Re: Howto setup netsnmp for proxying and failover

Thanks for your answers!

2010/5/20 Dave Shield <[email protected]>

> On 19 May 2010 22:01, Boris Zweimueller <[email protected]>
> wrote:
> >> - I don't know how you'd want to handle receiving different answers from
> >> - the parralel proxied agents, but the above should be a workable
> arrangement.
> > Answers from different parallel proxies are simply compared.
> > If they are different this shows some missconfiguration/error.
>
> Surely that would depend on what the MIB ibjects being queried were?
> If you were querying statistics, then the counts might well be different.
> (even if only by one or two packets for a heavily loaded system).
>

There will be only one management console accessing the device. It is used
just to query and set configuration of the device (e.g. like a router). I
don't understand why for this setting there could be different responses in
normal operation (devices operate correctly) ?

I only a problem, when e.g. for a SET command, one device ist doing it, and
the other can not for whatever reason. I'm not sure what to do then. Some
kind of undoing is probably to much work to implement. As far as I know,
SNMP dows not have any undo support?


> Then the device(s) are restarted and reconfigured somethin like that.

 That would feel overkill for such object counters.
> Deciding whether or not to restart the device feels more a question
> for the management console.   Which implies that it would need to
> be able to retrieve the settings for each device individually.
>
> So the first thing I'd suggest is that you register the two device
> subagents independently  (perhaps using different contexts).
> That would then provide the flexibility to manage them independently,
> as well as the "combined proxy" approach.
>

Yes, this will be the setup. If the supermaster detects that one of the
parallel agents is 'down' it will inform the managment console to query the
'slave' directly. Of course this will be done with contexts.

> I already looked at the current proxy code and made some simple
> examples sending the same request multiple times.
>
> I found that the easiest method would probably be, if
> 'doublicate registrations' would be possible

 My immediate reaction is that this is probably the wrong track to take.
> The basic model of the agent is that there is one (and only one) handler
> responsible for any given request.
>   Rather than change this fundamental assumption,  you'd probably be
> better working with a new "combined proxy" handler - similar to the
> existing one, but splitting the request into multiple destinations, and
> handling the multiple responses.   Either wait for them all to come in
> before returning a result,  or (probably simpler to start with) returning
> the first result and discarding the rest.
>
>
> But handling this in a single module feels much simpler (and safer!)
> than fiddling about with the inner workings of the core agent processing.
>
>
I think you'r right:-)  For a fast start I will use (I'm already
implementing it) synchronized communication with the parallel proxies. Then
it is much easier to know which responses belong together. I plan to extend
it to async. communication when the synchron approach works.


> Dave
>

Boris

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

_______________________________________________
Net-snmp-users mailing list
[email protected]
Please see the following page to unsubscribe or change other options:
https://lists.sourceforge.net/lists/listinfo/net-snmp-users