On Fri, Jun 8, 2012 at 3:27 PM, Greg Von Kuster <greg@bx.psu.edu> wrote:
Hi John,
On Jun 8, 2012, at 1:22 PM, John Chilton wrote:
Hello Greg,
Thanks for the prompt and detailed response (though it did make me sad). I think deploying tested, static components and configurations to production environments and having production environments not depending on outside services (like the tool shed) should be considered best practices.
I'm not sure I understand this issue. What processes are you using to upgrade your test and production servers with new Galaxy distributions? If you are pulling new Galaxy distributions from our Galaxy dist repository in bitbucket, then pulling tools from the Galaxy tool shed is not much different - both are outside services. Updating your test environment, determining it is functionally correct, and then updating your production environment using the same approach would generally follow a best practice approach. This is the approach we are currently using for our public test and main Galaxy instances at Penn State.
We don't pull down from bitbucket directly to our production environment, we pull galaxy-dist changes into our testing repository, merge (that can be quite complicated, sometimes a multihour process), auto-deploy to a testing server, and then finally we push the tested changes into a bare production repo. Our sys admins then pull in changes from that bare production repo in our production environment. We also prebuild eggs in our testing environment not live on our production system. Given the complicated merges we need to do and the configuration files that need to be updated each dist update it would seem making those changes on a live production system would be problematic. Even if one was pulling changes directly from bitbucket into a production codebase, I think the dependency on bitbucket would be very different than on N toolsheds. If our sys admin is going to update Galaxy and bitbucket is down, that is no problem he or she can just bring Galaxy back up and update later. Now lets imagine they shutdown our galaxy instance, updated the code base, did a database migration, and went to do a toolshed migration and that failed. In this case instead of just bringing Galaxy back up they will now need to restore the database from backup and pullout of the mercurial changes. Anyway all of that is a digression right, I understand that we will need to have the deploy-time dependencies on tool sheds and make these tool migration script calls part of our workflow. My lingering hope is for a way of programmatically importing and updating new tools that were never part of Galaxy (Qiime, upload_local_file, etc...) using tool sheds. My previous e-mail was proposing or positing a mechanism for doing that, but I think you read it like I was trying to describe a way to script the migrations of the existing official Galaxy tools (I definitely get that you have done that). Thanks again for your time and detailed responses, -John