Please route new work through -mm tree?

Please route new work through -mm tree?

[Stephen Hemminger]

There is lots of stuff going on that has a potential to make things less
stable. How about routing any future TCP, fib, statistics, stuff through
Andrew's tree and not directly to Linus. That way we have some buffer
and can stablize some more. Alternatively, Dave you could publize your
staging area some more.

Re: Please route new work through -mm tree?

[Nivedita Singhvi]

Stephen Hemminger wrote:
> There is lots of stuff going on that has a potential to make things less
> stable. How about routing any future TCP, fib, statistics, stuff through
> Andrew's tree and not directly to Linus. That way we have some buffer
> and can stablize some more. Alternatively, Dave you could publize your
> staging area some more.

That would help us out a bit - we're testing mainline
and -mm regularly. This came up for me yesterday, too,
with Dave's tso patches (which I'm testing right now,
after getting past some hw issues). I was going to
suggest a separate bk tree or something that would
hold this stuff till it stabilized, before it needed
to go into mainline. Having that picked up by mm would
help us to increase coverage quite a bit.

thanks,
Nivedita

Re: Please route new work through -mm tree?

[Marcos D. Marado Torres]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Wed, 29 Sep 2004, Nivedita Singhvi wrote:

> I was going to
> suggest a separate bk tree or something that would
> hold this stuff till it stabilized, before it needed
> to go into mainline. Having that picked up by mm would
> help us to increase coverage quite a bit.

What about doing a -mm*-net* tree?

Mind Booster Noori

- -- 
/* *************************************************************** */
    Marcos Daniel Marado Torres	     AKA	Mind Booster Noori
    http://student.dei.uc.pt/~marado   -	  marado@student.dei.uc.pt
    () Join the ASCII ribbon campaign against html email, Microsoft
    /\ attachments and Software patents.   They endanger the World.
    Sign a petition against patents:  http://petition.eurolinux.org
/* *************************************************************** */
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)
Comment: Made with pgp4pine 1.76

iD8DBQFBWv2XmNlq8m+oD34RAie+AKDR+MbEBQ0nrNVVZb3lM6oEisRt9wCfc7cI
8yFjdsZ2gXcb81yVpY8Ukgo=
=LjKk
-----END PGP SIGNATURE-----

Re: Please route new work through -mm tree?

[David S. Miller]

On Wed, 29 Sep 2004 11:13:55 -0700
Nivedita Singhvi <niv@us.ibm.com> wrote:

> I was going to suggest a separate bk tree or something that would
> hold this stuff till it stabilized, before it needed
> to go into mainline.

Nivedita, it was you specifically who complained to me
when I used a seperate BK tree for networking changes
last month whilst Linus was away for 2 weeks.  You said
that this gave IBM too many trees to work from, so I told
you that what I was doing was a temporary thing until Linus
came back from his trip.

Now you're suggesting that I specifically do what you asked
me not to do a month ago.

So figure out what you really want.

And I have to warn people if they think that the churn is fast
and the rate of change in the networking is high right now, you
have seen absolutely nothing yet. :-)

Re: Please route new work through -mm tree?

[Nivedita Singhvi]

David S. Miller wrote:

> Nivedita, it was you specifically who complained to me
> when I used a seperate BK tree for networking changes
> last month whilst Linus was away for 2 weeks.  You said

"complained" ? no, no, honestly, it was just a very polite
inquiry :) :)

> And I have to warn people if they think that the churn is fast
> and the rate of change in the networking is high right now, you
> have seen absolutely nothing yet. :-)

yep, this is to some extent driving this, and things
like picking up the tso patches where we needed to
test prior to you checking them in.

Sure, we'd benefit from having fewer trees to test,
and since the -mm tree is already getting tested
and consuming the hw resources, we'd have benefited
from having the networking tree go into -mm. Hence
my previous inquiry to see if you could go through
-mm. If you're feeding to Linus immediately, then
we could just as well test mainline.

There is a need to have mainline stable, but that
can be solved by stretching out the bk snapshots and
increasing the number of rc releases so mainline
releases are fairly stable (something Andrew Morton
mentioned yesterday). That has been a separate
ongoing discussion among a lot of people.

Your point about there being a lot of other stuff
in -mm and it not being stable is taken, and it would
be nice to have a networking contained tree.

Which brings us to having your interim bk tree
available to test - which would be fine with me :)

What we're setting up is throwing the nightly
release onto 2 big boxes and running some heavy
duty networking stress tests.

All in the spirit of wanting to be helpful, honestly :).
Any suggestions welcome..

thanks,
Nivedita

Re: Please route new work through -mm tree?

[David S. Miller]

On Wed, 29 Sep 2004 13:58:55 -0700
Nivedita Singhvi <niv@us.ibm.com> wrote:

> Which brings us to having your interim bk tree
> available to test - which would be fine with me :)

There are people who know the location of it and are pulling
from it all the time :-)

	bk://kernel.bkbits.net/davem/net-2.6

I know for a fact that people such as Herbert Xu ping that
tree multiple times a day.

When I don't have any networking work present, that tree is
simply deleted.  So if you pull and get a URL error, that just
means no changes pending and Linus has everything. :-)

Re: Please route new work through -mm tree?

[Jeff Garzik]

David S. Miller wrote:
> When I don't have any networking work present, that tree is
> simply deleted.  So if you pull and get a URL error, that just
> means no changes pending and Linus has everything. :-)

No need to delete it.  When Linus merges your stuff, net-2.6 tree 
becomes equivalent to linux-2.6 tree.

When you next create and push local changes, a few days later, it is 
guaranteed that there will be no conflicts between your local and 
bkbits.net trees.

	Jeff

Re: Please route new work through -mm tree?

[David S. Miller]

On Wed, 29 Sep 2004 17:37:47 -0400
Jeff Garzik <jgarzik@pobox.com> wrote:

> David S. Miller wrote:
> > When I don't have any networking work present, that tree is
> > simply deleted.  So if you pull and get a URL error, that just
> > means no changes pending and Linus has everything. :-)
> 
> No need to delete it.  When Linus merges your stuff, net-2.6 tree 
> becomes equivalent to linux-2.6 tree.
> 
> When you next create and push local changes, a few days later, it is 
> guaranteed that there will be no conflicts between your local and 
> bkbits.net trees.

Ok, works for me.  I'll try to do this from now on.

Re: Please route new work through -mm tree?

[Jeff Garzik]

David S. Miller wrote:
> On Wed, 29 Sep 2004 17:37:47 -0400
> Jeff Garzik <jgarzik@pobox.com> wrote:
> 
> 
>>David S. Miller wrote:
>>
>>>When I don't have any networking work present, that tree is
>>>simply deleted.  So if you pull and get a URL error, that just
>>>means no changes pending and Linus has everything. :-)
>>
>>No need to delete it.  When Linus merges your stuff, net-2.6 tree 
>>becomes equivalent to linux-2.6 tree.
>>
>>When you next create and push local changes, a few days later, it is 
>>guaranteed that there will be no conflicts between your local and 
>>bkbits.net trees.
> 
> 
> Ok, works for me.  I'll try to do this from now on.


Cool.  FWIW the only time you should need to blow away and reclone a 
tree is when you reorder changesets..

	Jeff

Re: Please route new work through -mm tree?

[Gerrit Huizenga]

On Wed, 29 Sep 2004 13:22:47 PDT, "David S. Miller" wrote:
> On Wed, 29 Sep 2004 11:13:55 -0700
> Nivedita Singhvi <niv@us.ibm.com> wrote:
> 
> > I was going to suggest a separate bk tree or something that would
> > hold this stuff till it stabilized, before it needed
> > to go into mainline.
> 
> Nivedita, it was you specifically who complained to me
> when I used a seperate BK tree for networking changes
> last month whilst Linus was away for 2 weeks.  You said
> that this gave IBM too many trees to work from, so I told
> you that what I was doing was a temporary thing until Linus
> came back from his trip.
> 
> Now you're suggesting that I specifically do what you asked
> me not to do a month ago.
> 
> So figure out what you really want.
> 
> And I have to warn people if they think that the churn is fast
> and the rate of change in the networking is high right now, you
> have seen absolutely nothing yet. :-)

Hi Dave, having a bk tree which akpm can merge into his would
mean that all of our standard testing would get run over networking
changes as well as everything else.  We typically run some pretty
exhaustive tests on a small number of trees.  We try to focus on
-rc trees from Linus and the -mm trees.  As a result, networking
testing (by us) is done really late in the cycle compared to your
rate of change in pushing to Linus.

So, ideally, if you had your own tree published in such a way
that akpm could pull in, we'd probably have the best of all
worlds.

gerrit

Re: Please route new work through -mm tree?

[David S. Miller]

On Wed, 29 Sep 2004 11:03:43 -0700
Stephen Hemminger <shemminger@osdl.org> wrote:

> There is lots of stuff going on that has a potential to make things less
> stable. How about routing any future TCP, fib, statistics, stuff through
> Andrew's tree and not directly to Linus. That way we have some buffer
> and can stablize some more. Alternatively, Dave you could publize your
> staging area some more.

I'm not changing how I do things because:

1) Putting the net stuff into -mm makes debugging of networking
   changes harder, as -mm has a ton of experimental stuff in it
   as well.  -mm frequently makes machines unbootable, and particularly
   this is felt on non-x86 platforms such as sparc64 which is where
   I do all of my work.

2) I am more than happy to post the location of my net-2.6 tree
   but last time I used this actively (whilst Linus was away for
   2 weeks) the IBM folks complained that this gave them
   too many trees to work from.

Furthermore, I've never put any change into my tree that I didn't
think was ready for upstream.  When I post something here before
pushing to Linus, that's the chance for folks to test things out.