The last project I ran here at Red Gate was on SQL Compare / Data Compare 11 project. SQL Compare is a superficially simple product designed, as it’s name suggests, to compare Microsoft SQL database schemas. Its cousin, SQL Data Compare compares the data in the tables.
Like the product itself, the Compare 11 project had a superficially simple goal: add support for the new features in SQL Server 2014. Easy, right? It turns out, not so much…
The largest new feature is Memory-Optimized Tables (MOTs): the ability to keep data in memory so you get significant performance improvements through reduced disk-access. But as you dig into what MOTs can and can’t do, you realise that you’ll need a lot of special-cases in the code to deal with the limitations. The details of those limitations is well outside the scope of this post, but one of my colleagues Michelle Taylor has written a useful post on our company blog for those who want to know more.
When the team came to start work on this story, which was before the release of SQL Server, some of the limitations weren’t immediately apparent. This led me (as project manager) to a false sense of security, and so I was far too lax on getting the user stories and acceptance criteria defined in detail. The lack of knowledge this created in the team caused the feature to become massively under-estimated: just 50 task-points (at their velocity, each task-point turned out to be just over an hour of real work).
The first tasks related to this story went on the Scrum board early in the project: ‘Add property to CreateOrAlterTable’, ‘Add property to DatabaseObjectsTable’, ‘Update DatabaseObjectsTable.Equals()’. You see what’s missing? We had dived straight into implementation with no form of planning first. That first sprint was estimated as 15 points. By the time we got to the end of it, we had uncovered quite how unexpectedly complex this feature was likely to become. We were nowhere near 30% (15 out of 50 points) complete on the feature. We carried some of the tasks forward to the following Sprint, however, trying to convince ourselves that we’d crack it soon. Before I knew it, we’d spent nearly a month on this feature, continually uncovering more and more complexity, and morale on the team was starting to become impacted.
Even then, my monthly project A3 report excessively optimistically promised that we’re complete both Inline Indexes (another 2014 feature) and Memory-Optimised Tables in the next four weeks.
It was time to pull the Andon cord.
Luckily, this also coincided with a slight change to business focus. The team was being asked to ship SQL Compare 10.7, giving support for connecting to SQL Server 2014 (but not the new features) to ship with the final release of SQL Server 2014.
Whilst this was a massive disruption to the team, it did give us some breathing space to find out more about MOTs before we had to work on them again.
By the time we were ready to start work on MOTs again, the Product Owner had defined the Acceptance Criteria in more detail. There were a series of tasks, themselves totalling 42 task-points in estimates, which centred on reviewing these ACs and making sure they made sense. This worked OK, and put us in a much better place, but with a large team, the meetings didn’t really work.
At this point, one of my colleagues pointed me to the idea of a 3-amigos meeting. I’d actually tried this once before on a different project, and I found myself wondering why I hadn’t tried this earlier on the MOTs work. The idea is to have three people (specifically the Product Owner, a Developer, and a Tester) in a room together, defining how the feature should work for users, in the code, and in test cases respectively. There’s a good write up of the meeting on the Scrum Alliance website; the only difference for us being that the Product Owner also doubled in the BA role for us. Although it felt slightly more waterfall than we were used to, the output from the meeting was a set of really well defined, clear Acceptance Criteria and a good common understanding of what MOTs involved, that was ready for sharing across the team.
The team re-estimated the work, and came up with an estimate of 250 points. In the end, it took 337 – still rather more than 250, but also far more than the original 50 points planned!
This process worked so well that we adopted the same meeting for other features in SQL Server 2014, rotating the role of the amigos through the dev team. As long as the communication about what was discussed is good immediately after the meeting, everyone became brought into the idea of 3-amigos, and ensured that the chaotic start to MOTs wasn’t repeated during the rest of the project.