Is it possible to deploy without having a repository? I currently am
not concerned with source control for a quick-and-dirty setup I'm
working on and I just need Capistrano to work. Honestly, all I'm
looking for is a quick *installer* and setup. What am I missing. . .?
Thanks,
Michael
scp -r my_site name@site:~/my_site
Then reconcile public/.htaccess and public_html/.htaccess, sym-link
public_html to my_site/public, and set the permissions on dispatch.*
to 755.
You are asking "how do I use a car to walk"; just walk!
--
Phlip
http://c2.com/cgi/wiki?ZeekLand <-- NOT a blog!!
Thanks,
Michael
> Phliphttp://c2.com/cgi/wiki?ZeekLand <-- NOT a blog!!
The answer is "yes, you can use Capistrano to deploy without an SCM".
However, you can't use any of the standard deployment tasks to do so;
you'll have to write your own tasks, which would wind up being pretty
much what Phlip suggested.
- Jamis
> --~--~---------~--~----~------------~-------~--~----~
> To unsubscribe from this group, send email to capistrano-
> unsub...@googlegroups.com
> For more options, visit this group at http://groups.google.com/
> group/capistrano
> -~----------~----~----~----~------~----~------~--~---
>
Thanks in advance,
Michael
> smime.p7s
> 3KDownload
> point me in the right direction so as to get all knowed up on it?
I want to know how to write
cap --tasks
and get a list of tasks. (I also then need to know which ones to
extend to add actions to each step in the process. Without reading the
documentation site, or Googling for [Capistrano Cheat Sheet].
But that don't work. I want to know which tasks to grep for, to read
their source and see what they do.
I'm no expert on Capistrano, but I can tell when code is coupled, and
Capistrano looks very tightly coupled to its SCM!
--
Phlip
Nah, not at all. Capistrano's deployment functionality is very
explicit in that it requires SCM's for deployment. If your project is
trivial enough that it doesn't require an SCM, then it's trivial
enough that you don't need Capistrano to deploy it. Phlip had a very
good suggestion for you in that case: scp works fine. Or rsync. Or
any other similar tool.
Don't call a tool "short sighted" if it doesn't do what you want.
Instead, spend your energy positively by looking for the right tool
for the job. Obviously, capistrano isn't that tool for you, unless
you don't mind writing a few tasks to do what you want it to do.
- Jamis
On Feb 22, 2007, at 2:15 PM, gberz3 wrote:
>
> Obviously you guys are much more in the know regarding Cap's ability.
> I was simply looking for all the management "recipes" without having
> SCM. I agree that it's good practice to have SCM, but I think it was
> short sighted to actually *require* it for deployment. I simply need
> to be able to point it a local directory and have it sync, etc. I
> suppose my main concern is really the initial setup for each app.
>
> Anywho, it appears that the main "manual" site is down. Can you gents
> point me in the right directly so as to get all knowed up on it?
>
> Thanks in advance,
> Michael
>> smime.p7s
>> 3KDownload
It turns out you could use my latest version of
local_subversion_rsync.rb to do what you want with some minor hacks...
just pointing you in the right direction here.
Follow the directions in my post, point the local rsync cache to your
folder with your source. Comment out the calls to subversion in my
code, and it should work.
http://blog.wolfman.com/articles/2007/02/18/updated-capistrano-local-subversion-and-perforce
Basically it will simply rsync your local copy with the server, so
long as you comment out all the svn accesses.
> smime.p7s
> 3KDownload
On Feb 22, 5:48 pm, "wolfmanjm" <wolfma...@gmail.com> wrote:
> Although I agree with Jamis (of course ;)
>
> It turns out you could use my latest version of
> local_subversion_rsync.rb to do what you want with some minor hacks...
> just pointing you in the right direction here.
>
> Follow the directions in my post, point the local rsync cache to your
> folder with your source. Comment out the calls to subversion in my
> code, and it should work.
>
> http://blog.wolfman.com/articles/2007/02/18/updated-capistrano-local-...
As misguided as it may be in case of Ruby (that has no separation of
source and executable form), IT of some public corporations in USA and
Canada sees it as a legal requirement (usually referring to Sarbannes-
Oxley Act).
I do appreciate the fact that Capistrano devs don't care about formal
IT environments and public corporations. But if someone on this list
does, a Capistrano recipe that does this in a nice way is worth
publishing. Myself, or somebody else in ThoughtWorks, may even have to
do it soon - by the nature of our work we do care about formal IT
environments more than we may want to admit. :)
Thoughts?
Alex Verkhovsky
I think the code in my local subversion scm module may be a good place
to start.
Basically I tar (you could easily zip instead) up a release directory,
send it to the servers and de-tar there.
I think you could easily add an option to skip the tar-ing of the
releases directory, and just point to a pre-tar'd (or zipped)
package instead, as you suggest.
On Feb 24, 12:34 pm, "alexey.verkhov...@gmail.com"
Would it be insufficient to only give deployment read-only access to
the repository, and not give development access to the deployment
servers?
-faisal
Would it be insufficient to only give deployment read-only access to
the repository, and not give development access to the deployment
servers?
I can't entirely disagree with that sentiment.
At a former employer the makefile spit out tarballs of the final
product, and marked those tarballs (and their contents) with the
date, building user, and CVS tags related to that source. A subset
of users had write access to the (semi-publicly readable) directory
these tarballs were distributed from, and end-users could type a
specific command to figure out what release they were running. This
let support figure out exactly what they were debugging with. The
company was doing software distribution rather than web services, but
there might be some lessons there.
Capistrano's current model for deploying new code is:
local machine talks to app server, server pulls from svn repo
I could see use for
local machine exports from svn repo, scp's to app server
All the other moving parts (disable_web, restart, symlink,
enable_web, etc.) would be more or less the same. I'd be more
comfortable with this approach, but don't need it right now. Which
is to say: if someone builds it I'll want to play with it. Failing
that, I may try to whip something up before the heat death of the
universe.
-faisal
it is documented here... http://blog.wolfman.com/articles/2007/02/18/updated-capistrano-local-subversion-and-perforce
as well as various posts to this group, which makes me think is google
groups losing posts???
That is exactly what the local_subversion scm module does, with the
option of using rsync or scp to transfer the code to the server.