[SNMP4J] Protocol mismatch (or socket errors in general) leadto many connection attempts using snmp4j 1.10.2

[Warren Cooper]

(Resending to include the mailing list in case you missed my followup 
questions. Thanks)

 >...Any suggestions are welcome...
How about if the the queued entries obey the retry configuration 
setting? i.e. if the retry count is set to 1, then a message in the 
pending list will only register for OP_CONNECT one more time. After that 
it is simply discarded from the pending list.

I am seeing cases where the socket entry is null in connectChannel( ) 
because the socket channel fails this check:
--> if (sc.equals(entry.getSocket().getChannel())) { <--
So no event is fired and that warning message
--> logger.warn("Message not found on finish connection");  <--
is sent instead.  Once a few messages are queued into the pending member 
and the server socket connection is down, this really churns cpu 
streaming an endless amount of these 'Message not found...' messages and 
eats memory quickly if the traps include variables w/ alot of data.

I'm investigating overriding the DefaultTcpTransportMapping class, but I 
just wanted to make certain you understood the potential impact of the 
problem I am reporting.
-thanks

Frank Fock wrote:
> Warren,
>
> You can (and should) handle such a situation in your code
> by implementing a TransportStateListener and add it
> to your TcpTransportMapping.
>
> I have no idea how this could be solved in a generic
> way without a negative impact on existing code.
> Any suggestions are welcome.
>
> Regards,
> Frank
>
> On 28.12.2009 15:57, Warren Cooper wrote:
>> If there is a connection problem between an SNMP agent configured to
>> send TCP  in its attempt to send events to an SNMP server, and there is
>> a problem making a connection, many attempts to send the events will
>> fail.  ( It's a problem because the pending list (member in
>> DefaultTCPTransportMapping) will continue to grow.
>>
>> To see the problem, configure an SNMP client to send 40 events over TCP
>> to a SNMP server expecting UDP.
>> (But this problem would happen in any case where  if
>> (entry.getSocket().isConnected()) ) check in processPending() is not 
>> true.
>> As you know, this is a serious problem since CPU resources are impacted
>> and memory could eventually be exhausted if the problem continues for
>> over an extended period of time and enough events are written to the
>> 'pending' list object instance.
>>
>> thanks
>> _______________________________________________
>> SNMP4J mailing list
>> SNMP4J at agentpp.org
>> http://lists.agentpp.org/mailman/listinfo/snmp4j
>