[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: snmpconf A non-trivial policyAction example

Glenn Waters wrote:

> [gww] Actually, a MIB compiler is not needed by the agent - just the
> output from the MIB compiler. The output tends to be one or more files
> that could easily used by an agent (at least in systems I have used
> and designed in the past). Also, if the agent does have access to the
> compiled MIB the agent would have access to the type (e.g.: string,
> integer, etc.) information of the objects.

True. My main point is that the SNMP architecture has never required
agents to be datastores of this information (in any format) and that has
been a good thing. This isn't a good time to start.

I'd like to go further before anyone says "Given that this mib is
on the agent why can't the agent determine an object's type given its

First, there are various implementation-specific reasons why a tight
between the execution environment and the agent may not be possible. The

policy agent might be on a different system than the target system. Even
on the
same system the policy agent and the target agent might be in separate
independent processes (and not using an extensible agent). Even if part
of the same
(potentially virtual) agent, the agent might not offer a name to type

Second, even if the PM scripting engine can get access to that
information from
the local agent, what if the script is trying to request an object that
exist on the local system? Yes, the request will eventually fail, but I
think it
is important to be able to actually create the PDU and send it and get
proper failure back from the agent.