On 05/10/10 11:53 AM, John Fischer wrote:
Sarah and Dave,
Thanks for the discussion. I think that this, /etc/nodename, is the
controversial
part of the project. Thus before going to the committee it needs to be
really
hashed out.
To me the real question is whether or not to keep the /etc/nodename
around for read-only purposes. Setting the host using this interface is
just
not tenable since (as I think Dave stated) that would give us 2
interfaces to
set the name. Keeping it around as a read-only file allows for scripts and
other programs to still get the name via this method (i.e., we don't break
a bunch of stuff). This also fits in well with various ARC case log history
of keeping interfaces around that are used by scripts. Perhaps adding
a do not use this file to set the name comment, updating nodename(1) and
making it read only would be sufficient.
/etc/nodename is not authoritative for all cases, as noted in
nodename(4). Any application that references it is already broken in
many existing cases. Deleting it is quite reasonable.
A question... when a user uses uname -S does this update the SMF repo
with this?
That is an interesting question. It would make sense. If this was done
then uname -S would probably check to see if the stored name matches
the name being set. If it doesn't match then update the SMF property.
Alternatively, it could be treated like it is today. Today if uname -S is
used to update the name then the name is updated but the storage
is left alone. This kind of bothers me though. It is kind of like a train
that looses a car but keeps on going. You know, oh well there goes a
car from the train. Also then when the system reboots again the name
is lost as if it was simply a temporary name for the system. If someone
uses uname -S I would think that they really want the name changed
period.
The semantics of uname -S have always been that it is not persistent.
Changing that seems unnecessary and unwise, as it would lead to
unexpected side-effects in any existing usage.
Dave
Additional opinions?
Thanks,
John
On 05/10/10 07:52 AM, Sarah Jelinek wrote:
One other option.. if we say SMF is the definitive source for
nodename, then we file an ARC case to state that, and still keep
/etc/nodename for compatibility, while letting users know that SMF is
the source for for this data. This might be confusing, but it is an
option.
A question... when a user uses uname -S does this update the SMF repo
with this?
thanks,
sarah
*****
On 05/10/10 08:38 AM, Sarah Jelinek wrote:
On 05/10/10 08:30 AM, Dave Miner wrote:
On 05/10/10 10:17 AM, Sarah Jelinek wrote:
I thought part of this proposal was to deprecate the use of this file.
As Erik points out it is documented today in nodename(4), and has been
around a really long time. I would think that we would file an ARC
case
to EOL this interface along with this proposal. If we are keeping it
around, I think the read-only option is a good choice. but, I think
even
if we do this, EOL'ing this as a writeable interface needs to be done.
I'm not sure how any proposal to make this read-only would work, since
any scripts that wrote here in the past are privileged and would just
blow right through mode bits.
Fair point. Then it seems as if we have to either remove
/etc/nodename(EOL it) or use the SMF properties in a transient way,
removing them after configuration.
sarah
Dave
thanks,
sarah
*****
On 05/10/10 06:21 AM, John Fischer wrote:
Erik,
Thanks!! You are not the first one to ask about /etc/nodename
and scripts. I think that we should probably kept it around in
some form too.
Thanks,
John
On 05/10/10 05:08 AM, Erik Nordmark wrote:
On 05/ 6/10 03:17 PM, John Fischer wrote:
All,
Attached is the first revision of the System Configuration SMF
service
nodename
design document.
Comments/Questions welcomed.
We currently document the existence of /etc/nodename in nodename(4).
Thus it wouldn't be unreasonable for a customer to have developed
scripts which write the nodename to /etc/nodename instead of using
uname -S.
What happens if such scripts are run in the new world?
Would it be helpful to provide a read-only /etc/nodename which
contains just a comment along the lines of:
# This file has been obsoleted. Use uname(1m) to set and get the
nodename
That way such a script would at least fail, instead of silently
continuing without updating the nodename.
Erik
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
