14
March
2007

The Continuous Integration Barometer

This is an idea spun out of conversation with Ben and Paul at last night’s LRUG - a physical device for alerting developers working in a continuous build environment as to how far they are from the trunk.

Because of the rate of change on an agile project, it’s important to keep your own personal codebase as close to the trunk as possible, otherwise you’ll spend a long while working on code that doesn’t fit with the current state of the system, or working around issues which may have been fixed. It’d be good to have some sort of physical display to alert you to this, rather than relying on typing svn up every five minutes.

Continuous Integration Barometer

Enter the continuous build barometer. It’s dead simple: it’s a dial you connect to your computer via a USB connection. The moment you do a checkout, the needle sets itself to 0. And then, as your codebase diverges from the trunk, the needle indicates how far (proportionally) your code diversifies from the original. And, of course, the moment you check in, the needle resets itself to zero.

Ideally, the continuous-build barometer needs to be made out of something reasonably heavy - I have this image of brass and wood, much like this example, if only to convey the weight that “checking in often” actually carries. In terms of making one… dial indicators seem to be quite pricey. It’d be easy to prototype with something like this ten-bar LED display, an Arduino board, and a short-ish script to compare the trunk with your local version, cronned to run regularly. Of course, I think the physical nature of the object is important, but it’d still be interesting to see what a working version in any form looked like.

That was one of my favourite things this month’s LRUG gave me, and it’s not really even code. Now: who wants one?

23
February
2007

code

Tagged as:
, , , .

The latency of software development, pre- and post-launch.

In a recent post about offshoring development work, Ryan Carson explains that the way you know when you’ve outgrown a freelance developer is this:

Getting bugs fixed and new features implemented starts taking fricken’ forever.

There’s some interesting discussion on the post - about eastern-European wage rates, about freelancing, etcetera - but there was one elephant in the room that nobody really mentioned, and I really think it’s necessary. And that’s this:

once your application is live, everything will take longer.

Whilst you’re in the build process, you can turn on a dime; you can commit new code at a moment’s notice, change the direction of the codebase or even the application, churn out features at a remarkable rate. And, when something doesn’t work (despite having passed the test suite… which you do have, right?), you patch it and move on. It’s a doozy.

Once you’re live, everything changes. You’ve got an active audience using the site - you might even have paying customers. You can’t afford for a single thing to go wrong. New features are no longer about build it, run the tests, check it in, deploy. Your testing on the development box has to be more fastidious. The integration between new functionality and the existing needs to be really well thought through. The design needs to be seamless with that which exists already. And, if it goes live and there’s still one of those totally unforseen bugs… you need to be ready to roll back at a moment’s notice, and put the feature on hold until it’s ready.

This is not new. Some teams are lucky enough to make many deploys daily; most can muster up daily fixes; some might go to weekly. But the effort that a feature requires post-launch is far more than it does pre-launch.

So it’s inevitable - freelancers or no - that the pace of development is going to fall off a bit post-launch. Hopefully, only a bit - but it’s not going to be the same rate unless you’re extremely lucky. You’ve also got to keep your users in mind, and break them into new functionality slowly.

It’s a hard transition, going from the development site to live. To you, as an owner, a developer, it feels like the same product - so the change in pace is a bit frustrating. And it’s important to keep the pace as fast as possible. Some fixes are critical, and need to go out the door the instant they’re done; others can wait a little longer - a day or two, if necessary - for release. It’s a rare, focused, hugely talented team that can splurge out new features like a machinegun.

Now, I’m sure that the problem is compounded by avoiding maintenance contracts (and so having to re-contract and re-hire) and having developers whose time is precious. But it’s always important to bear in mind that the latency of any web software development shoots up post-launch. So if it does: don’t immediately reach for the outsource button. Take stock, have some patience, and see what the product needs - how often it needs to be released, how often it needs to be improved. And then you can make that call a little better informed.

The other golden rule, incidentally, is never deploy on Friday. I hope the reasoning behind that is obvious.

Links & notes for this month

Endnotes