On 02/18/11 16:46, John Fischer wrote:
Ethan,
Thanks for looking into these issues.
See below.
Thanks,
John
On 02/18/11 02:49 PM, Ethan Quach wrote:
544-550 - does this logic account for the fact that if
exclude_network == false, and networks == 0.0.0.0/0, (these are the
default values) then we really want to include all of the systems
networks? This block comment above at 528-539 does not indicate that.
It does account for the exclude_network == false and networks ==
0.0.0.0/0. The main difference
here was that the code needed to handle the case where these were
uninitialized. I will be handling
that case differently now, see the answer below.
544 - for the case where networks == 0.0.0.0/0, is self.networks
None here or will we enter this if clause?
I am handling it differently. In __init__() I set the networks and
exclude_networks to their defaults
(0.0.0.0/0, false respectively). Then in the case that the tool is
registering I check to see if
these have truly been initialized and if not then I do the original
function calls from __init__().
I'm still not quite following here. Let's say we're currently
processing some interface address, 10.0.0.1, and the 'self.exclude'
and 'self.networks' are set to their default values. At 543,
wouldn't 'in_net' be set to False? And if so, we'd not enter the if
at 545, and we wouldn't set 'include_it' to True. Which means we
just continue at 549. Doesn't that mean we're skipping interface
10.0.0.1 ?
No. If the interface is 10.0.0.1 and networks is the default
(0.0.0.0/0) then in_net will
be true.
johnf@solaris:507 $ python2.6
Python 2.6.4 (r264:75706, Jan 16 2011, 12:34:46) [C] on sunos5
Type "help", "copyright", "credits" or "license" for more information.
>>> from aimdns_mod import _convert_ipv4 as _convert_ipv4
>>> from aimdns_mod import _convert_cidr_mask as _convert_cidr_mask
>>> from aimdns_mod import compare_ipv4 as compare_ipv4
>>> from aimdns_mod import in_networks as in_networks
>>> in_net = in_networks('10.0.0.1/24', ['0.0.0.0/0'])
>>> print in_net
True
Since it is True the evaluation at line 545 causes the include_it to
be True.
The above was done from usr/src/cmd/installadm. I also tried the IP
address
with a mask of 32 and without a mask (10.0.0.1) which both succeeded.
OK, I think the part I missed was that with the default networks being
'0.0.0.0/0', every IP address is going to be calculated as "in" that
network, because its netmasks is 0.
So I think I am okay with your changes for this bugfix.
806-812 - For browse, why do we need to ask the SMF service for the
number of install services that are configured on the local
system? Browse, is supposed to be an outward looking query, so the
number of services we have configured on the running host seems
irrelevant.
Right, this is more of an educated guess only.
Some history....
As you will recall dns-sd loops indefinitely that is because it is
attempting to get all the
services that are registered. However, that has the undesirable
side effect of our various
tools needing to kill the application after some set time (typically
it was 5 seconds).
Yeah, I recall this, but that was a command-line interface. I was
under the impression that with us now calling into a library to
interact with mdns, there'd be more finer grained interfaces for us
to get exactly what we need without having to query the open wild to
verify local services.
We can take the remaining offline, or at least off this CR line.
thanks,
-ethan
Right. But we are only making an "educated guess" as to how many
times we need
to loop within the processing cycle. If we under process then we
might miss some
services. For the case of the client we are only interested in the
single service anyhow.
In the case of being on the server we get a reasonable number without
looping indefinitely
nor too little.
If I recall correctly we discussed this a while back and even though
neither of us liked it
it was better then setting an arbitrary number which is what I did
originally.
aimdns was designed to loop a reasonable number of times.
Originally that was 5 times with
a timeout of 1 second for the select call within the process loop.
What I found was that sometimes
that was insufficient. So I came up with an educated guess of the
number of local services * the
number of interfaces should be sufficient for all the registered
services to get processed.
So it seems to me that perhaps this browse function isn't a generic
"browse" function at all (which could be my disconnect here, because
I'm assuming it's supposed to be).
Well again it is just a guess. It could be a number like 5 * number
of interfaces if you like but that
might miss some services. Again this is only a server side "educated
guess".
Its internals (per this loop count) seem to be catered toward the use
on the AIl of the install services from the local machine. Would you
agree?
Sure. So how long to loop then? Or what's the alternative
algorithm? Again, this is really only
server side since the client side will fail the smf.AISCF() call with
the exception because the
service does not exist.
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
