Fedora

During my package source control talk, Toshio and I started chatting about how to solve the current "sending cvs commit mail is slow" problem. I let slip something that I had been cooking in my head for quite a while now.

A unified message bus between applications / systems we use in Fedora infrastructure. What does this mean? Well I'm not nearly the developer enough to describe how it would actually look in applications, I just thing conceptually this would be very valuable. Lets take for example our source control system:

Consider that right now when you commit something, we figure out (by querying Fedora Account System) who all should be notified of your change. Our SCM has specific knowledge of how to do this, as it's post-commit hook. Lets say later we want to add the ability to modify bugzilla states given specific terms found in commit logs or spec file changelogs. We'd have to modify the scm some more, and make things even slower, and more importantly, even harder to replace later.

What if instead, our SCM just dropped a notification on a communications bus that a change was made, with the details of the change. That could be pretty instant, and the client would return quickly. We could have various tools listening on the bus, and when they see this commit message, one could send mail to appropriate people, another could key off of bug details and touch on bugzilla, a third could notice something about it and start a bodhi update request, so on and so forth. When we want to add functionality all we have to do is create another listener on the bus that does the explicit task in mind. We don't have to add more domain specific info to our source control system, and if we ever have to replace the source control system all we have to ensure is that it can drop messages on the bus.

To me, this seems like an incredibly powerful development model, and really builds on that "Voltronic" thing that folks like Michael Dehaan talk about.

Unfortunately I missed Mike McGrath's "CSI" talk and he may have already talked about something like this, but Toshio got really excited when I briefly mentioned my system bus.

Now, I'm no where near developer enough to know if this is feasible, how difficult it would be, whether we can truly guarantee message delivery and execution, etc.. etc.. I'm just offering up my vision to get folks thinking and talking.

It's Sunday, the day after our BarCamp day for FUDCon Boston 2008. I'm still trying to recover on sleep and brain power, so I'm sitting on my couch watching my son play with his toy cars while listening to the radio. I'm resisting the urge to do actual work right now, so instead I'll blog about the last few days.

As always, FUDCon is incredibly energetic for me. I get to see a lot of people, hear snippits of great conversations, and always end up trying to spread myself way too thin with lots of things going on. Something I noticed this year, and a number of people have pointed out to me, I seem to be something of a catalyst. When I bring people together to talk about something, rapid progress is made, and somehow I manage to convince other people to go do a bunch of great work, without me (:

A great example of this is the Fedora QA/QE/Whatever they want to be called team of rock stars. On Thursday they showed me where they had gotten with a Testopia install on a Fedora public test instance. I think my eyes literally lit up when I saw this, and immediately I started overreaching my bounds and basically said we will use this for Fedora 10 testing, and volunteered them to give a talk at the BarCamp about Testopia and the progress made at FUDCon. We quickly whipped up a list of 5 or 10 things to focus on during the hackfest and when they all went to go work on it I had to run away to go talk to some other people. I feel kind of bad that I didn't actually /do/ anything for it, but man it's great to see that kind of work happen.

John Poelstra had posted a few days ago about "What is Rawhide?". John has this great knack of asking the really hard questions. A few of us got together with a whiteboard and really started plowing through this line of questioning. I asked John for his list of big issues and I wrote them down. We chatted at a high level about the questions and came to a list we'd tackle in depth. So then question by question we just went at it. Not getting lost in any implementation details, nor trying to do any blame game, just identifying problems and brainstorming solutions. John kept notes through it all so that we could continue the creativity. I think we managed to get through quite a few of the issues and even identified and blocked out the need for a few new tools for our contributors to make use of. Seth, since he was around, was asked about maybe an existing yum-util to do one of our needs and since it didn't exist, Seth whipped it up in no time. We didn't even have to ask him, this guy totally rocks. I can't wait for John to post up his notes from the little jam session as we really accomplished a lot of good.

I popped into the Spins talk and was flat out amazed at how well Mo was running things there. Together we walked through a few different user experience stories and along the way covered some crucial areas of how spins as a program should work. When you come at things from actual use experiences it really helps to answer the hard questions. Unfortunately I couldn't stay for the whole meeting, but I'm extremely confident in that team. I think they've blocked out enough work for the next year.

BarCamp day was incredibly entertaining. The amount of people pitching talks was mind blowing. I heard somebody comment that there were more people in line to give talks than there were people in the audience listening to the pitches. This is definitely a good problem to have. Somehow the BarCamp chaos worked itself out into a schedule of talks and off I went. First stop was the Free Java / Eclipse talk which was pretty awesome. Then I went to Jeremy's git talk, and even though I've been using git for a while now, I still learned a couple things. More git training would be awesome. Due to a scheduling conflict, I was not able to attend the Secondary Arches talk, because I was giving my talk on requirements gathering for Fedora Package SCM 2.0. This was a very helpful talk for me, both from getting lists of requirements and also getting some data on the current pain points of our system today. From those we actually identified some things we could potentially fix right now. Next up was Testopia. I let James Laska and Jon Stanley run this because, well, they did most the work. I did however get up and provide some context as to how I'd like to see us use Testopia in Fedora's future. Finally I went to the "Ask a Kernel Guy" session which was pretty informative. Paul gave a closing speech where he both gave praise for some pretty key and cool areas seeing huge amounts of good, and then identified some areas we could really use some improvement. I think feedback like this is quite helpful for the project. FUDPub followed which is always a good time. Later on I biked home with Jeremy (after spending about 20 minutes trying to actually leave, lots of good byes).

Over the next few weeks I'll be trying to dig through all the info that was shoved into my head these last few days and start making progress on some things. The project is really running strong now and as Jim said in his talk with us, this is just the beginning.