Home

[ I write this for I found the solution applied to my case at the 3rd link of the 2nd page of my Google search. I had never gone beyond the first page for years. ]

So there was this gitosis repository that lived on an Ubuntu server that had to move to a CentOS one.

Thankfully EPEL still carries gitosis and I did not built it from the source. I copied over the files and was done with it.

That is until I needed to change access to one of the repositories to add a user. Said user was denied any kind of access with:

ERROR:gitosis.serve.main:Repository read access denied

There are plenty of reasons for this occur (examples), but none seemed to fit in my case, so I brute forced the search results. And at the first comment that was posted on the third link at the second page Google came up with, the solution was found.

# cd ~gitosis/repositories/gitosis-admin.git/hooks/
# ls -l
total 48
-rwxr-xr-x 1 gitosis gitosis  452 Sep 11  2011 applypatch-msg.sample
-rwxr-xr-x 1 gitosis gitosis  896 Sep 11  2011 commit-msg.sample
-rwxr-xr-x 1 gitosis gitosis  160 Sep 11  2011 post-commit.sample
-rwxr-xr-x 1 gitosis gitosis  552 Sep 11  2011 post-receive.sample
lrwxrwxrwx 1 gitosis gitosis   61 Dec 21 09:53 post-update -> /usr/share/pyshared/gitosis/templates/admin/hooks/post-update
-rwxr-xr-x 1 gitosis gitosis  189 Sep 11  2011 post-update.sample
-rwxr-xr-x 1 gitosis gitosis  398 Sep 11  2011 pre-applypatch.sample
-rwxr-xr-x 1 gitosis gitosis 1578 Sep 11  2011 pre-commit.sample
-rwxr-xr-x 1 gitosis gitosis 4971 Sep 11  2011 pre-rebase.sample
-rwxr-xr-x 1 gitosis gitosis 1239 Sep 11  2011 prepare-commit-msg.sample
-rwxr-xr-x 1 gitosis gitosis 3611 Sep 11  2011 update.sample

Yeah, you’ve guessed it by now:

/usr/share/pyshared/gitosis/templates/admin/hooks/post-update

did not exist on CentOS but

/usr/lib/python2.6/site-packages/gitosis/templates/admin/hooks/post-update

did. Fixing the symbolic link fixed the problem.

/* Oh what fun it is to chase bugs into the night */

I think I wrote the following playbook a few months ago, when I was halfway through watching the 2 hour introduction (now replaced with this):

---
- name: automatic deploy new version
  hosts: ruby-production
  user: root
  sudo: yes
  tasks:

  - name: stop monit
    command: service monit stop

  - name: checkout correct version
    sudo_user: projectuser
    command: chdir=/workspace/project git checkout release-1.0

  - name: grab latest sources for that release
    sudo_user: projectuser
    command: chdir=/workspace/project git pull origin release-1.0

  - name: run db:migrate
    sudo_user: projectuser
    sudo: no
    script: migrate.sh

  - name: restart apache
    command: service httpd restart

  - name: start monit
    command: service monit start

What the above playbook does is simple:

– Stop monit to avoid any automatic restarts when you’re half way through updating stuff that it monitors.

– It makes sure that you grab the latest updates from the correct branch

– It runs a script on all servers that takes care of any bundle install and db:migrate stuff.

– It restarts apache. Yes, a touch tmp/restart.txt should do the trick but sometimes it does not.

– It restarts monit.

There is plenty of room for improvement here, for example using the git module instead of running an explicit command and even making use of roles as the project expands and becomes more demanding and of course get rid of the script in favor of a more ansible specific play.

So why post it now? Basically at the request of @nanobeep and as a means of self-pressure to improve the playbook. Maybe I should promise this to someone?

So there, nothing complex or elaborate. BTW, here is a similar way to do this with git and Capistrano that I bumped into.