Re: Peer twiddling

From: Adrian Chadd <adrian@dont-contact.us>
Date: Sat, 15 Sep 2001 10:21:56 -0600

On Sat, Sep 15, 2001, Henrik Nordstrom wrote:
> Adrian Chadd wrote:
>
> > *nod* So should they be formatted to be read by a computer, and have
> > some external program/programs handle converting them to other
> > formats?
>
> For which SNMP does a very good job with only a minimal amount of coding.

Ok.

> > What about comments on the original idea? :-) The guy who I'm doing it
> > for would like it to appear in -HEAD and then 2.4 so he doesn't have to
> > run a "custom squid" (I like people like this. :-)
>
> I assume this is the peer on/off toggle you were talking about?

Yup.

> If done in todays interfaces, my vote is the cache_object interface as
> indicated earlier.
>
> If a new interface for active control is to be set up, my votes is in order of
> preference:
> * UNIX socket command channel
> * SNMP

WHat would the UNIX socket message format look like?

> most of the -k commands we have today does not belong in SNMP I think. These
> are mostly of interest to the host, not your NMS. Some fits very well in a
> UNIX socket control channel.

Well, you'd be surprised what people want to do via an NMS.
This does include rotating logs, flushing dns/ipcache, and shutting
down squid. :)

> The peer on/off function also fits well in SNMP. It is of interest to your
> average network operator, and SNMP is well known. It is only a pity that most
> NMS systems is somewhat closed and does not allow the operator to easily
> extend them with new SNMP mibs and representations thereof. SNMP is defenitely
> easier to integrade in the average NMS than telnet or HTTP interfaces.

You won't be able to turn on a given peer:port via SNMP with only one
SET- you'll need to walk a list of index->peer mappings, and then
SET someoid.peeroid.port.enableoid={1|0} .

That kind of thing requires some NMS magic - this'll come in the form
of scripting.

Personally, I don't think SNMP should be the default method of
squid management. I think it should be HTTP - which allows it to
inherit all the existing ACL types when it comes to management -
and push out user-format-friendly and SNMP management via external
agents.

Adrian
Received on Sat Sep 15 2001 - 10:21:56 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:14:21 MST