[Starcluster] More on the rebundling issue

Justin Riley jtriley at MIT.EDU
Thu Jan 14 10:15:55 EST 2010


Hi Mark/Dan,

You're correct about not nuking the old configs if you're attempting to image an
already running StarCluster instance.

I should have mentioned earlier that I generally tend to launch a single
starcluster ami instance outside of starcluster that does not go through the
launch phase to avoid these config files being bundled with the image.

My typical workflow for imaging is this:

1. Start a single starcluster ami instance from elasticfox
2. Log in to that instance and apt-get update/upgrade and install pkgs to your
heart's desire.
3. Logout and then run create_image.py after I'm satisfied with the software
stack.

Imaging an active StarCluster node (ie that was launched by starcluster -s) is
problematic because of the fact that configs arent cleaned up. Furthermore, you
wouldn't want create_image.py to clean them up because then your active cluster
would no longer work.

In the new version I'm considering having the createimage action also launch an
instance and drop you in a ssh shell to do the updating before doing the actual
packaging. After you exit that shell, the script would then continue on with the
packaging. This would make sure that you're always working with a standalone
instance outside of starcluster.

Hope that helps...

BTW the next version is ready for testing. However, I need to do a bit of
documentation first, otherwise it'll be a bit of a mystery on how to use it.
The current version in github has a new createimage command for bundling EC2
images on S3.

I'm hoping to get some documentation done sometime between now and the weekend
and then send out a call for testing.

~Justin



More information about the StarCluster mailing list