You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Garrett Rooney <ro...@electricjellyfish.net> on 2004/10/10 04:53:57 UTC
Version of Subversion on svn.collab.net
So I've been wondering... Why are we sticking to released versions of
Subversion on svn.collab.net?
In the past we used to run reasonably current development snapshots (ok,
so that's the only kind of snapshot that existed at the time, but
still), and it let us catch problems that would not ordinarily be
noticed during development. If we'd been doing the same thing with
1.1.0 release candidates we would have had a shot at noticing the 'svn
list' slowdown before releasing it into the wild.
I'll note that the Apache guys seem to have a policy of running release
candidates of httpd on apache.org before cutting a release, and it seems
like a reasonable thing to do.
Plus, with our backwards compatability policy we shouldn't be causing
problems for people trying to access the repository by running more
bleading edge versions of svn on it, and even if we do have an
occasional problem it would be a good way to shake out version specific
issues, since I suspect svn.collab.net gets hit with a wider variety of
svn versions than anyone actually tests with on a regular basis.
So I'd like to propose a policy of testing RCs on svn.collab.net before
rolling out an actual release, and beyond that, I think it might be
worthwhile to upgrade to dev snapshots of trunk from time to time in
order to shake out problems before they become too intrenched. Perhaps
that would only be done as we get closer to 1.2.x being released, I'm
not sure how the best way to do this would be, but it seems like
something worth discussing.
Any thoughts?
-garrett
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by kf...@collab.net.
Ben Reser <be...@reser.org> writes:
> On Mon, Oct 11, 2004 at 07:38:49PM -0500, kfogel@collab.net wrote:
> > But the new piece of data we have today is the number of people
> > encountering it, and how severe they find it (both anecdotal, of
> > course, but still worth something IMHO).
>
> Agreed, which is why I'm cutting a 1.0.9 on Wednesday and did the work
> to milestone 1.1.1 today. :)
You, sir, rock my socks :-).
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by Ben Reser <be...@reser.org>.
On Mon, Oct 11, 2004 at 07:38:49PM -0500, kfogel@collab.net wrote:
> But the new piece of data we have today is the number of people
> encountering it, and how severe they find it (both anecdotal, of
> course, but still worth something IMHO).
Agreed, which is why I'm cutting a 1.0.9 on Wednesday and did the work
to milestone 1.1.1 today. :)
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by kf...@collab.net.
Ben Reser <be...@reser.org> writes:
> I'd like to remind everyone that we knew about the performance
> regression at the 1.1.0 release. Having 1.1.0-rc4 on svn.collab.net
> might have made more people aware of it. But it was not something that
> wasn't known. We made a judgement call on how serious it was based upon
> the data we had at the time. I suspect that it wouldn't have changed
> anything though had 1.1.0-rc4 been on svn.collab.net. Most of us don't
> use `svn ls`.
I think you're right.
But the new piece of data we have today is the number of people
encountering it, and how severe they find it (both anecdotal, of
course, but still worth something IMHO).
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by Ben Reser <be...@reser.org>.
On Mon, Oct 11, 2004 at 02:16:12PM -0700, Justin Erenkrantz wrote:
> No. This is the perfect case of why we want to eat our dog food. In a
> perfect world, we would have deployed rc4 to svn.collab.net for a few days.
> Then, we likely would have spotted this regression and then fixed it
> instead of including the regression in 1.1.0.
>
> We wrote the code, so we sure better be able to cope with the latest
> versions! We *want* to see these problems before our other users do.
> svn.collab.net has the highest percentage of users that'll be able to fix
> any problem in Subversion. It's unlikely any other public repository does.
>
> I really don't like the implication that the people who wrote the code
> don't trust it while telling everyone else to upgrade. That's bad.
I'd like to remind everyone that we knew about the performance
regression at the 1.1.0 release. Having 1.1.0-rc4 on svn.collab.net
might have made more people aware of it. But it was not something that
wasn't known. We made a judgement call on how serious it was based upon
the data we had at the time. I suspect that it wouldn't have changed
anything though had 1.1.0-rc4 been on svn.collab.net. Most of us don't
use `svn ls`.
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by Garrett Rooney <ro...@electricjellyfish.net>.
C. Michael Pilato wrote:
> Justin Erenkrantz <ju...@erenkrantz.com> writes:
>
>
>>I really don't like the implication that the people who wrote the
>>code don't trust it while telling everyone else to upgrade. That's
>>bad. -- justin
>
>
> Um. Go read the rest of the thread. Especially the part where I said:
>
> This is why we should have rolled *1.1.0-rc4* out on svn.collab.net
> *immediately after* releasing it. This is rather why we have RCs at
> all.
I think we're all in violent agreement at this point ;-)
-garrett
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by "C. Michael Pilato" <cm...@collab.net>.
Justin Erenkrantz <ju...@erenkrantz.com> writes:
> I really don't like the implication that the people who wrote the
> code don't trust it while telling everyone else to upgrade. That's
> bad. -- justin
Um. Go read the rest of the thread. Especially the part where I said:
This is why we should have rolled *1.1.0-rc4* out on svn.collab.net
*immediately after* releasing it. This is rather why we have RCs at
all.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
--On Sunday, October 10, 2004 10:03 AM -0500 "C. Michael Pilato"
<cm...@collab.net> wrote:
> "C. Michael Pilato" <cm...@collab.net> writes:
>
>> If a problem is found in one of our release candidates or final
>> releases that is so bad that we would be tempted to upgrade
>> svn.collab.net, then that problem is bad enough to require us to
>> roll a new release candidate or bugfix release, at which time our
>> server can be upgraded accordingly.
>
> For example, this ra->get_dirs() performance death that's living in
> our 1.1.0 release. Had I upgraded svn.collab.net to 1.1.0, we'd all
> be suffering this awful performance. And why haven't we rolled 1.1.1
> with this fix?
No. This is the perfect case of why we want to eat our dog food. In a
perfect world, we would have deployed rc4 to svn.collab.net for a few days.
Then, we likely would have spotted this regression and then fixed it
instead of including the regression in 1.1.0.
We wrote the code, so we sure better be able to cope with the latest
versions! We *want* to see these problems before our other users do.
svn.collab.net has the highest percentage of users that'll be able to fix
any problem in Subversion. It's unlikely any other public repository does.
I really don't like the implication that the people who wrote the code
don't trust it while telling everyone else to upgrade. That's bad. --
justin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by "C. Michael Pilato" <cm...@collab.net>.
Garrett Rooney <ro...@electricjellyfish.net> writes:
> > For example, this ra->get_dirs() performance death that's living in
> > our 1.1.0 release. Had I upgraded svn.collab.net to 1.1.0, we'd all
> > be suffering this awful performance. And why haven't we rolled 1.1.1
> > with this fix?
> >
>
> I realize this, that's why I think we should have rolled 1.1.0 out on
> svn.collab.net before releasing it, so we would have noticed the
> problem and corrected it.
Close, but still wrong. :-)
This is why we should have rolled *1.1.0-rc4* out on svn.collab.net
*immediately after* releasing it. This is rather why we have RCs at
all.
> I imagine that having us be exposed to the problem directly would
> provide more of a motivation, and the issue would have been
> corrected more rapidly as a result.
Oh, absolutely.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by Garrett Rooney <ro...@electricjellyfish.net>.
C. Michael Pilato wrote:
> "C. Michael Pilato" <cm...@collab.net> writes:
>
>
>>If a problem is found in one of our release candidates or final
>>releases that is so bad that we would be tempted to upgrade
>>svn.collab.net, then that problem is bad enough to require us to
>>roll a new release candidate or bugfix release, at which time our
>>server can be upgraded accordingly.
>
>
> For example, this ra->get_dirs() performance death that's living in
> our 1.1.0 release. Had I upgraded svn.collab.net to 1.1.0, we'd all
> be suffering this awful performance. And why haven't we rolled 1.1.1
> with this fix?
>
I realize this, that's why I think we should have rolled 1.1.0 out on
svn.collab.net before releasing it, so we would have noticed the problem
and corrected it. I imagine that having us be exposed to the problem
directly would provide more of a motivation, and the issue would have
been corrected more rapidly as a result. As I recall this happened with
at least one release back in the day. We released a version, rolled it
out on svn.collab.net, realized some new code in the fs was unusably
slow, and fixed it. Until we rolled it out nobody had realized the
performance problem.
As for why 1.1.1 hasn't been released yet, I believe that discussion is
happening in other threads...
-garrett
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by "C. Michael Pilato" <cm...@collab.net>.
"C. Michael Pilato" <cm...@collab.net> writes:
> If a problem is found in one of our release candidates or final
> releases that is so bad that we would be tempted to upgrade
> svn.collab.net, then that problem is bad enough to require us to
> roll a new release candidate or bugfix release, at which time our
> server can be upgraded accordingly.
For example, this ra->get_dirs() performance death that's living in
our 1.1.0 release. Had I upgraded svn.collab.net to 1.1.0, we'd all
be suffering this awful performance. And why haven't we rolled 1.1.1
with this fix?
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Version of Subversion on svn.collab.net
Posted by "C. Michael Pilato" <cm...@collab.net>.
Garrett Rooney <ro...@electricjellyfish.net> writes:
> So I've been wondering... Why are we sticking to released versions of
> Subversion on svn.collab.net?