
Emnico Technologies Ltd - http://www.emnico.com
Dauntsey House, Stonehill Green, Westlea, Swindon SN5 7HB
_______________________________________________
SNMP4J mailing list
[email protected]
http://lists.agentpp.org/mailman/listinfo/snmp4j

Reply via email to