You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@commons.apache.org by Phil Steitz <ph...@gmail.com> on 2007/08/25 07:54:51 UTC
[all][poll] How should nigthlies / CI work?
Continuum works differently from the old bash script in a couple of
ways. First, it only executes builds when svn changes have happened.
So if there are no changes, there will be no "nightly build" for a
component. It also looks for changes and executes builds hourly, so
there can be multiple builds in a given day.
We need to decide what we jars we publish to the snapshot repo on
people.apache.org and what zips/tarballs we make available on the
"nightlies" page. Seems to me we have two choices.
1. Publish only the latest successful build (with build number and
date in the metadata but only commons-foo-x.y.z-SNAPSHOT.jar as the
jar name) and do the same with the zips/tarballs (only one is
available at a given time)
2. Publish jars with uniqueVersion=true, so they are each named
differently and keep a fixed number of them around, say 5. Similarly
for the zips/tarballs.
I think 1 is simpler and will be easier to maintain (no cron cleanups
necessary), but we could get 2. to work if others feel strongly that
we should be maintaining a stack of successful builds. Thoughts?
To get the zips/tarballs to work, components need to get m2 assemblies
defined for all of the components.
Phil
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Henri Yandell <fl...@gmail.com>.
On 8/25/07, Martin Cooper <ma...@apache.org> wrote:
> On 8/25/07, Phil Steitz <ph...@gmail.com> wrote:
> >
> > On 8/25/07, Noel J. Bergman <no...@devtech.com> wrote:
> > > Martin Cooper wrote:
> > >
> > > > > GUMP builds are deemed non-trusted, since GUMP downloads from
> > > > > non-ASF sites and includes them in builds without any vetting
> > > > > of the third party dependencies.
> > >
> > > > True, but it's not clear that everything in the public Maven repo
> > > > should be considered as "vetted" either.
> > >
> > > Exactly. Maven continues to be remiss in delivering on their goal of
> > > ensuring authenticated packages. I view anyone who uses the public
> > Maven
> > > repository as being foolish; competent Maven users have their own
> > private
> > > repositories.
> > >
> > > And, yes, the corollary that GUMP is building from the latest of
> > everything
> > > is another key reason not to use it for nightly builds.
> > >
> >
> > Another reason is that it is a little easier for us to manage
> > "publication" of the CI artifacts using Continuum / vmbuild. We can
> > publish both jars and zips/tarballs to a local maven repo on vmbuild
> > and set up rsynch to people.apache.org, eliminating some of the
> > ugliness in the bash setup.
> >
> > So can I get some feedback on the "what to publish" question?
>
>
> I'd say #1. The nightly builds are purely a convenience. If someone really
> needs something prior to the most recent nightly, they should be able to
> build it themselves, and it's not clear that us picking an arbitrary number
> would provide what they need in any case.
+1. #1 Most successful build only.
Hen
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Martin Cooper <ma...@apache.org>.
On 8/25/07, Phil Steitz <ph...@gmail.com> wrote:
>
> On 8/25/07, Noel J. Bergman <no...@devtech.com> wrote:
> > Martin Cooper wrote:
> >
> > > > GUMP builds are deemed non-trusted, since GUMP downloads from
> > > > non-ASF sites and includes them in builds without any vetting
> > > > of the third party dependencies.
> >
> > > True, but it's not clear that everything in the public Maven repo
> > > should be considered as "vetted" either.
> >
> > Exactly. Maven continues to be remiss in delivering on their goal of
> > ensuring authenticated packages. I view anyone who uses the public
> Maven
> > repository as being foolish; competent Maven users have their own
> private
> > repositories.
> >
> > And, yes, the corollary that GUMP is building from the latest of
> everything
> > is another key reason not to use it for nightly builds.
> >
>
> Another reason is that it is a little easier for us to manage
> "publication" of the CI artifacts using Continuum / vmbuild. We can
> publish both jars and zips/tarballs to a local maven repo on vmbuild
> and set up rsynch to people.apache.org, eliminating some of the
> ugliness in the bash setup.
>
> So can I get some feedback on the "what to publish" question?
I'd say #1. The nightly builds are purely a convenience. If someone really
needs something prior to the most recent nightly, they should be able to
build it themselves, and it's not clear that us picking an arbitrary number
would provide what they need in any case.
--
Martin Cooper
1. Most recent successful build only
> 2. A stack of the n (probably = 5) most recent successful builds
>
> I guess if we really want to hold on to the "nightly" idea, we could do
>
> 3. Symlink nightly the subset of 2 that correspond to the last n nights.
>
> 3. gets us back into bash/cron more deeply. 1. is bash/cron free
> (other than rsynch) and 2. requires cron cleanup. All are simpler
> than the current bash script, though. I am happy help implement any
> of these.
>
> Phil
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>
RE: [all][poll] How should nigthlies / CI work?
Posted by Jörg Schaible <Jo...@Elsag-Solutions.com>.
Phil Steitz wrote on Saturday, August 25, 2007 7:00 PM:
[snip]
> So can I get some feedback on the "what to publish" question?
>
> 1. Most recent successful build only
> 2. A stack of the n (probably = 5) most recent successful builds
+1, do not build if nothing has changed.
- Jörg
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Phil Steitz <ph...@gmail.com>.
On 8/25/07, Noel J. Bergman <no...@devtech.com> wrote:
> Martin Cooper wrote:
>
> > > GUMP builds are deemed non-trusted, since GUMP downloads from
> > > non-ASF sites and includes them in builds without any vetting
> > > of the third party dependencies.
>
> > True, but it's not clear that everything in the public Maven repo
> > should be considered as "vetted" either.
>
> Exactly. Maven continues to be remiss in delivering on their goal of
> ensuring authenticated packages. I view anyone who uses the public Maven
> repository as being foolish; competent Maven users have their own private
> repositories.
>
> And, yes, the corollary that GUMP is building from the latest of everything
> is another key reason not to use it for nightly builds.
>
Another reason is that it is a little easier for us to manage
"publication" of the CI artifacts using Continuum / vmbuild. We can
publish both jars and zips/tarballs to a local maven repo on vmbuild
and set up rsynch to people.apache.org, eliminating some of the
ugliness in the bash setup.
So can I get some feedback on the "what to publish" question?
1. Most recent successful build only
2. A stack of the n (probably = 5) most recent successful builds
I guess if we really want to hold on to the "nightly" idea, we could do
3. Symlink nightly the subset of 2 that correspond to the last n nights.
3. gets us back into bash/cron more deeply. 1. is bash/cron free
(other than rsynch) and 2. requires cron cleanup. All are simpler
than the current bash script, though. I am happy help implement any
of these.
Phil
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
RE: [all][poll] How should nigthlies / CI work?
Posted by "Noel J. Bergman" <no...@devtech.com>.
Martin Cooper wrote:
> > GUMP builds are deemed non-trusted, since GUMP downloads from
> > non-ASF sites and includes them in builds without any vetting
> > of the third party dependencies.
> True, but it's not clear that everything in the public Maven repo
> should be considered as "vetted" either.
Exactly. Maven continues to be remiss in delivering on their goal of
ensuring authenticated packages. I view anyone who uses the public Maven
repository as being foolish; competent Maven users have their own private
repositories.
And, yes, the corollary that GUMP is building from the latest of everything
is another key reason not to use it for nightly builds.
--- Noel
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Martin Cooper <ma...@apache.org>.
On 8/25/07, Noel J. Bergman <no...@devtech.com> wrote:
>
> > Why not use the Gump output for the nightlies?
>
> Because GUMP builds are deemed non-trusted, since GUMP downloads from
> non-ASF sites and includes them in builds without any vetting of the third
> party dependencies.
True, but it's not clear that everything in the public Maven repo should be
considered as "vetted" either.
What did occur to me - _after_ I sent my previous message, of course - is
that Gump builds against the latest of everything, and not a specified set
of versions (unless we're circumventing that deliberately). That's a pretty
good reason not to use it for nightlies.
--
Martin Cooper
--- Noel
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>
RE: [all][poll] How should nigthlies / CI work?
Posted by "Noel J. Bergman" <no...@devtech.com>.
> Why not use the Gump output for the nightlies?
Because GUMP builds are deemed non-trusted, since GUMP downloads from
non-ASF sites and includes them in builds without any vetting of the third
party dependencies.
--- Noel
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Rahul Akolkar <ra...@gmail.com>.
On 8/25/07, Dion Gillard <di...@trongus.com> wrote:
> On 8/26/07, Martin Cooper <ma...@apache.org> wrote:
> > On 8/24/07, Phil Steitz <ph...@gmail.com> wrote:
> > >
> > > Continuum works differently from the old bash script in a couple of
> > > ways. First, it only executes builds when svn changes have happened.
> > > So if there are no changes, there will be no "nightly build" for a
> > > component. It also looks for changes and executes builds hourly, so
> > > there can be multiple builds in a given day.
> >
> >
> > Given that, isn't Gump producing something closer to what we want for
> > nightly builds than Continuum? Why not use the Gump output for the
> > nightlies?
>
> On the other hand, why produce a new nightly if nothing's changed?
>
<snip/>
I think the net of it is that we need to flesh out what purposes Gump
and Continuum are serving, and what purposes they should continue to
serve. For better or worse, Gump may tend to more closely match our
bash nightlies.
-Rahul
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Paul Benedict <pb...@apache.org>.
I don't see a problem with having builds generated only on changes. I would
prefer that. But will all of ASF do the same? I would find it difficult to
determine what group is doing what: either nightly auto-builds or builds
daily only on commit.
Paul
On 8/26/07, Brett Porter <br...@gmail.com> wrote:
>
> On 26/08/07, Dion Gillard <di...@trongus.com> wrote:
> > > Given that, isn't Gump producing something closer to what we want for
> > > nightly builds than Continuum? Why not use the Gump output for the
> > > nightlies?
> >
> > On the other hand, why produce a new nightly if nothing's changed?
>
> Yah, Continuum certainly can spit out a whole new one every night if
> you want - it just doesn't make sense if the code hasn't changed
> (unless you are including development versions of dependencies that
> aren't in Continuum - which I don't believe would be the case for
> Commons).
>
> - Brett
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>
Re: [all][poll] How should nigthlies / CI work?
Posted by Brett Porter <br...@gmail.com>.
On 26/08/07, Dion Gillard <di...@trongus.com> wrote:
> > Given that, isn't Gump producing something closer to what we want for
> > nightly builds than Continuum? Why not use the Gump output for the
> > nightlies?
>
> On the other hand, why produce a new nightly if nothing's changed?
Yah, Continuum certainly can spit out a whole new one every night if
you want - it just doesn't make sense if the code hasn't changed
(unless you are including development versions of dependencies that
aren't in Continuum - which I don't believe would be the case for
Commons).
- Brett
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Dion Gillard <di...@trongus.com>.
On 8/26/07, Martin Cooper <ma...@apache.org> wrote:
> On 8/24/07, Phil Steitz <ph...@gmail.com> wrote:
> >
> > Continuum works differently from the old bash script in a couple of
> > ways. First, it only executes builds when svn changes have happened.
> > So if there are no changes, there will be no "nightly build" for a
> > component. It also looks for changes and executes builds hourly, so
> > there can be multiple builds in a given day.
>
>
> Given that, isn't Gump producing something closer to what we want for
> nightly builds than Continuum? Why not use the Gump output for the
> nightlies?
On the other hand, why produce a new nightly if nothing's changed?
>
> --
> Martin Cooper
>
>
> We need to decide what we jars we publish to the snapshot repo on
> > people.apache.org and what zips/tarballs we make available on the
> > "nightlies" page. Seems to me we have two choices.
> >
> > 1. Publish only the latest successful build (with build number and
> > date in the metadata but only commons-foo-x.y.z-SNAPSHOT.jar as the
> > jar name) and do the same with the zips/tarballs (only one is
> > available at a given time)
> > 2. Publish jars with uniqueVersion=true, so they are each named
> > differently and keep a fixed number of them around, say 5. Similarly
> > for the zips/tarballs.
> >
> > I think 1 is simpler and will be easier to maintain (no cron cleanups
> > necessary), but we could get 2. to work if others feel strongly that
> > we should be maintaining a stack of successful builds. Thoughts?
> >
> > To get the zips/tarballs to work, components need to get m2 assemblies
> > defined for all of the components.
> >
> > Phil
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> > For additional commands, e-mail: dev-help@commons.apache.org
> >
> >
>
--
dIon Gillard
Rule #131 of Acquisition: Information is Profit.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [all][poll] How should nigthlies / CI work?
Posted by Martin Cooper <ma...@apache.org>.
On 8/24/07, Phil Steitz <ph...@gmail.com> wrote:
>
> Continuum works differently from the old bash script in a couple of
> ways. First, it only executes builds when svn changes have happened.
> So if there are no changes, there will be no "nightly build" for a
> component. It also looks for changes and executes builds hourly, so
> there can be multiple builds in a given day.
Given that, isn't Gump producing something closer to what we want for
nightly builds than Continuum? Why not use the Gump output for the
nightlies?
--
Martin Cooper
We need to decide what we jars we publish to the snapshot repo on
> people.apache.org and what zips/tarballs we make available on the
> "nightlies" page. Seems to me we have two choices.
>
> 1. Publish only the latest successful build (with build number and
> date in the metadata but only commons-foo-x.y.z-SNAPSHOT.jar as the
> jar name) and do the same with the zips/tarballs (only one is
> available at a given time)
> 2. Publish jars with uniqueVersion=true, so they are each named
> differently and keep a fixed number of them around, say 5. Similarly
> for the zips/tarballs.
>
> I think 1 is simpler and will be easier to maintain (no cron cleanups
> necessary), but we could get 2. to work if others feel strongly that
> we should be maintaining a stack of successful builds. Thoughts?
>
> To get the zips/tarballs to work, components need to get m2 assemblies
> defined for all of the components.
>
> Phil
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>