Saturday, March 19, 2011

Programming in sprint cycles (currently sprint 9)

This post is meant to be a bit more observational than my usual how-to posts. It describes what my first experiencing of programming in agile sprints has been like.


The sprints are set up on a weekly basis, starting on Thursday mornings with a demo to the stakeholders of the completed user stories of the just-completed sprint. Each demo is guided from a list of (JIRA-based) tickets stored in a spreadsheet. Read out the ticket name, go over the acceptance criteria, hide the spreadsheet, and demo the completed feature. The demos are recorded in Camtasia the day before, and stored online for ease of access by all members of the team for later reference (eg. an additional stakeholder added, or a team member back from 2 weeks vacation.)

With the demo completed, a sprint retrospective is held, guided by our team lead / SCRUM person. We go into more or less detail, depending on the week. (I've found this process to be amazingly clarifying and effective.) We then launch JIRA, go into the planning view, add any additional stories/tasks/bugs that have become evident, assign any unassigned tickets, and place them in likely order of completion. We then go through each ticket and give it a story point rating. This rating has been much easier to arrive at with each subsequent sprint. We look at the total story points, knowing what our average velocity has become per sprint, and either push tickets back to the later sprint, or perhaps bring tickets forward into the current sprint if we are a bit short.

By this point it is typically noon, so we grab some lunch. After lunch I go to my desk, launch JIRA, flip into task board view, and look at my backlog. I drag over (typically the top) item. Most of the time, I try and keep only one item in the active development swimlane (although by the end of a busy day (several smaller story-point tickets may have traversed from backlog to development to tech review).

I'm now ready to start, so I launch git bash, change to the git repo containing the area where I need to work (typically web applications, with a submodule to shared libraries). I do a git fetch to get the latest, checkout the dev branch, and do a git merge --ff-only origin/dev if necessary to catch up to latest. I think create a local branch on that commit, typically with the JIRA ticket number in the branch name, something like prd1234addMissingNullCheckGuardCondition (I've always liked long descriptive branch names or for that matter, method names in code, since tab in git or Intellisense in VS will be happy to save the typing effort for me later).

At this point, I try and gain an understanding of what's needed. Before long, I'll be creating an NUnit test. The test may be in a Nnnn.Tests.Unit project or Nnnn.Tests.Integration project, depending on the legacy code situation and the ability (or inability) to substitute dependencies. The general rule of thumb at the moment is that if it is new functionality altogether, or if it is abstracting/creating utility or helper classes that work with the legacy code, then unit tests will be possible. For new functionality, dependencies are DI injected and tested with mocks or just simple fake implementation classes.

When the work is completed, I do git fetch again, fast-forward the dev branch if new work has been added, swtich to my local branch, rebase it onto the dev branch, switch to dev branch, merge in my local branch, check the results in gitk --all GUI view, and if everything looks clean, push it to origin. I do this in the submodule / shared libraries first, and then again at the outer level. This push to origin is picked up on by the TeamCity integration server, and for the current project that typically means I want to wait for TeamCity to create the artifact (a combination of pages and xcopy-deploy style DLLs whose structure I have setup and verified on TeamCity awhile back). When TeamCity, compiles, passes all tests, and generates the artifact successfully, I save the artifact zip file to the web server, unzip it, move it to the deploy directory, update the permissions, and repoint the IIS home directory for that project to the new artifact.

With that complete, I drag the ticket from the development swimlane to the tech review swimlane, alert QA that it is ready, typically verbally, often supplemented with an explanation either in JIRA or just in email, and look to my backlog for the next ticket.

This is of course interrupted constantly by emails, discussions, whiteboard sketch sessions, morning standup, and afternoon coffee runs, but in the midst of daily chaos, the ongoing process of the tickets moving across the swimlanes gives a steady rhythm to the week.

We're in sprint 9 now, and probably one more sprint will finish this project. Another project is in the works. This is a most interesting, satisfying, and fascinating process.