hi sam

On Sunday, January 27, 2002, at 03:03 PM, Sam Ruby wrote:

> For security and performance reasons, builds of web pages are not to be
> done on daedalus (the machine that hosts Apache's web presence).  Also for
> security reasons, new accounts are not automatically being created on
> daedalus.  In fact, Brian has indicated that he ultimately wants to reduce
> the number of accounts on daedalus.

i think that the only reason i use the account on daedalus (i'm right in 
thinking that the cvs login uses another machine, aren't i?) is to update 
sub-project web sites. doing these updates makes me nervous and i'd gladly 
give up my account if there was another way of doing them.

> Here's how I would like to incrementally address the issue:
>
> 1) I'll create a cron job to do a cvs update of the jakarta site several
>    times a day.  I will expand it to include other subprojects on request.
>    While I may adjust the frequency with which this job runs, I expect to
>    run at least once a day
>
>    Ramifications of this: first the existence of this job does not 
> preclude
>    manual updates out of cycle by those with access to do so.  What it 
> does
>    mean, however, is that if someone goes and directly updates the web 
> site
>    (which despite all the warnings, does happen from time to time), this
>    will inevitably result in a merge conflict and need to be manually
>    resolved.

+1
i'd be happy for anyone who can't play by the rules to lose access rights 
to daedalus.

is there any way that the conflict could be recognized and a warning 
posted somewhere so that the problem can be fixed as quickly as possible?

> 2) Gump already builds the site and captures jars and javadocs.  See
>    http://gump.covalent.net/jars/ and
>    http://nagoya.apache.org/~rubys/gump/javadoc.html .  I'll add logic to
>    capture sites and publish them separately (i.e., not on the main site)
> .
>    Subprojects who want to participate need only ensure that the target
>    that gump is building for them does cause the site to be built, and 
> then
>    need to identify the directory into which this site is placed.  I
>    anticipate that this will be done in a quite similar manner to the way
>    that javadocs are done; see
>    http://jakarta.apache.org/gump/project.html#javadoc for details.  For
>    those who wish to get involved, please click on any of the get involved
>    links on the left hand side of the page.

+1

> 3) Once that is stable, I plan to convert the cron job on a subproject to
>    subproject basis from a cvs update to using rsync.  This means that the
>    generated html will no longer need to be checked into cvs.  It also
>    addresses the problem identified in #1 above.  I expect the rsync
>    process to occur more frequently than the cvs update as it consumes 
> less
>    resources, but I expect the builds to be done less frequently.  I still
>    expect the minimum frequency to be once a day.

+1

> 4) Ultimately, I'd like to explore means of allowing this process to be
>    triggered by other means; either on demand or by virtue of the cvs
>    commits themselves.  Any suggestions along these lines would be
>    appreciated - once again, feel free to get involved!

i quite like the idea of scheduled updates. with the main site we seem to 
be moving towards posting up changes so that people get a chance to 
improve them before they go live. i think that this some merit for sub 
projects as well.


your plan sounds like a good one to me :)

- robert


--
To unsubscribe, e-mail:   <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>

Reply via email to