Re: OPS-NM status

[Tom Barron <tbarron@cisco.com>]

On Wed, Jan 23, 2002 at 05:20:19PM -0800, Steve Feldman wrote:
...
> During the meeting we had consensus that this effort is worthwhile,
> and that we should go over each of the points in the draft to
> decide whether to keep, modify, or drop them.
...
> 12. Devices must support both line-by-line and bulk commit, with
> syntax checking and rollback either line-by-line or bulk, as
> appropriate to the mode.
> 
> 13. Devices should support a ?syntax check only? mode, bulk or
> line-by-line, which only reports errors and does not actually apply
> changes to the configuration.
> 

  While I like the spirit of #12 and #13, I'm not convinced of the
usefulness of requiring ability to commit on boundaries corresponding
to physical input lines.  The critical thing is that a configuration
action be complete prior to a commit.  An operator intends to make a
config change from state A to state Z. What the operator does not want
to do is make incremental config state changes B, C, D, ... on the way
from A to Z just because these incremental changes correspond to
physical lines of config text input. One can "bulk" commit all the
lines resulting in the change from state A to state Z to achieve the
desired effect if each line itself would if individually committed
yield a coherent (though not currently desired) config state.  But
what if multiple physical lines are required to describe any coherent
new state?

  I'd like to see #12 and #13 put in a way that is decoupled from
the presentation layer matter of input lines that are approximately
80 chars or less - perhaps by talking about commit and rollback of
single or multiple configuration *objects* or some such.

- Tom Barron