You are viewing a plain text version of this content. The canonical link for it is here.
Posted to infrastructure-dev@apache.org by Jukka Zitting <ju...@gmail.com> on 2009/01/21 12:19:17 UTC

Zone for hosting git clones at Apache

Hi,

As discussed, I'd like to start moving the git clones from my server
to Apache infrastructure. The best way to do this seems to be to set
up a new Solaris zone for this purpose. Here's a rough starting point
for documenting such a zone:

<draft>
git
-----
PMC: infrastructure
Maintainers: Jukka Zitting <ju...@apache.org>,
             Grzegorz Kossakowski <gk...@apache.org>
Docs: TBD (perhaps use http://wiki.apache.org/general/GitAtApache ?)

  This zone is used to host git clones of Apache codebases.
</draft>

Any thoughts about this before I push forward with the idea?

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
On Mon, Jan 26, 2009 at 5:53 PM, Steve Loughran <st...@apache.org> wrote:
> Justin Erenkrantz wrote:
>> On Thu, Jan 22, 2009 at 9:18 AM, Steve Loughran <st...@apache.org> wrote:
>>> May want to think about (at least short term) an open git-at-apache mail
>>> list for all the hand-holding people like me will need
>>
>> Well, AIUI, git will only be optional - that is, SVN can still be
>> used.  (Jukka, please correct me if I'm misunderstanding your
>> intention.)
>
> I understand that, but I have effectively got a private branch of bits of
> hadoop right now, and git appeals to me just so that I can keep my desktop
> and laptop in sync with relative ease

OK. For now it's easiest to just use this list for peer support on git.

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Steve Loughran <st...@apache.org>.
Justin Erenkrantz wrote:
> On Thu, Jan 22, 2009 at 9:18 AM, Steve Loughran <st...@apache.org> wrote:
>> May want to think about (at least short term) an open git-at-apache mail
>> list for all the hand-holding people like me will need
> 
> Well, AIUI, git will only be optional - that is, SVN can still be
> used.  (Jukka, please correct me if I'm misunderstanding your
> intention.)

I understand that, but I have effectively got a private branch of bits 
of hadoop right now, and git appeals to me just so that I can keep my 
desktop and laptop in sync with relative ease

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

On Thu, Jan 22, 2009 at 6:37 PM, Justin Erenkrantz
<ju...@erenkrantz.com> wrote:
> On Thu, Jan 22, 2009 at 9:18 AM, Steve Loughran <st...@apache.org> wrote:
>> May want to think about (at least short term) an open git-at-apache mail
>> list for all the hand-holding people like me will need
>
> Well, AIUI, git will only be optional - that is, SVN can still be
> used.  (Jukka, please correct me if I'm misunderstanding your
> intention.)

Exactly. The Git mirrors are purely an add-on service on top of the
existing Subversion infrastructure. Changes to existing svn workflows
are neither needed nor planned.

> Hence, I think infra-dev should likely suffice to bootstrap
> documentation.  (More mailing lists is always a bad idea.)

Agreed.

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On Thu, Jan 22, 2009 at 9:18 AM, Steve Loughran <st...@apache.org> wrote:
> May want to think about (at least short term) an open git-at-apache mail
> list for all the hand-holding people like me will need

Well, AIUI, git will only be optional - that is, SVN can still be
used.  (Jukka, please correct me if I'm misunderstanding your
intention.)

Hence, I think infra-dev should likely suffice to bootstrap
documentation.  (More mailing lists is always a bad idea.)  -- justin

Re: Zone for hosting git clones at Apache

Posted by Steve Loughran <st...@apache.org>.
Paul Querna wrote:
> Jukka Zitting wrote:
>> Docs: TBD (perhaps use http://wiki.apache.org/general/GitAtApache ?)
>>
> 
> I think as we move this to a production service, it would make sense to 
> have authoritative documents that live under www.apache.org/dev/
> 
> -Paul

May want to think about (at least short term) an open git-at-apache mail 
list for all the hand-holding people like me will need

Re: Zone for hosting git clones at Apache

Posted by Paul Querna <pq...@apache.org>.
Jukka Zitting wrote:
> Docs: TBD (perhaps use http://wiki.apache.org/general/GitAtApache ?)
> 

I think as we move this to a production service, it would make sense to 
have authoritative documents that live under www.apache.org/dev/

-Paul

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,


2009/1/21 Justin Erenkrantz <ju...@erenkrantz.com>:
> From the peanut gallery, I think leveraging either harmonia (EU SVN
> mirror) or thor (new box @ OSU OSL) might make more sense than using a
> Solaris zone off helios/gaea/hyperion/etc.

That seems fine as well. I essentially need an account where I can
install git, run cronjobs, trigger actions based on received email
(commit notifications), and serve stuff over HTTP (with CGI support
for gitweb). It would be nice (but not necessary) if I could also run
git-daemon to allow cloning and pulling over the git: protocol (more
efficient than HTTP access).

On Wed, Jan 21, 2009 at 5:22 PM, Norman Maurer <no...@apache.org> wrote:
> I would like to know how much resources it needs before think about
> where to put it on.
> So any stats you are able to share with us (top, iostat etc) ?

I'm running on a virtual server, so I unfortunately don't have
accurate IO statistics. At current load levels I don't normally even
notice the effect of the git clones on the server.

For CPU usage, here's the output of vmstat at 60 second intervals when
I remotely request a new clone of the largest git mirror (cocoon.git,
at 500MB). During the first minute, when git looks through the
repository the CPU usage goes up to about 25% of the dedicated CPU
core (in an Intel Xeon processor) I have for the server. After that it
just pushes data out as fast as I can receive it (250kB/s) with
essentially zero CPU load.

# vmstat -a -n 60
procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----
 r  b   swpd   free  inact active   si   so    bi    bo   in    cs us sy id wa
 1  0      0 1698144      0      0    0    0     0     0    0     0  0  3 96  1
 2  0      0 1633228      0      0    0    0     0     0    0  4040 22  2 76  0
 1  0      0 1624044      0      0    0    0     0     0    0  3248  2  1 98  0
 1  0      0 1621568      0      0    0    0     0     0    0  3156  1  1 99  0
 1  0      0 1611760      0      0    0    0     0     0    0  3665  0  0 99  0
...

(The IO and swap stats are zero since I'm on a virtual server.)

After the initial clone an individual git pull doesn't even register
on the vmstat output. Similarly, at current usage rates, the gitweb
CGI puts essentially no load on the server.

Here's the vmstat output for running the daily git-svn update on all
the mirrors. It takes a few minutes, during which CPU usage goes up to
10%. Most of the time the server is just waiting for more data from
svn.eu.apache.org.

# vmstat -a -n 60
procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----
 r  b   swpd   free  inact active   si   so    bi    bo   in    cs us sy id wa
 0  0      0 1712016      0      0    0    0     0     0    0     0  0  3 96  1
 1  0      0 1703416      0      0    0    0     0     0    0  4167  4  4 88  4
 0  1      0 1694948      0      0    0    0     0     0    0  2981  3  5 79 13
 0  1      0 1698068      0      0    0    0     0     0    0  4439  5  6 79 10
 1  0      0 1699172      0      0    0    0     0     0    0  5577  4  5 80 11
 1  0      0 1710788      0      0    0    0     0     0    0  4544  5  6 78 11
 0  0      0 1712064      0      0    0    0     0     0    0  4675  2  2 94  2
 0  0      0 1712060      0      0    0    0     0     0    0  4629  0  0 99  0

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Grzegorz Kossakowski <gk...@apache.org>.
Norman Maurer pisze:
> I would like to know how much resources it needs before think about
> where to put it on.
> So any stats you are able to share with us (top, iostat etc) ?

See: http://markmail.org/message/bdwxha457y2fd6ez (second paragraph)

I guess that even if we won't consume all space at the beginning it would be reasonable to allocate around 10-15GB of
space for Git activity.

-- 
Grzegorz Kossakowski

Re: Zone for hosting git clones at Apache

Posted by Norman Maurer <no...@apache.org>.
I would like to know how much resources it needs before think about
where to put it on.
So any stats you are able to share with us (top, iostat etc) ?

Cheers,
Norman


2009/1/21 Justin Erenkrantz <ju...@erenkrantz.com>:
> On Wed, Jan 21, 2009 at 3:19 AM, Jukka Zitting <ju...@gmail.com> wrote:
>> As discussed, I'd like to start moving the git clones from my server
>> to Apache infrastructure. The best way to do this seems to be to set
>> up a new Solaris zone for this purpose. Here's a rough starting point
>> for documenting such a zone:
>
> From the peanut gallery, I think leveraging either harmonia (EU SVN
> mirror) or thor (new box @ OSU OSL) might make more sense than using a
> Solaris zone off helios/gaea/hyperion/etc.  -- justin
>

Re: Zone for hosting git clones at Apache

Posted by Santiago Gala <sa...@gmail.com>.
El jue, 26-02-2009 a las 13:36 +0100, Jukka Zitting escribió:
(...)
> * Figure out a way to run gitweb, cgit, or another git web interface
> in a way that doesn't require separate processes to be spawned for
> each HTTP request.

gitweb is a perl cgi, so I guess standard mod_perl trickery might work.
I'm not a perl or mod_perl expert, but whoever has been maintaining
bugzilla, for instance, should know.

Regards
Santiago


Re: Project update scripts

Posted by Grzegorz Kossakowski <gk...@apache.org>.
Jukka Zitting pisze:
> Hi,
> 
> On Sat, Feb 28, 2009 at 12:31 AM, Grzegorz Kossakowski
> <gk...@apache.org> wrote:
>> Paul Querna pisze:
>>> Can't we can figure out what needs to be done, and just add another
>>> notification method to the post-commit hook?
>> In my opinion we need to figure out how to deal with mismatch that we will face.
>> The problem is that we have one big svn repository mapped into several smaller
>> Git repositories.
> 
> That's no big problem, as the Git mirrors already include enough
> git-svn metadata for an update script to determine whether a
> particular commit affects the mirror. All you need is a list of paths
> affected in the svn commit and a script that matches those paths to
> the relevant Git mirrors.

Yep, that was my idea as well but I still think this won't be just a few lines of scripting but we'll see what I'll come
up with.

BTW. I'll fork your scripts hosted on GitHub. Since they were not updated for a few months I guess they are out-dated.
Could you upload the latest version?

-- 
Best regards,
Grzegorz Kossakowski

Re: Project update scripts

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

On Sat, Feb 28, 2009 at 12:31 AM, Grzegorz Kossakowski
<gk...@apache.org> wrote:
> Paul Querna pisze:
>> Can't we can figure out what needs to be done, and just add another
>> notification method to the post-commit hook?
>
> In my opinion we need to figure out how to deal with mismatch that we will face.
> The problem is that we have one big svn repository mapped into several smaller
> Git repositories.

That's no big problem, as the Git mirrors already include enough
git-svn metadata for an update script to determine whether a
particular commit affects the mirror. All you need is a list of paths
affected in the svn commit and a script that matches those paths to
the relevant Git mirrors.

BR,

Jukka Zitting

Re: Project update scripts

Posted by Grzegorz Kossakowski <gk...@apache.org>.
Paul Querna pisze:
> Can't we can figure out what needs to be done, and just add another
> notification method to the post-commit hook?

In my opinion we need to figure out how to deal with mismatch that we will face. The problem is that we have one big svn
repository mapped into several smaller Git repositories.

If big repository is being modified by commit we have to find out which Git repositories are affected by that commit. I
don't know how replacing monitoring e-mail notifications with post-commit hook will help us with this particular problem.

-- 
Best regards,
Grzegorz Kossakowski

Re: Project update scripts

Posted by Paul Querna <pq...@apache.org>.
Grzegorz Kossakowski wrote:
> Jukka Zitting pisze:
>>> Is it possible to subscribe to some notification list that notifies about
>>> any commit made into public part of our svn? If so I would like to
>>> subscribe to it and then I could experiment with writing such a script.
>> I don't think we have that, but the existing per-project commit
>> mailing lists should be a good a source of example notifications.
> 
> Yep. Good idea.
> 
> The problem is that there is no 1-1 mapping between per-project notification list and actual projects in SVN. What I
> mean here is (for example) situation with Cocoon where we have standard layout for 2.1 and 2.2 versions of Cocoon but
> Cocoon 3.0 that is written from scratch is subproject with its own branches/tags/trunk layout.
> 
> Thus we would have to either:
> 1. Update all projects associated with given e-mail address
> 2. extract the path of modified  files and try to scan all .git/config files for matching paths
> 
> The second way is totally automatic so no mapping between repositories and e-mail addresses has to be maintained but
> this way is much more tricky.
> 
> Do we need this trickery or we are fine with first option?
> 

Can't we can figure out what needs to be done, and just add another 
notification method to the post-commit hook?

Project update scripts (was: Re: Zone for hosting git clones at Apache)

Posted by Grzegorz Kossakowski <gk...@apache.org>.
Jukka Zitting pisze:
>> Is it possible to subscribe to some notification list that notifies about
>> any commit made into public part of our svn? If so I would like to
>> subscribe to it and then I could experiment with writing such a script.
> 
> I don't think we have that, but the existing per-project commit
> mailing lists should be a good a source of example notifications.

Yep. Good idea.

The problem is that there is no 1-1 mapping between per-project notification list and actual projects in SVN. What I
mean here is (for example) situation with Cocoon where we have standard layout for 2.1 and 2.2 versions of Cocoon but
Cocoon 3.0 that is written from scratch is subproject with its own branches/tags/trunk layout.

Thus we would have to either:
1. Update all projects associated with given e-mail address
2. extract the path of modified  files and try to scan all .git/config files for matching paths

The second way is totally automatic so no mapping between repositories and e-mail addresses has to be maintained but
this way is much more tricky.

Do we need this trickery or we are fine with first option?

-- 
Best regards,
Grzegorz Kossakowski

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

On Thu, Feb 26, 2009 at 10:37 PM, Grzegorz Kossakowski
<gk...@apache.org> wrote:
> Jukka Zitting pisze:
>> * Document things like how to request a new mirror (file an INFRA
>> issue with project name, svn URL, and the trunk/branches/tags layout
>> used by the project) and how to request a mirror update for a project
>> being moved (INFRA issue with svn URLs before and after the change).
>
> I guess we can take care of it once we have JIRA project in place.

A few days ago I created a "Git" component in the INFRA project. Let's use that.

> Is it possible to subscribe to some notification list that notifies about
> any commit made into public part of our svn? If so I would like to
> subscribe to it and then I could experiment with writing such a script.

I don't think we have that, but the existing per-project commit
mailing lists should be a good a source of example notifications.

> May I have an access (read-only is enough at this stage) to your current
> server so I can check all small details of the current setup?

Sure, I'll send you the account details in a moment.

>> * Update the mirror scripts to use svn.eu.apache.org internally, but
>> record svn.apache.org in the commit messages.
>
> That seems to be very easy as git-svn supports it.

I think so, yes.

>> * Create a script that updates all svn.eu.apache.org references in the
>> commit messages of the current mirrors to svn.apache.org.
>
> May have a look into this as well.

OK, cool.

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Grzegorz Kossakowski <gk...@apache.org>.
Gavin pisze:
> 
> Perhaps you meant when Git is listed as a sub-component of INFRA issues, if
> so, it is already there.
> 
> If you didn't then I am confused, Git is not an Apache project, and so wont
> be getting its own Jira Project.

Jukka and Gavien: Yep I meant sub-component and it's my fault that I didn't check if it already exists but only based my
opinion on no e-mail notification informing about its creation coming to my inbox.

Sometimes I'm too fast on sending e-mail. Sorry for confusion.

-- 
Best regards,
Grzegorz Kossakowski

RE: Zone for hosting git clones at Apache

Posted by Gavin <ga...@16degrees.com.au>.

> -----Original Message-----
> From: Grzegorz Kossakowski [mailto:gkossakowski@apache.org]
> Sent: Friday, 27 February 2009 7:37 AM
> To: infrastructure-dev@apache.org
> Subject: Re: Zone for hosting git clones at Apache
> 
> Jukka Zitting pisze:
> > Hi,
> >
> > On Wed, Feb 25, 2009 at 5:09 PM, Grzegorz Kossakowski
> > <gk...@apache.org> wrote:
> >> Since I have some spare time I wonder if I could help with Git
> introduction into Apache.
> >>

<snip>

> >
> > * Document things like how to request a new mirror (file an INFRA
> > issue with project name, svn URL, and the trunk/branches/tags layout
> > used by the project) and how to request a mirror update for a project
> > being moved (INFRA issue with svn URLs before and after the change).
> 
> I guess we can take care of it once we have JIRA project in place.
> 

Perhaps you meant when Git is listed as a sub-component of INFRA issues, if
so, it is already there.

If you didn't then I am confused, Git is not an Apache project, and so wont
be getting its own Jira Project.

Gav...

> --
> Best regards,
> Grzegorz Kossakowski
> 
> 


Re: Zone for hosting git clones at Apache

Posted by Grzegorz Kossakowski <gk...@apache.org>.
Jukka Zitting pisze:
> Hi,
> 
> On Wed, Feb 25, 2009 at 5:09 PM, Grzegorz Kossakowski
> <gk...@apache.org> wrote:
>> Since I have some spare time I wonder if I could help with Git introduction into Apache.
>>
>> Are there any free tasks?
> 
> Yeah, here's some on my TODO list:
> 
> * Create https://svn.apache.org/repos/asf/infrastructure/trunk/git/
> and copy the mirror scripts from my Github project to there
> 
> * Clean up the mirror scripts and add some usage documentation
> 
> * Create http://www.apache.org/dev/git.html and seed it with the
> information we currently have in the wiki.
> 
> * Document things like how to request a new mirror (file an INFRA
> issue with project name, svn URL, and the trunk/branches/tags layout
> used by the project) and how to request a mirror update for a project
> being moved (INFRA issue with svn URLs before and after the change).

I guess we can take care of it once we have JIRA project in place.

> * Create a commit hook (or email update) script that we could use to
> automatically sync a Git mirrors whenever a commit affects the
> mirrored subtree. Unlike my current email update script, this script
> should be able to automatically detect (using the git-svn metadata)
> which Git mirror needs to be updated based on the paths updated in the
> svn commit.

Is it possible to subscribe to some notification list that notifies about any commit made into public part of our svn?
If so I would like to subscribe to it and then I could experiment with writing such a script.

May I have an access (read-only is enough at this stage) to your current server so I can check all small details of the
current setup?

> * Figure out a way to run gitweb, cgit, or another git web interface
> in a way that doesn't require separate processes to be spawned for
> each HTTP request.
> 
> * Update the mirror scripts to use svn.eu.apache.org internally, but
> record svn.apache.org in the commit messages.

That seems to be very easy as git-svn supports it.

> * Create a script that updates all svn.eu.apache.org references in the
> commit messages of the current mirrors to svn.apache.org.

May have a look into this as well.

-- 
Best regards,
Grzegorz Kossakowski

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

On Wed, Feb 25, 2009 at 5:09 PM, Grzegorz Kossakowski
<gk...@apache.org> wrote:
> Since I have some spare time I wonder if I could help with Git introduction into Apache.
>
> Are there any free tasks?

Yeah, here's some on my TODO list:

* Create https://svn.apache.org/repos/asf/infrastructure/trunk/git/
and copy the mirror scripts from my Github project to there

* Clean up the mirror scripts and add some usage documentation

* Create http://www.apache.org/dev/git.html and seed it with the
information we currently have in the wiki.

* Document things like how to request a new mirror (file an INFRA
issue with project name, svn URL, and the trunk/branches/tags layout
used by the project) and how to request a mirror update for a project
being moved (INFRA issue with svn URLs before and after the change).

* Create a commit hook (or email update) script that we could use to
automatically sync a Git mirrors whenever a commit affects the
mirrored subtree. Unlike my current email update script, this script
should be able to automatically detect (using the git-svn metadata)
which Git mirror needs to be updated based on the paths updated in the
svn commit.

* Figure out a way to run gitweb, cgit, or another git web interface
in a way that doesn't require separate processes to be spawned for
each HTTP request.

* Update the mirror scripts to use svn.eu.apache.org internally, but
record svn.apache.org in the commit messages.

* Create a script that updates all svn.eu.apache.org references in the
commit messages of the current mirrors to svn.apache.org.

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Grzegorz Kossakowski <gk...@apache.org>.
Jukka Zitting pisze:
> Hi,
> 
> Picking this up again...
> 
> On Sat, Jan 31, 2009 at 1:33 PM, Norman Maurer <no...@apache.org> wrote:
>> I think it would make sense to put it on a zone hosted on odyne. Odyne
>> is located in the in the same datacenter (SARA) as harmonia (EU SVN).
> 
> Seems reasonable. See INFRA-1914 for the zone request.
> 
> I've also created INFRA-1913 as a top-level issue for tracking
> everything we need to do move the Git mirrors to Apache infra.

Hi Jukka,

Since I have some spare time I wonder if I could help with Git introduction into Apache.

Are there any free tasks?

-- 
Best regards,
Grzegorz Kossakowski

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

Picking this up again...

On Sat, Jan 31, 2009 at 1:33 PM, Norman Maurer <no...@apache.org> wrote:
> I think it would make sense to put it on a zone hosted on odyne. Odyne
> is located in the in the same datacenter (SARA) as harmonia (EU SVN).

Seems reasonable. See INFRA-1914 for the zone request.

I've also created INFRA-1913 as a top-level issue for tracking
everything we need to do move the Git mirrors to Apache infra.

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Luciano Resende <lu...@gmail.com>.
On Sat, Jan 31, 2009 at 4:33 AM, Norman Maurer <no...@apache.org> wrote:
> Hi guys,
>
> Sorry for the late response :-/
> I think it would make sense to put it on a zone hosted on odyne. Odyne
> is located in the in the same datacenter (SARA) as harmonia (EU SVN).
>
> Cheers,
> Norman
>
> 2009/1/21 Justin Erenkrantz <ju...@erenkrantz.com>:
>> On Wed, Jan 21, 2009 at 3:19 AM, Jukka Zitting <ju...@gmail.com> wrote:
>>> As discussed, I'd like to start moving the git clones from my server
>>> to Apache infrastructure. The best way to do this seems to be to set
>>> up a new Solaris zone for this purpose. Here's a rough starting point
>>> for documenting such a zone:
>>
>> From the peanut gallery, I think leveraging either harmonia (EU SVN
>> mirror) or thor (new box @ OSU OSL) might make more sense than using a
>> Solaris zone off helios/gaea/hyperion/etc.  -- justin
>>
>

One of the issues I see when using EU SVN is that committing multiple
git commits is a painful manual process of multiple git svn rebase +
git svn dcommit as it seems we need to wait for EU SVN to resync with
EUA SVN.

Are we still going to have restrictions to only use EU SVN once we
bring the git clones more close to the apache infra ?


-- 
Luciano Resende
Apache Tuscany, Apache PhotArk
http://people.apache.org/~lresende
http://lresende.blogspot.com/

Re: Zone for hosting git clones at Apache

Posted by Norman Maurer <no...@apache.org>.
Hi guys,

Sorry for the late response :-/
I think it would make sense to put it on a zone hosted on odyne. Odyne
is located in the in the same datacenter (SARA) as harmonia (EU SVN).

Cheers,
Norman

2009/1/21 Justin Erenkrantz <ju...@erenkrantz.com>:
> On Wed, Jan 21, 2009 at 3:19 AM, Jukka Zitting <ju...@gmail.com> wrote:
>> As discussed, I'd like to start moving the git clones from my server
>> to Apache infrastructure. The best way to do this seems to be to set
>> up a new Solaris zone for this purpose. Here's a rough starting point
>> for documenting such a zone:
>
> From the peanut gallery, I think leveraging either harmonia (EU SVN
> mirror) or thor (new box @ OSU OSL) might make more sense than using a
> Solaris zone off helios/gaea/hyperion/etc.  -- justin
>

Re: Zone for hosting git clones at Apache

Posted by Henning Schmiedehausen <he...@schmiedehausen.org>.
Having something close to svn.apache.org, running on ASF infrastructure
would probably allow the git folks to actually hook into the svn commit
hooks so that the git stuff no longer needs to be triggered by commit
mails. 

	Ciao
		Henning


On Wed, 2009-01-21 at 08:01 -0800, Justin Erenkrantz wrote:
> On Wed, Jan 21, 2009 at 3:19 AM, Jukka Zitting <ju...@gmail.com> wrote:
> > As discussed, I'd like to start moving the git clones from my server
> > to Apache infrastructure. The best way to do this seems to be to set
> > up a new Solaris zone for this purpose. Here's a rough starting point
> > for documenting such a zone:
> 
> From the peanut gallery, I think leveraging either harmonia (EU SVN
> mirror) or thor (new box @ OSU OSL) might make more sense than using a
> Solaris zone off helios/gaea/hyperion/etc.  -- justin
-- 
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen       -- Java, J2EE, Linux
Mail: henning@schmiedehausen.org    -- Consultant, Architect, Developer
Web:  http://henning.schmiedehausen.org/



Re: Zone for hosting git clones at Apache

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On Wed, Jan 21, 2009 at 3:19 AM, Jukka Zitting <ju...@gmail.com> wrote:
> As discussed, I'd like to start moving the git clones from my server
> to Apache infrastructure. The best way to do this seems to be to set
> up a new Solaris zone for this purpose. Here's a rough starting point
> for documenting such a zone:

>From the peanut gallery, I think leveraging either harmonia (EU SVN
mirror) or thor (new box @ OSU OSL) might make more sense than using a
Solaris zone off helios/gaea/hyperion/etc.  -- justin

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

On Thu, Jan 22, 2009 at 6:47 PM, Jukka Zitting <ju...@gmail.com> wrote:
> On Thu, Jan 22, 2009 at 6:26 PM, Paul Querna <pq...@apache.org> wrote:
>> Have you looked at cgit?
>
> It looks nice, though I haven't tried it yet. I'll set it up on my
> server to see how it works in practice.

See a quick and dirty setup at http://jukka.zitting.name/cgit/

cgit seems like a solid piece of work and is reasonably configurable.
However, even though it doesn't spawn extra git processes to gather
information, it still is a standalone CGI script which needs to be run
once for each HTTP request.

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

On Thu, Jan 22, 2009 at 6:26 PM, Paul Querna <pq...@apache.org> wrote:
> I'm not sure I'm comfortable using the straight up gitweb right now, and
> using forked one off projects sucks.

Fine by me.

> Have you looked at cgit?

It looks nice, though I haven't tried it yet. I'll set it up on my
server to see how it works in practice.

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Paul Querna <pq...@apache.org>.
Jukka Zitting wrote:
> Hi,
> 
> On Wed, Jan 21, 2009 at 8:59 PM, Paul Querna <pq...@apache.org> wrote:
>> I agree with Justin, we should try to productionize this as much as
>> possible, including using role users, and putting it on a real machine, not
>> a shared zone box (might still run it on a zone on thor, but not shared with
>> other random TLP things).
> 
> Sounds OK to me. My original intent was to impose as few changes to
> existing infrastructure as possible, but I'd be happy also to go for a
> more integrated solution.
> 
>> How are you currently running gitweb?
> 
> As a standalone CGI script.
> 
> The only caching I needed to add is a rewrite rule that maps the main
> page (http://jukka.zitting.name/git/) to a static file that I
> periodically update.
> 
>     RewriteCond %{QUERY_STRING} =""
>     RewriteRule ^$ gitweb.html
> 
> Otherwise the main page would have taken too long to generate with the
> large number of repositories I host.
> 
>> I started looking, and it seems to only support CGI or, running inside
>> mod_perl -- it seems we would want to use the mod_perl version of course for
>> reasonable performance.
> 
> Agreed. I haven't tried it yet, but it should be possible to run
> gitweb with mod_perl.
> 
>> I'm also concerned that it appears kernel.org has forked gitweb for
>> performance enhancements, and the upsteam project has not taken these in:
>> <http://lkml.org/lkml/2007/2/22/338>
> 
> Yep, when investigating alternatives for the main page cache setup I
> did, I got the impression that the main gitweb developers aren't that
> eager to add any form of caching.

I talked to the guy who runs kernel.org's gitweb on freenode IRC, and 
they still run their fork.

I'm not sure I'm comfortable using the straight up gitweb right now, and 
using forked one off projects sucks.

>> Did we want to run gitweb, or another git web based frontend?
> 
> No particular preference, I just set up gitweb as it was easily and
> readily available.
> 

Have you looked at cgit?

http://hjemli.net/git/cgit/tree/README

It powers a few bigger sites like freedesktop:
http://cgit.freedesktop.org/

Which makes me think it might be viable for our uses.

Thanks,

Paul


Re: Zone for hosting git clones at Apache

Posted by Jukka Zitting <ju...@gmail.com>.
Hi,

On Wed, Jan 21, 2009 at 8:59 PM, Paul Querna <pq...@apache.org> wrote:
> I agree with Justin, we should try to productionize this as much as
> possible, including using role users, and putting it on a real machine, not
> a shared zone box (might still run it on a zone on thor, but not shared with
> other random TLP things).

Sounds OK to me. My original intent was to impose as few changes to
existing infrastructure as possible, but I'd be happy also to go for a
more integrated solution.

> How are you currently running gitweb?

As a standalone CGI script.

The only caching I needed to add is a rewrite rule that maps the main
page (http://jukka.zitting.name/git/) to a static file that I
periodically update.

    RewriteCond %{QUERY_STRING} =""
    RewriteRule ^$ gitweb.html

Otherwise the main page would have taken too long to generate with the
large number of repositories I host.

> I started looking, and it seems to only support CGI or, running inside
> mod_perl -- it seems we would want to use the mod_perl version of course for
> reasonable performance.

Agreed. I haven't tried it yet, but it should be possible to run
gitweb with mod_perl.

> I'm also concerned that it appears kernel.org has forked gitweb for
> performance enhancements, and the upsteam project has not taken these in:
> <http://lkml.org/lkml/2007/2/22/338>

Yep, when investigating alternatives for the main page cache setup I
did, I got the impression that the main gitweb developers aren't that
eager to add any form of caching.

> Did we want to run gitweb, or another git web based frontend?

No particular preference, I just set up gitweb as it was easily and
readily available.

Note that a git web interface isn't that essential, as you can also
browse all the same information using a local clone of a git
repository. A git web frontend on apache.org would essentially
duplicate the functionality of our viewvc interface, just with a
different perspective (per project view, commits instead of files as
primary entities, etc.).

BR,

Jukka Zitting

Re: Zone for hosting git clones at Apache

Posted by Paul Querna <pq...@apache.org>.
Jukka Zitting wrote:
> Hi,
> 
> As discussed, I'd like to start moving the git clones from my server
> to Apache infrastructure. The best way to do this seems to be to set
> up a new Solaris zone for this purpose. Here's a rough starting point
> for documenting such a zone:

I agree with Justin, we should try to productionize this as much as 
possible, including using role users, and putting it on a real machine, 
not a shared zone box (might still run it on a zone on thor, but not 
shared with other random TLP things).

How are you currently running gitweb?

I started looking, and it seems to only support CGI or, running inside 
mod_perl -- it seems we would want to use the mod_perl version of course 
for reasonable performance.

I'm also concerned that it appears kernel.org has forked gitweb for 
performance enhancements, and the upsteam project has not taken these in:
<http://lkml.org/lkml/2007/2/22/338>

Did we want to run gitweb, or another git web based frontend?

Thanks,

Paul