Thoughts on ostree and EC2

[Owen Taylor <otaylor redhat com>]

A couple of days ago tried following Colin's instructions for running an
EC2 build of gnome-ostree. Wanted to dump out a few thoughts from that
before I forget.

* Mostly went quite smoothly - there were a couple of snags in the setup
  scripts, that I sent to Colin and are now fixed. Other problems:

  - There is a mismatch in 'ln -s ~/build/ostbuild/work/logs logs'
    vs. 'workdir=~/build/ostbuild' in ostree-user-install.sh.
    Not sure what one is right - I changed the symlink on my system.

  - The instructions in your mail didn't mention the need to 
    run 'bitbake ostree-native' before the other bitbake commands.

* Stats for my build were around 20 hours runtime and $4 of usage. This
  is reasonable, but it's certainly not instant gratification.

  If we want to fire off build servers frequently or encourage people
  to do so, we definitely need to figure out how to get this faster -
  presumably by not starting from scratch but using the repo/ and 
  maybe poky sstate from the "official build".
  
* I ran my build on a VM backed with an Elastic Block Storage device -
  this got picked by default, but then I was wondering if it was
  the right choice - in the end it seemed fine - it was fast enough to
  keep the build CPU bound and the IO charges were about $0.30.

  The nice thing about a EBS-backed VM is that you can shut it down,
  and start it again - but the caution is that 150GB's of provisioned
  storage is $15/mo.

* I found the directory structure and naming to be quite confusing when
  I dove in - it might be good to have a wiki page somewhere documenting
  what all the different names are - e.g. to figure out why the repo
  ended up in:

   ~/build/gnomeos-build/tmp-eglibc/deploy/images/repo

  (I know there's a symlink - it didn't really unconfuse me.)
  
  Maybe some reorganization would be good as well - the current
  organization feels very much dependent on technology details, 
  while if the toplevel organization was on results, it would be easier
  to get oriented - something like:

   build/
      shared/
         ostbuild/
             src-mirror/
      gnomeos/
         i686/
             repo/
             poky/
             ostbuild/

  Just a thought.

* [Writing down something that came up in conversation with Colin]

  One of the cool things we could do with EC2 is start test instances
  there - the ability to boot from a EBS volume should make that easy -
  we can mount an EBS volume into an instance, ostree pull into the
  root directory, then detach the volume and fire it up and see if it
  works.

  (To get hardlink goodness, you'd probably want the EBS volume to have
  a git style organization where the repository is in /.ostree.)